Atomic Media

Publishers report ‘negligible’ traffic impact of Google AI Overviews

Thursday, August 15th, 2024

Two large publishers – Dotdash Meredith and Ziff Davis – say Google AI Overviews haven’t significantly impacted their traffic.

What Dotdash Meredith is saying. Here’s what the IAC (owner of Dotash) wrote its Q2 2024 shareholder letter:

• “Google began to roll out AI Overviews in mid-May, the impact on our traffic has been negligible. Referrals from Google search queries produce less than half of our traffic, and based on our analysis, AI-generated answers are being served on roughly 15% of searches across our categories, with the highest frequency in Health, Technology, and Finance. Click-through rate differentials between pages with and without AI Overviews are minor so far, but it is still early and products change quickly, so the past isn’t prologue.”

What Ziff Davis is saying. Here’s what Ziff Davis CEO Vivek Shah said on the company’s earnings call:

• “I know there continues to be investor interest in the impacts of AI on search. …Now that the AI Overview experience is in full circulation, we wanted to revisit the analysis we did in Q3 of 2023 relating to the frequency of AI overviews being presented to users.
• “We analyzed thousands of queries across our key domains that generate the lion’s share of our organic search referrals and the percentage of times that an AI Overview appeared. It fell to 8%, meaning 92% of the time the search engine results page did not include an AI Overview for the queries that matter to us most. Analysis by other industry experts indicate a similar percentage of overall searches resulting in AI Overviews.
• “At this point, we don’t see it as a significant change to the search experience.”

But. While this all makes sense, it’s important to remember that the rollout of Google AI Overviews has been fairly limited and volatile so far. Overall, AI Overviews appeared for 7% of queries, as of the end of July, according to BrightEdge research. The presence of AI Overviews peaked at 15% in May.

While Ziff Davis may not see AI Overviews as a “significant change” to Google’s search experience, I would add the word “yet” to the end of that sentence. Because Google has clearly stated search is evolving toward AI Overviews.

Why we care. Generative AI has just begun reshaping search and AI will continue to do so over the next decade. Granted, these two reports form a line, not a trend. But AI Overviews (and Search Generative Experience before it) caused a lot of anxiety for SEOs, publishers and content creators. So it’s helpful to get some insight into how publishers are being impacted in these early days of AI Overviews.

Courtesy of Search Engine Land: News & Info About SEO, PPC, SEM, Search Engines & Search Marketing

Google August 2024 core update rolling out now

Thursday, August 15th, 2024

Google released the August 2024 core update today. It will take about a month to fully roll out.

This update is not just a normal core update. The August 2024 core update takes into account the feedback Google heard since the September 2023 helpful content update that seemed to have a negative impact on many small and independent publishers.

What Google is saying. John Mueller, Search Advocate at Google, wrote.

• “Today, we launched our August 2024 core update to Google Search. This update is designed to continue our work to improve the quality of our search results by showing more content that people find genuinely useful and less content that feels like it was made just to perform well on Search.”

Google said this update aims to promote useful content from small and independent publishers, after Google listened to feedback it received since the release of the March 2024 core update. Mueller added:

• “This latest update takes into account the feedback we’ve heard from some creators and others over the past few months. As always, we aim to connect people with a range of high quality sites, including ‘small’ or ‘“’independent’ sites that are creating useful, original content on relevant searches. This is an area we’ll continue to address in future updates.”

This August 2024 core update “aims to better capture improvements that sites may have made, so we can continue to surface the best of the web,” Mueller added.

Guidance updated. Google posted several updates to its help page about core updates, including more in-depth guidance for those who may see changes after an update.

More details. Google told us we should expect a core update soon, after many publishers have become concerned and anxious about the next update.

Since then we have seen a tremendous amount of Google search ranking volatility without a confirmation from Google on a core update or any update of its kind. In fact, this morning, I posted about even more intense Google Search ranking volatility on Search Engine Roundtable.

What to do if you are hit. Google has given advice on what to consider if you are negatively impacted by a core update in the past. Google has not really given much new advice here.

• There aren’t specific actions to take to recover. A negative rankings impact may not signal anything is wrong with your pages.
• Google has offered a list of questions to consider if your site is hit by a core update.
• Google said you can see a bit of a recovery between core updates but the biggest change would be after another core update.

In short, write helpful content for people and not to rank in search engines.

• “There’s nothing new or special that creators need to do for this update as long as they’ve been making satisfying content meant for people. For those that might not be ranking as well, we strongly encourage reading our creating helpful, reliable, people-first content help page,” Google said previously.

Previous core updates. The previous core update – the March 2024 core update – was the largest core update, according to Google. It started March 5 and completed 45 days later on April 19.

Here’s a timeline and our coverage of recent core updates:

• The March 2024 core update was on March 5th and ended on April 19.
• The November 2023 core update was on November 2nd and ended on November 28.
• The October 2023 core update was on October 5th and ended on October 19
• The August 2023 core update was on August 22nd and ended on September 7.
• The March 2023 core update was on March 15th and ended on March 28th.

Other updates. We did have a spam update between the last core update and this core update. It was the June 2024 spam update that started on June 20 and took 7 days to finish rolling out, completing on June 27.

Why we care. Many sites are hoping, and have been hoping, to see improvements with the last core update ever since the September 2023 helpful content update rolled out. Most, if not all, of those sites that were hit in September did not see recoveries. They were hoping to see recoveries with the March 2024 core update, but did not.

Now, with this August 2024 core update, many of those sites hit by previous updates will be watching closely to see if their sites recover over the next few weeks.

Courtesy of Search Engine Land: News & Info About SEO, PPC, SEM, Search Engines & Search Marketing

Google AI Overviews now show for signed-out users in the US

Thursday, August 15th, 2024

Google AI Overviews will now appear for all users in the United States, even if they are not signed into their Google account. Google has confirmed with Search Engine Land that AI Overviews are available for signed-out users in the US after we noticed Google testing it this morning.

What it looks like. I just conducted a search while not signed into Google using Chrome and Google showed me an AI Overview at the top of the search results.

Here is a screenshot:

More details. As a reminder, Google launched AI Overviews to US searchers back in May, after the Google I/O event.

Today, Google expanded AI Overviews to six new countries, added new links to the AI Overviews and added new Search Labs experimental tests for AI Overviews.

Jon Henshaw posted on LinkedIn yesterday he was seeing these while signed out, so I dug in and Google confirmed, it is 100% rolled out to all US searchers. I should note, I personally do not see AI Overviews when searching logged in to my Google Workspace account but Google did tell me that Workspace accounts can see AI Overviews.

Why we care. As more and more searchers see AI Overviews, it might lead to different click behavior from the Google Search results. Google touts how AI Overviews should lead to more traffic to publishers, but publishers have good reason to be doubtful of such statements, especially since Google is not showing publishers distinct impression and click data from Search Console on AI Overviews.

This also means that SEO tools will be able to better track these AI Overviews, as Lily Ray pointed out. Mark Traphagen from seoClarity also confirmed that his toolset is seeing AI Overviews in incognito mode.

AI Overviews seem to be here for the foreseeable future and it is our job to ensure publishers and content creators get traffic from Google Search, even with AI Overviews being found at the top of those results.

Courtesy of Search Engine Land: News & Info About SEO, PPC, SEM, Search Engines & Search Marketing

Google AI Overviews volatility continues: Rise, fall, repeat

Tuesday, August 13th, 2024

The visibility of AI Overviews in Google’s search results increased to 12% in July, only to fall back to 7% by the end of the month, according to new data from enterprise SEO platform BrightEdge.

While AI Overviews now have a “stronger presence” compared to June, the visibility of Google’s AI-generated answers previously appeared on 15% of queries in May.

Why we care. Google’s AI Overviews continue to be a huge area of interest and concern among SEOs and content creators. Google is continuing to test and refine formats, citations and the amount of real estate AI Overviews occupy in the search results. The continued volatility clearly indicates AI Overviews are an unsolved problem for Google.

Decreases. AI Overviews no longer appear for the travel and entertainment queries BrightEdge is tracking using its Generative Parser. AI Overviews also no occupy 12.5% less vertical space in Google’s search results.

In July, AI Overviews were less likely to appear for:

• General product searches (e.g., “shoes,” “furniture”).
• Brand-specific queries (e.g., Google, Apple).
• Basic how-to queries (e.g., “how to do,” “how to find”).
• Healthcare searches – Google removed AI results for keywords related to “rare,” “genetic” and “complex” disorders. However, healthcare still has the highest number of AI Overviews (61%, down from 68%).

Increases. Salary-related queries (e.g., “nurse salary,” “human resources manager salary”) saw a significant increase in AI Overviews – jumping from 7% to 85% in July. AI Overviews were also more likely to be shown for:

• Complex technical keywords (e.g., “confirmatory factor analysis”).
• Long-tail keywords (e.g., “types of data analysis”).
• Specific product features (e.g., gooseneck electric kettle”)

Other growth and volatility. AI Overviews overlap with Google’s search results increased overall:

• Pages ranking in positions 31-100 experienced a “general increase in overlap, sometimes exceeding 3%.”
• Citations of top 10 results decreased, and “significant spikes” were seen.
• The most stable area was for pages in positions 11-30, but there were some fluctuations here as well.

The sources cited by AI Overviews also experienced significant fluctuation, according to BrightEdge. Some websites losing citations included Wikipedia (-5%), CDC (-15%), USA Today (-60%) and Forbes (-30%).

Other findings from the latest data found that AI Overviews showed for:

• 45% of B2B tech queries for most of July, up from 29% – however, there was a significant drop and increase volatility at the end of the month.
• 38% of education queries, up from 16% – but this sector also saw a sharp decline and volatility at the end of July.
• 35% of insurance queries in the second half of July, but dropped significantly by the end of the month.
• 17% of ecommerce queries, up from 11%, with significant volatility at the end of July.
• 6.5% of finance queries, up from 6%.

Courtesy of Search Engine Land: News & Info About SEO, PPC, SEM, Search Engines & Search Marketing

Google Ads API streamlines conversion adjustment uploads

Tuesday, August 13th, 2024

Google will soon make significant changes to its Google Ads API, simplifying the process of uploading conversion adjustments.

Why we care. This update removes a key friction point for advertisers and API users, allowing for more immediate data adjustments and potentially more accurate campaign optimization.

Driving the news. Starting Sept. 9, users can upload conversion adjustments immediately after the original conversion is recorded, eliminating the previous 24-hour waiting period.

Key changes:

• Removal of error codes related to “too recent” conversions.
• Adjustments will be counted as “pending” until processed (up to 24 hours).
• Affects Google Ads API versions v15, v16, and v17.

What to watch:

1. API users should remove logic that enforces waiting periods for adjustment uploads.
2. Applications should be modified to no longer track the removed error codes.
3. Users relying on successful or failed event count metrics may need to adjust their logic, especially for v17.

Between the lines. This update streamlines the conversion adjustment process, potentially leading to more timely and accurate campaign data for advertisers.

What’s next. Google advised users to update their applications and processes ahead of Sept. 9.

Bottom line. While this change simplifies the upload process, it may require some adjustments to existing workflows and applications for Google Ads API users.

Courtesy of Search Engine Land: News & Info About SEO, PPC, SEM, Search Engines & Search Marketing

Google extends deadline for Hotel Ads commission bidding sunset

Tuesday, August 13th, 2024

Google announced a grace period for its planned phase-out of commission-based bidding in Hotel Ads campaigns. The new deadline is Feb. 20, 2025

Why we care. This extension gives hotel advertisers more time to adapt and carefully plan and implement new bidding strategies to ensure uninterrupted campaign performance.

Key details:

• Existing Hotel ads campaigns using commission-based strategies will continue until Feb. 20, 2025.
• Reconciliation for commissions (per stay) available until Nov. 30, 2025.

How to prepare:

• Shift existing campaigns to new bid strategies before the sunset date.
• Consider creating new Performance Max campaigns for travel goals.
• Plan for reconciliation of any outstanding commissions.

---

Courtesy of Search Engine Land: News & Info About SEO, PPC, SEM, Search Engines & Search Marketing

How Google Search ranking works

Tuesday, August 13th, 2024

It should be clear to everyone that the Google documentation leak and the public documents from antitrust hearings do not really tell us exactly how the rankings work. 

The structure of organic search results is now so complex – not least due to the use of machine learning – that even the Google employees who work on the ranking algorithms say they can no longer explain why a hit is at one or two. We do not know the weighting of the many signals and the exact interplay.

Nevertheless, it is important to familiarize yourself with the structure of the search engine to understand why well-optimized pages do not rank or, conversely, why seemingly short and non-optimized results sometimes appear at the top of the rankings. The most important aspect is that you need to broaden your view of what is really important.

All the information available clearly shows that. Anyone who is even marginally involved with ranking should incorporate these findings into their own mindset. You will see your websites from a completely different point of view and incorporate additional metrics into your analyses, planning and decisions.

To be honest, it is extremely difficult to draw a truly valid picture of the systems’ structure. The information on the web is quite different in its interpretation and sometimes differs in terms, although the same thing is meant. 

An example: The system responsible for building a SERP (search results page) that optimizes space use is called Tangram. In some Google documents, however, it is also referred to as Tetris, which is probably a reference to the well-known game.

Over weeks of detailed work, I have viewed, analyzed, structured, discarded and restructured almost 100 documents many times. 

This article is not intended to be exhaustive or strictly accurate. It represents my best effort (i.e., “to the best of my knowledge and belief”) and a bit of Inspector Columbo’s investigative spirit. The result is what you see here.

A new document waiting for Googlebot’s visit

When you publish a new website, it is not indexed immediately. Google must first become aware of the URL. This usually happens either via an updated sitemap or via a link placed there from an already-known URL. 

Frequently visited pages, such as the homepage, naturally bring this link information to Google’s attention more quickly. 

The trawler system retrieves new content and keeps track of when to revisit the URL to check for updates. This is managed by a component called the scheduler. The store server decides whether the URL is forwarded or whether it is placed in the sandbox. 

Google denies the existence of this box, but the recent leaks suggest that (suspected) spam sites and low-value sites are placed there. It should be mentioned that Google apparently forwards some of the spam, probably for further analysis to train its algorithms. 

Our fictitious document passes this barrier. Outgoing links from our document are extracted and sorted according to internal or external outgoing. Other systems primarily use this information for link analysis and PageRank calculation. (More on this later.) 

Links to images are transferred to the ImageBot, which calls them up, sometimes with a significant delay, and they are placed (together with identical or similar images) in an image container. Trawler apparently uses its own PageRank to adjust the crawl frequency. If a website has more traffic, this crawl frequency increases (ClientTrafficFraction).

Alexandria: The great library

Google’s indexing system, called Alexandria, assigns a unique DocID to each piece of content. If the content is already known, such as in the case of duplicates, a new ID is not created; instead, the URL is linked to the existing DocID.

Important: Google differentiates between a URL and a document. A document can be made up of multiple URLs that contain similar content, including different language versions, if they are properly marked. URLs from other domains are also sorted here. All the signals from these URLs are applied via the common DocID. 

For duplicate content, Google selects the canonical version, which appears in search rankings. This also explains why other URLs may sometimes rank similarly; the determination of the “original” (canonical) URL can change over time.

As there is only this one version of our document on the web, it is given its own DocID. 

Individual segments of our site are searched for relevant keyword phrases and pushed into the search index. There, the “hit list” (all the important words on the page) is first sent to the direct index, which summarizes the keywords that occur multiple times per page. 

Now an important step takes place. The individual keyword phrases are integrated into the word catalog of the inverted index (word index). The word pencil and all important documents containing this word are already listed there. 

In simple terms, as our document prominently contains the word pencil multiple times, it is now listed in the word index with its DocID under the entry “pencil.” 

The DocID is assigned an algorithmically calculated IR (information retrieval) score for pencil, later used for inclusion in the Posting List. In our document, for example, the word pencil has been marked in bold in the text and is contained in H1 (stored in AvrTermWeight). Such and other signals increase the IR score. 

Google moves documents considered important to the so-called HiveMind, i.e., the main memory. Google uses both fast SSDs and conventional HDDs (referred to as TeraGoogle) for long-term storage of information that doesn’t require quick access. Documents and signals are stored in the main memory. 

Notably, experts estimate that before the recent AI boom, about half of the world’s web servers were housed at Google. A vast network of interconnected clusters allows millions of main memory units to work together. A Google engineer once noted at a conference that, in theory, Google’s main memory could store the entire web. 

It’s interesting to note that links, including backlinks, stored in HiveMind seem to carry significantly more weight. For example, links from important documents are given much greater importance, while links from URLs in TeraGoogle (HDD) may be weighted less or possibly not considered at all.

• Hint: Provide your documents with plausible and consistent date values. BylineDate (date in the source code), syntaticDate (extracted date from URL and/or title) and semanticDate (taken from the readable content) are used, among others.
• Faking topicality by changing the date can certainly lead to downranking (demotion). The lastSignificantUpdate attribute records when the last significant change was made to a document. Fixing minor details or typos does not affect this counter.

Additional information and signals for each DocID are stored dynamically in the repository (PerDocData). Many systems access this later when it comes to fine-tuning relevance. It is useful to know that the last 20 versions of a document are stored there (via CrawlerChangerateURLHistory). 

Google has the ability to evaluate and assess changes over time. If you want to completely change the content or topic of a document, you would theoretically need to create 20 intermediate versions to override the old content signals. This is why reviving an expired domain (a domain that was previously active but has since been abandoned or sold, perhaps due to insolvency) does not offer any ranking advantage.

If a domain’s Admin-C changes and its thematic content changes at the same time, a machine can easily recognize this at this point. Google then sets all signals to zero, and the supposedly valuable old domain no longer offers any advantages over a completely newly registered domain.

QBST: Someone is looking for ‘pencil’

When someone enters “pencil” as a search term in Google, QBST begins its work. The search phrase is analyzed, and if it contains multiple words, the relevant ones are sent to the word index for retrieval. 

The process of term weighting is quite complex, involving systems like RankBrain, DeepRank (formerly BERT) and RankEmbeddedBERT. The relevant terms, such as “pencil,” are then passed on to the Ascorer for further processing. 

Ascorer: The ‘green ring’ is created

The Ascorer retrieves the top 1,000 DocIDs for “pencil” from the inverted index, ranked by IR score. According to internal documents, this list is referred to as a “green ring.” Within the industry, it is known as a posting list. 

The Ascorer is part of a ranking system known as Mustang, where further filtering occurs through methods such as deduplication using SimHash (a type of document fingerprint), passage analysis, systems for recognizing original and helpful content, etc. The goal is to refine the 1,000 candidates down to the “10 blue links” or the “blue ring.” 

Our document about pencils is on the posting list, currently ranked at 132. Without additional systems, this would be its final position.

Superroot: Turn 1,000 into 10!

The Superroot system is responsible for re-ranking, carrying out the precision work of reducing the “green ring” (1,000 DocIDs) to the “blue ring” with only 10 results.

Twiddlers and NavBoost perform this task. Other systems are probably in use here, but their exact details are unclear due to vague information.

• Google Caffeine no longer exists in this form. Only the name has remained.
• Google now works with countless micro-services that communicate with each other and generate attributes for documents that are used as signals by a wide variety of ranking and re-ranking systems and with which the neural networks are trained to make predictions.

Filter after filter: The Twiddlers

Various documents indicate that several hundred Twiddler systems are in use. Think of a Twiddler as a plug-in similar to those in WordPress. 

Each Twiddler has its own specific filter target. They are designed this way because they are relatively easy to create and don’t require changes to the complex ranking algorithms in Ascorer.

Modifying these algorithms is challenging and would involve extensive planning and programming due to potential side effects. In contrast, Twiddlers operate in parallel or sequentially and are unaware of the activities of other Twiddlers.

There are basically two types of Twiddlers.

• PreDoc Twiddlers can work with the entire set of several hundred DocIDs because they require little or no additional information. 
• In contrast, Twiddlers of the “Lazy” type require more information, for example, from the PerDocData database. This takes correspondingly longer and is more complex. 

For this reason, the PreDocs first reduce the posting list to significantly fewer entries and then start with slower filters. This saves an enormous amount of computing capacity and time. 

Some Twiddlers adjust the IR score, either positively or negatively, while others modify the ranking position directly. Since our document is new to the index, a Twiddler designed to give recent documents a better chance of ranking might, for instance, multiply the IR score by a factor of 1.7. This adjustment could move our document from the 132nd place to the 81st place.

Another Twiddler enhances diversity (strideCategory) in the SERPs by devaluing documents with similar content. As a result, several documents ahead of us lose their positions, allowing our pencil document to move up 12 spots to 69. Additionally, a Twiddler that limits the number of blog pages to three for specific queries boosts our ranking to 61.

Our page received a zero (for “Yes”) for the CommercialScore attribute. The Mustang system identified a sales intention during analysis. Google likely knows that searches for “pencil” are frequently followed by refined searches like “buy pencil,” indicating a commercial or transactional intent. A Twiddler designed to account for this search intent adds relevant results and boosts our page by 20 positions, moving us up to 41.

Another Twiddler comes into play, enforcing a “page three penalty” that limits pages suspected of being spam to a maximum rank of 31 (Page 3). The best position for a document is defined by the BadURL-demoteindex attribute, which prevents ranking above this threshold. Attributes like DemoteForContent, DemoteForForwardlinks and DemoteForBacklinks are used for this purpose. As a result, three documents above us are demoted, allowing our page to move up to Position 38.

Our document could have been devalued, but to keep things simple, we’ll assume it remains unaffected. Let’s consider one last Twiddler that assesses how relevant our pencil page is to our domain based on embeddings. Since our site focuses exclusively on writing instruments, this works to our advantage and negatively impacts 24 other documents.

For instance, imagine a price comparison site with a diverse range of topics but with one “good” page about pencils. Because this page’s topic differs significantly from the site’s overall focus, it would be devalued by this Twiddler. 

Attributes like siteFocusScore and siteRadius reflect this thematic distance. As a result, our IR score is boosted once more, and other results are downgraded, moving us up to Position 14.

As mentioned, Twiddlers serve a wide range of purposes. Developers can experiment with new filters, multipliers or specific position restrictions. It’s even possible to rank a result specifically either in front of or behind another result. 

One of Google’s leaked internal documents warns that certain Twiddler features should only be used by experts and after consulting with the core search team.

“If you think you understand how they work, trust us: you don’t. We’re not sure that we do either.”

– Leaked “Twiddler Quick Start Guide – Superroot” document

There are also Twiddlers that only create annotations and add these to the DocID on the way to the SERP. An image then appears in the snippet, for example, or the title and/or description are dynamically rewritten later.

If you wondered during the pandemic why your country’s national health authority (such as the Department of Health and Human Services in the U.S.) consistently ranked first in COVID-19 searches, it was due to a Twiddler that boosts official resources based on language and country using queriesForWhichOfficial.

You have little control over how Twiddler reorders your results, but understanding its mechanisms can help you better interpret ranking fluctuations or “inexplicable rankings.” It’s valuable to regularly review SERPs and note the types of results. 

For example, do you consistently see only a certain number of forum or blog posts, even with different search phrases? How many results are transactional, informational, or navigational? Are the same domains repeatedly appearing, or do they vary with slight changes in the search phrase?

If you notice that only a few online stores are included in the results, it might be less effective to try ranking with a similar site. Instead, consider focusing on more information-oriented content. However, don’t jump to conclusions just yet, as the NavBoost system will be discussed later.

Google’s quality raters and RankLab

Several thousand quality raters work for Google worldwide to evaluate certain search results and test new algorithms and/or filters before they go “live.”

Google explains, “Their ratings don’t directly influence ranking.” 

This is essentially correct, but these votes do have a significant indirect impact on rankings.

Here’s how it works: Raters receive URLs or search phrases (search results) from the system and answer predetermined questions, typically assessed on mobile devices. 

For example, they might be asked, “Is it clear who wrote this content and when? Does the author have professional expertise on this topic?” The answers to these questions are stored and used to train machine learning algorithms. These algorithms analyze the characteristics of good and trustworthy pages versus less reliable ones.

This approach means that instead of relying on Google search team members to create criteria for ranking, algorithms use deep learning to identify patterns based on the training provided by human evaluators.

Let’s consider a thought experiment to illustrate this. Imagine people intuitively rate a piece of content as trustworthy if it includes an author’s picture, full name, and a LinkedIn biography link. Pages lacking these features are perceived as less trustworthy.

If a neural network is trained on various page features alongside these “Yes” or “No” ratings, it will identify this characteristic as a key factor. After several positive test runs, which typically last at least 30 days, the network might start using this feature as a ranking signal. As a result, pages with an author image, full name, and LinkedIn link might receive a ranking boost, potentially through a Twiddler, while pages without these features could be devalued.

Google’s official stance of not focusing on authors could align with this scenario. However, leaked information reveals attributes like isAuthor and concepts such as “author fingerprinting” through the AuthorVectors attribute, which makes the idiolect (the individual use of terms and formulations) of an author distinguishable or identifiable – again via embeddings. 

Raters’ evaluations are compiled into an “information satisfaction” (IS) score. Although many raters contribute, an IS score is only available for a small fraction of URLs. For other pages with similar patterns, this score is extrapolated for ranking purposes.

Google notes, “A lot of documents have no clicks but can be important.” When extrapolation isn’t possible, the system automatically sends the document to raters to generate a score.

The term “golden” is mentioned in relation to quality raters, suggesting there might be a gold standard for certain documents or document types. It can be inferred that aligning with the expectations of human testers could help your document meet this gold standard. Additionally, it’s likely that one or more Twiddlers might provide a significant boost to DocIDs deemed “golden,” potentially pushing them into the top 10.

Quality raters are typically not full-time Google employees and may work through external companies. In contrast, Google’s own experts operate within the RankLab, where they conduct experiments, develop new Twiddlers and evaluate whether these or refined Twiddlers improve result quality or merely filter out spam. 

Proven and effective Twiddlers are then integrated into the Mustang system, where complex, computationally intensive and interconnected algorithms are used.

---

But what do users want? NavBoost can fix that!

Our pencil document hasn’t fully succeeded yet. Within Superroot, another core system, NavBoost, plays a significant role in determining the order of search results. NavBoost uses “slices” to manage different data sets for mobile, desktop, and local searches.

Although Google has officially denied using user clicks for ranking purposes, FTC documents reveal an internal email instructing that the handling of click data must remain confidential.

This shouldn’t be held against Google, as the denial of using click data involves two key aspects. Firstly, acknowledging the use of click data could provoke media outrage over privacy concerns, portraying Google as a “data octopus” tracking our online activity. However, the intent behind using click data is to obtain statistically relevant metrics, not to monitor individual users. While data protection advocates might view this differently, this perspective helps explain the denial.

FTC documents confirm that click data is used for ranking purposes and frequently mention the NavBoost system in this context (54 times in the April 18, 2023 hearing). An official hearing in 2012 also revealed that click data influences rankings.

It has been established that both click behavior on search results and traffic on websites or webpages impact rankings. Google can easily evaluate search behavior, including searches, clicks, repeat searches and repeat clicks, directly within the SERPs.

There has been speculation that Google could infer domain movement data from Google Analytics, leading some to avoid using this system. However, this theory has limitations. 

First, Google Analytics does not provide access to all transaction data for a domain. More importantly, with over 60% of people using the Google Chrome browser (over three billion users), Google collects data on a substantial portion of web activity. 

This makes Chrome a crucial component in analyzing web movements, as highlighted in hearings. Additionally, Core Web Vitals signals are officially collected through Chrome and aggregated into the “chromeInTotal” value.

The negative publicity associated with “monitoring” is one reason for the denial, while another is the concern that evaluating click and movement data could encourage spammers and tricksters to fabricate traffic using bot systems to manipulate rankings. While the denial might be frustrating, the underlying reasons are at least understandable.

• Some of the metrics that are stored include badClicks and goodClicks. The length of time a searcher stays on the target page and the information on how many other pages they view there and at what time (Chrome data) are most likely included in this evaluation.
• A short detour to a search result and a quick return to the search results and further clicks on other results can increase the number of bad clicks. The search result that had the last “good” click in a search session is recorded as the lastLongestClick.
• The data is squashed (i.e., condensed), so that it is statistically normalized and less susceptible to manipulation.
• If a page, a cluster of pages or the start page of a domain generally has good visitor metrics (Chrome data), this has a positive effect via NavBoost. By analyzing movement patterns within a domain or across domains, it is even possible to determine how good the user guidance is via the navigation.
• Since Google measures entire search sessions, it is theoretically even possible in extreme cases to recognize that a completely different document is considered suitable for a search query. If a searcher leaves the domain that they clicked on in the search result within a search and goes to another domain (because it may even have been linked from there) and remains there as the recognizable end of the search, this “end” document could be flushed to the front via NavBoost in the future, provided it is available in the selection ring set. However, this would require a strong statistically relevant signal from many searchers.

Let’s first examine clicks in search results. Each ranking position in the SERPs has an average expected click-through rate (CTR), serving as a performance benchmark. For example, according to an analysis by Johannes Beus presented at this year’s CAMPIXX in Berlin, the organic Position 1 receives an average of 26.2% of clicks, while Position 2 gets 15.5%.

If a snippet’s actual CTR significantly falls short of the expected rate, the NavBoost system registers this discrepancy and adjusts the ranking of the DocIDs accordingly. If a result historically generates significantly more or fewer clicks than expected, NavBoost will move the document up or down in the rankings as needed (see Figure 6).

This approach makes sense because clicks essentially represent a vote from users on the relevance of a result based on the title, description and domain. This concept is even detailed in official documents, as illustrated in Figure 7.

Since our pencil document is still new, there are no available CTR values yet. It’s unclear whether CTR deviations are ignored for documents with no data, but this seems likely, as the goal is to incorporate user feedback. Alternatively, the CTR might initially be estimated based on other values, similar to how the quality factor is handled in Google Ads.

• SEO experts and data analysts have long reported that they have noticed the following phenomenon when comprehensively monitoring their own click-through rates: If a document for a search query newly appears in the top 10 and the CTR falls significantly short of expectations, you can observe a drop in ranking within a few days (depending on the search volume). 
• Conversely, the ranking often rises if the CTR is significantly higher in relation to the rank. You only have a short time to react and adjust the snippet if the CTR is poor (usually by optimizing the title and description) so that more clicks are collected. Otherwise, the position deteriorates and is subsequently not so easy to regain. Tests are thought to be behind this phenomenon. If a document proves itself, it can stay. If searchers don’t like it, it disappears again. Whether this is actually related to NavBoost is neither clear nor conclusively provable.

Based on the leaked information, it appears that Google uses extensive data from a page’s “environment” to estimate signals for new, unknown pages.

For instance, NearestSeedversion suggests that the PageRank of the home page HomePageRank_NS is transferred to new pages until they develop their own PageRank. Additionally, pnavClicks seems to be used to estimate and assign the probability of clicks through navigation.

Calculating and updating PageRank is time-consuming and computationally intensive, which is why the PageRank_NS metric is likely used instead. “NS” stands for “nearest seed,” meaning that a set of related pages shares a PageRank value, which is temporarily or permanently applied to new pages.

It’s probable that values from neighboring pages are also used for other critical signals, helping new pages climb the rankings despite lacking significant traffic or backlinks. Many signals are not attributed in real-time but may involve a notable delay.

• Google itself set a good example of freshness during a hearing. For example, if you search for “Stanley Cup,” the search results typically feature the famous mug. However, when the Stanley Cup ice hockey games are actively taking place, NavBoost adjusts the results to prioritize information about the games, reflecting changes in search and click behavior.
• Freshness does not refer to new (i.e., “fresh”) documents but to changes in search behavior. According to Google, there are over a billion (that’s not a typo) new behaviors in the SERPs every day! So every search and every click contributes to Google’s learning. The assumption that Google knows everything about seasonality is probably not correct. Google recognizes fine-grained changes in search intentions and constantly adapts the system – which creates the illusion that Google actually “understands” what searchers want. 

The click metrics for documents are apparently stored and evaluated over a period of 13 months (one month overlap in the year for comparisons with the previous year), according to the latest findings. 

Since our hypothetical domain has strong visitor metrics and substantial direct traffic from advertising, as a well-known brand (which is a positive signal), our new pencil document benefits from the favorable signals of older, successful pages. 

As a result, NavBoost elevates our ranking from 14th to 5th place, placing us in the “blue ring” or top 10. This top 10 list, including our document, is then forwarded to the Google Web Server along with the other nine organic results.

• Contrary to expectations, Google does not actually deliver many personalized search results. Tests have probably shown that modeling user behavior and making changes to it delivers better results than evaluating the personal preferences of individual users. 
• This is remarkable. The prediction via neural networks is now better suited to us than our own surfing and clicking history. However, individual preferences, such as a preference for video content, are still included in the personal results. 

The GWS: Where everything comes to an end and a new beginning

The Google Web Server (GWS) is responsible for assembling and delivering the search results page (SERP). This includes 10 blue links, along with ads, images, Google Maps views, “People also ask” sections and other elements.

The Tangram system handles geometric space optimization, calculating how much space each element requires and how many results fit into the available “boxes.” The Glue system then arranges these elements in their proper places.

Our pencil document, currently in 5th place, is part of the organic results. However, the CookBook system can intervene at the last moment. This system includes FreshnessNode, InstantGlue (reacts in periods of 24 hours with a delay of around 10 minutes) and InstantNavBoost. These components generate additional signals related to topicality and can adjust rankings in the final moments before the page is displayed.

Let’s say a German TV program about 250 years of Faber-Castell and the myths surrounding the word “pencil” begins to air. Within minutes, thousands of viewers grab their smartphones or tablets to search online. This is a typical scenario. FreshnessNode detects the surge in searches for “pencil” and, noting that users are seeking information rather than making purchases, adjusts the rankings accordingly. 

In this exceptional situation, InstantNavBoost removes all transactional results and replaces them with informational ones in real time. InstantGlue then updates the “blue ring,” causing our previously sales-oriented document to drop out of the top rankings and be replaced by more relevant results.

Unfortunate as it may be, this hypothetical end to our ranking journey illustrates an important point: achieving a high ranking isn’t solely about having a great document or implementing the right SEO measures with high-quality content. 

Rankings can be influenced by a variety of factors, including changes in search behavior, new signals for other documents and evolving circumstances. Therefore, it’s crucial to recognize that having an excellent document and doing a good job with SEO is just one part of a broader and more dynamic ranking landscape.

The process of compiling search results is extremely complex, influenced by thousands of signals. With numerous tests conducted live by SearchLab using Twiddler, even backlinks to your documents can be affected.

These documents might be moved from HiveMind to less critical levels, such as SSDs or even TeraGoogle, which can weaken or eliminate their impact on rankings. This can shift ranking scales even if nothing has changed with your own document.

Google’s John Mueller has emphasized that a drop in ranking often doesn’t mean you’ve done anything wrong. Changes in user behavior or other factors can alter how results perform.

For instance, if searchers start preferring more detailed information and shorter texts over time, NavBoost will automatically adjust rankings accordingly. However, the IR score in the Alexandria system or Ascorer remains unchanged.

One key takeaway is that SEO must be understood in a broader context. Optimizing titles or content won’t be effective if a document and its search intent don’t align.

The impact of Twiddlers and NavBoost on rankings can often outweigh traditional on-page, on-site or off-site optimizations. If these systems limit a document’s visibility, additional on-page improvements will have minimal effect.

However, our journey doesn’t end on a low note. The impact of the TV program about pencils is temporary. Once the search surge subsides, FreshnessNode will no longer affect our ranking, and we’ll settle back at 5th place. 

As we restart the cycle of collecting click data, a CTR of around 4% is expected for Position 5 (based on Johannes Beus from SISTRIX). If we can maintain this CTR, we can anticipate staying in the top ten. All will be well.

Key SEO takeaways

• Diversify traffic sources: Ensure you receive traffic from various sources, not just search engines. Traffic from less obvious channels, like social media platforms, is also valuable. Even if Google’s crawler can’t access certain pages, Google can still track how many visitors come to your site through platforms like Chrome or direct URLs.
• Build brand and domain awareness: Always work on strengthening your brand or domain name recognition. The more familiar people are with your name, the more likely they are to click on your site in search results. Ranking for many long-tail keywords can also boost your domain’s visibility. Leaks suggest that “site authority” is a ranking signal, so building your brand’s reputation can help improve your search rankings.
• Understand search intent: To better meet your visitors’ needs, try to understand their search intent and journey. Use tools like Semrush or SimilarWeb to see where your visitors come from and where they go after visiting your site. Analyze these domains – do they offer information that your landing pages lack? Gradually add this missing content to become the “final destination” in your visitors’ search journey. Remember, Google tracks related search sessions and knows precisely what searchers are looking for and where they’ve been searching.
• Optimize your titles and descriptions to improve CTR: Start by reviewing your current CTR and making adjustments to enhance click appeal. Capitalizing a few important words can help them stand out visually, potentially boosting CTR; test this approach to see if it works for you. The title plays a critical role in determining whether your page ranks well for a search phrase, so optimizing it should be a top priority.
• Evaluate hidden content: If you use accordions to “hide” important content that requires a click to reveal, check if these pages have a higher-than-average bounce rate. When searchers can’t immediately see they’re in the right place and need to click multiple times, the likelihood of negative click signals increases.
• Remove underperforming pages: Pages that nobody visits (web analytics) or that do not achieve a good ranking over longer periods of time should be removed if necessary. Bad signals are also passed on to neighboring pages! If you publish a new document in a “bad” page cluster, the new page has few chances. “deltaPageQuality” apparently actually measures the qualitative difference between individual documents in a domain or cluster.
• Enhance page structure: A clear page structure, easy navigation and a strong first impression are essential for achieving top rankings, often thanks to NavBoost.
• Maximize engagement: The longer visitors stay on your site, the better the signals your domain sends, which benefits all of your subpages. Aim to be the final destination by providing all the information they need so visitors won’t have to search elsewhere. 
• Expand existing content rather than constantly creating new ones: Updating and enhancing your current content can be more effective. ContentEffortScore measures the effort put into creating a document, with factors like high-quality images, videos, tools and unique content all contributing to this important signal.
• Align your headings with the content they introduce: Ensure that (intermediate) headings accurately reflect the text blocks that follow. Thematic analysis, using techniques like embeddings (text vectorization), is more effective at identifying whether headings and content match correctly compared to purely lexical methods.
• Utilize web analytics: Tools like Google Analytics lets you track visitor engagement effectively and identify and address any gaps. Pay particular attention to the bounce rate of your landing pages. If it’s too high, investigate potential causes and take corrective actions. Remember, Google can access this data through the Chrome browser.
• Target less competitive keywords: You can also focus on ranking well for less competitive keywords first and thus more easily build up positive user signals.
• Cultivate quality backlinks: Focus on links from recent or high-traffic pages stored in HiveMind, as these provide more valuable signals. Links from pages with little traffic or engagement are less effective. Additionally, backlinks from pages within the same country and those with thematic relevance to your content are more beneficial. Be aware that “toxic” backlinks, which negatively impact your score, do exist and should be avoided.
• Pay attention to the context surrounding links: The text before and after a link, not just the anchor text itself, are considered for ranking. Make sure the text naturally flows around the link. Avoid using generic phrases like “click here,” which has been ineffective for over twenty years. 
• Take note of the Disavow tool’s limitations: The Disavow tool, used to invalidate bad links, is not mentioned in the leak at all. It seems that algorithms do not consider it, and it serves mainly a documentary purpose for spam fighters. 
• Consider author expertise: If you use author references, ensure they are also recognized on other websites and demonstrate relevant expertise. Having fewer but highly qualified authors is better than having many less credible ones. According to a patent, Google can assess content based on the author’s expertise, distinguishing between experts and laypeople.
• Create exclusive, helpful, comprehensive and well-structured content: This is especially important for key pages. Demonstrate your genuine expertise on the topic and, if possible, provide evidence of it. While it’s easy to have someone write content just to have something on the page, setting high ranking expectations without real quality and expertise may not be realistic.

A version of this article was originally published in German in August 2024 in Website Boosting, Issue 87.

Courtesy of Search Engine Land: News & Info About SEO, PPC, SEM, Search Engines & Search Marketing

The complete guide to optimizing content for SEO (with checklist)

Monday, August 12th, 2024

Creating high-quality, helpful content that drives search traffic is more important than ever.

There’s nothing worse than pouring your heart and soul into a wonderful guide or blog post that nobody ever sees or churning out pieces with little value that disappoint or alienate your potential audience.

Standing out from the crowd requires a delicate balance between optimizing content for search and creating pieces that truly resonate with users. 

This comprehensive guide will walk you through the essential steps to achieve this balance, ensuring your content ranks well and provides genuine value to your audience.

Follow along and check off each step as you go to create content that both search engines and readers will love.

Step 1: Optimize your content strategy for search

Your SEO and content strategy should complement each other. Leaning too heavily on one or the other could impact your share of organic traffic, so a balanced approach is best for optimal search engine results.  

Content marketing is a competitive game, so you must also ensure your piece is memorable or “sticky.” A great way to review this is by using the “Made to Stick SUCCESS model” to ensure your content has as many of these traits as possible:

• Simple. 
• Unexpected.
• Concrete. 
• Credible. 
• Emotional. 
• Stories. 

High-quality, helpful content: 

• Understands your audience but takes search data into account: At its heart, great content is written for people but also performs well in search, helping your audience discover it. Here are some tips on writing for people and search engines. 
• Helps the reader complete one specific task: Be thorough yet focused. If you find there’s too much to say, consider creating a hub and spoke model with a series of interlinked articles. 
• Connects with other content: Consider how a planned topic relates to the content you have already created or other pieces you plan to create.
• Guides your readers toward an action: Consider the place in the user’s journey you’re supporting and guide them appropriately.
• Is sticky: How will your content be remembered by your audience? Does it have some striking or unusual elements that help it to stick in people’s minds?

Step 2: Research and design helpful content

SEO is only successful with thorough research, which should help shape your ideas before you start creating content. 

It’s important to go beyond search volume, using data to help you better understand your target audience. 

Before you start to write content for search engines or users, you should carry out:

• Keyword research: Understanding the search queries that relate to your topic can help you learn about your audience and also give your search engine marketing a solid foundation. 
• Audience research: What are your users’ pain points and challenges? What does their journey look like? There are various ways to understand the people you are creating for, but keeping them at the forefront of your mind when creating content is important. 
• User intent analysis: Creating content that doesn’t match the search intent will likely set you up for failure. Understand what users are looking for before you begin, and make sure you are fulfilling their needs. 
• Entity research: On-page SEO is more than just keywords. It involves understanding the connections related to your topic to improve search results.
• Content gap analysis: checking where your website or piece of content is lacking compared to the competition is a really useful market research exercise. 
• A review of the helpful content guidelines: Follow Google’s guidelines for creating helpful, reliable content. The better you understand what search engines want, the more likely you will succeed.

Step 3: Human vs. AI content generation

AI isn’t a shortcut to writing content, and high-quality content always needs a human touch. Most of today’s content likely uses both humans and AI as part of the process, and a combined approach can be effective. 

If you have used AI during your content creation process, you should: 

• Personalize and humanize: Make sure it has a human touch. Add some personal experience, unique perspectives and even emotional content where suitable. Find out how to make your AI content sound more human.  
• Fact-check: AI is programmed on data and writes based on common patterns rather than information accuracy. It can’t yet verify its claims, so it’s crucial to fact-check any content you’ve generated with AI. It could risk your or your company’s reputation if you publish incorrect information. 
• Check for bias: Thoroughly review your content for any bias that may have been introduced by the use of AI. 
• Review for overuse of keywords or certain phrases: AI-generated content often repeats certain phrases and keywords. This can affect the overall feel of your content, making it appear mechanical to readers and hampering its chances of success. 

Step 4: Create correct, accessible and readable content

When you’re working on content, there are some basics that shouldn’t be ignored. Good spelling, grammar and readability are a must. 

Helpful content of a high standard should always be:

• Free of spelling and grammatical errors: While spelling mistakes and errors won’t directly affect your chances of ranking, they will erode trust and credibility. It’s important to proofread and get others to check your work too. 
• Easy to follow and read: This helps users and also makes your content more search engine-friendly. Search engines read by parsing text and complicated sentence structures can lose their meaning because they are more difficult to process. 
• Structured well: Content should make appropriate use of headings and formatting. This will help skim readers find the information they’re looking for quickly and easily. It’s also helpful for those using screen readers, making sure it remains accessible to all.    
• Enticing to click on: Don’t forget about the title tag and meta description. While meta descriptions aren’t a ranking factor, they are a key area to help your content stand out in the search engine results pages. 
• Up to date: Ensure your information, statistics and references are current to keep your content relevant.

---

Step 5: Keyword and entity usage

If you’ve followed the steps above, you’ll be armed with a content plan based on research and many relevant keywords. But you can still fall into common traps, like focusing too much on a primary keyword or overusing search terms.   

Helpful, great quality website content should:

• Make natural use of keywords: Try not to overuse your target keyword. Avoid keyword stuffing by aiming for natural use. If a word or phrase doesn’t feel right, don’t try to make it work.
• Include synonyms and close variants. There are many different ways of saying the same thing. Using equivalent words can make your writing more varied and interesting and avoid repetition. 
• Include related terms: Every topic includes a series of related keywords frequently used together. This frequency of use helps to contextualize which in turn helps search engines understand your meaning. 
• Use key entities and related entities: The “things” related to your topic can help you create a more accurate, complete or well-rounded web page. Entity SEO helps search engines by providing context and disambiguation, helping you to optimize content effectively.   

Step 5: Consider E-E-A-T

Having strong E-E-A-T (experience, expertise, authoritativeness and trustworthiness) can greatly impact SEO success. Trustworthy, authoritative content should:  

• Demonstrate experience: Use author bios, “checked by” references or personal experience to show the author and publisher entities behind your content. Use first-hand experience and first-person pronouns where appropriate, this has been found to have a strong correlation with SEO success in a recent study. 
• Display expertise: Create in-depth, accurate content that shows you really understand a topic. Go beyond the surface level and add insight and information that only subject matter experts are likely to know. 
• Build authoritativeness: Content must be on-topic to help build your authority in a certain field. Create something that others are likely to reference for an extra authority boost. 
• Be trustworthy and transparent: Keep things straightforward and open. Don’t try to mislead or make claims you can’t support. Be clear about who is behind the content. 

Step 6: Add multimedia elements

Depending on your topic, other media may help to enrich your content for users and improve your SEO efforts. 

Check the search engines to see what type of content they are already ranking for a certain search query. Perhaps video or visual content is prevalent. 

If so, make sure your content includes some original video or imagery too, where possible. Consider whether your pages would benefit from:

• Images: Imagery can inspire and enhance text-heavy content. However, stock images can negatively impact search rankings, so use images thoughtfully rather than just to fill space.
• Diagrams and graphics: Sometimes words aren’t enough or can be complemented by a graphic or diagram. Take instructions for example, it can be easier to see how something works than to simply read the words. 
• Video: Ideal for “how to’s”, showing a product or process in greater detail or helping users to see movement of any kind. Video can be a really powerful way to enhance content if used appropriately. 
• Audio: Less widely used, audio elements can add to storytelling or create an emotional response to enrich written content. 

Step 7: A piece of a bigger picture

Every piece of content is related to others. Great content will add to your area of expertise and will relate to other pieces of content on your site. 

It will also draw references from other trusted resources on a topic. And in time, it might become a piece that others refer to.

While creating your content, it’s important to consider:

• Internal linking: How does this piece relate to other content you have produced? Have you created internal links, connected related pages and made it easy for readers to navigate to other pages that would enhance their journey?
• External linking: Are any facts, statistics or quotes backed up by links to their original sources? Are there other useful resources by experts that might provide further detail for your readers?
• Backlink potential: Even at the early stages, consider your audience and whether there is any potential to gain backlinks from other websites. Have you included original information or quotes? Perhaps you have a unique perspective, new expertise or even artwork that sets your content apart. 
• Is it shareable?: Social shares can help you reach a larger audience. Have you created something others would want to share, and have you made it easy for them to do so?

Step 8: Technical SEO content issues

While technical SEO issues can affect your whole site, there are some specific ones to look out for to help optimize for SEO performance. As a quick overview, you should make sure:

• Content loads quickly: Optimize images and ensure other media files aren’t slowing down the content’s performance.
• Displays well on mobile: With a responsive site, this shouldn’t be too much of an issue. But check how your content looks and responds on mobile. You might need to break down large paragraphs of text or improve formatting. Make sure all elements can still be clicked easily.  
• Page is included in the site’s XML sitemap: This helps search engine crawlers find and understand the content.
• URLs are descriptive: Include wording rather than non-descript numbers or letters in the URLs. 
• Content passes Core Web Vitals: An indication of user experience and a ranking factor, you should check and resolve any issues with core web vitals. 
• Includes appropriate schema markup: Add markup for entities, FAQs, reviews, article elements like authorship and publishing date. Keep it relevant, though; don’t include structured data markup that doesn’t relate to the content on that page. 

Step 9: Review and rework

The best content is never finished. The world changes fast, and that leaves content online that is dated, inaccurate or, at worst, misleading. It’s important to be aware of this and revisit your content often. 

Here are some of the key times to evaluate your content: 

• Create a schedule: It might be as simple as having your own set timescale for reviewing content. Depending on how fast-moving your industry is, that might be monthly, quarterly or even annually. 
• Keep an eye on industry updates: Any big changes in your industry could cause information to become inaccurate or even create new areas that should be covered in your content.  
• Evaluate after internal changes: Updates within your company can have an impact, too. Perhaps you’ve stopped selling a particular product, changed your internal focus or added new services to your offering. Good content should reflect this accurately. 
• Assess after Google algorithm updates: When search engine algorithms have updated, it’s always worth checking how this has impacted your content. Look at other things to do after a core update for more information. 

Why content quality matters

Today’s search results are in danger of being cluttered with AI-generated content.

It’s also more competitive than ever because it’s possible to create faster than ever. That’s why the quality of your content matters. If it’s high quality and helpful, it’s likely to:

• Rank better and drive more traffic to your website.
• Fulfill a need so users will be satisfied and may progress through their user journey.
• Add to your authority to build trust and credibility in your business both online and offline. 
• Increase brand visibility by slowly building up your presence and creating something valuable in your niche. 
• Gain your audience’s trust by being genuinely helpful, so even if you’re not clinching a sale right now, you’re likely to be the place they go when they’re ready to take the next step. 

Ultimately, quality content is about your reputation. But it’s also crucial for gaining a competitive edge in the search engine results pages.

Next time you’re optimizing content, remember to think past keyword ideas and take the big picture into account. 

Courtesy of Search Engine Land: News & Info About SEO, PPC, SEM, Search Engines & Search Marketing

The high cost of PPC efficiency

Monday, August 12th, 2024

Addressing the tension between running efficient paid search campaigns and driving business growth is crucial. While building paid search accounts has become easier, using them effectively to grow a business is more challenging than ever. 

Focusing solely on optimizing for maximum return on ad spend (ROAS) and the lowest cost-per-lead (CPL) may yield short-term gains, but overemphasizing these efficiency metrics can hinder long-term business growth.

Understanding the efficiency vs. growth trade-off lets you make smarter decisions for your business and create paid marketing campaigns that align with long-term business goals.

The importance of aligning paid search with business goals

Search campaigns, once known for delivering highly efficient, immediate returns, have long been the cornerstone of the paid search industry.

Google Ads’ real-time bidding model and emphasis on transactional search intent made search campaigns groundbreaking for marketers, especially since this channel was historically highly measurable.

However, PPC has evolved, and today’s advertisers face a more complex environment. Increasingly complex match types, a broader range of advertising channels and new all-encompassing campaign types like Performance Max and Demand Gen have challenged the channel’s traditional efficiency.

While cost efficiency is important, a narrow focus hinders long-term growth and makes advertising repetitive and uninspired. The outdated mindset of “I’ll give you $5, and you must give me $10 in return,” oversimplifies the complex dynamics of effective marketing.

Successful paid search advertising in 2024 requires a balanced approach that aligns with overall business goals. Marketing needs to be centered around customer acquisition, retention, brand building and, ultimately, creative campaigns that move humans toward action.

The pitfalls of overemphasizing efficiency

Short-term gains and long-term growth are often at odds in paid search marketing. While a singular focus on metrics like the lowest CPL and the highest ROAS can yield immediate results, these metrics may also restrict an advertiser’s ability to grow the business. 

This narrow focus often leads to a conservative approach that prioritizes lower-funnel campaign types at the expense of upper-funnel campaigns. 

Building brand awareness and nurturing potential customers through upper-funnel efforts is essential for sustained growth. However, the challenge of directly attributing revenue to campaigns like YouTube has made these ads less attractive to many advertisers. 

As a result, the industry has seen a reliance on short-term, conversion-focused campaigns, which are very binary and limiting to business growth. 

The value of upper-funnel campaigns

Upper-funnel campaigns are crucial in building brand equity and nurturing potential customers. At a time when there is more AI-generated bland content than ever before, humans are craving connection, humanity and community. 

Video is a fantastic way to start building this familiarity and trust and increase the likelihood that customers will consider a brand when they are ready to purchase. Leverage YouTube advertising to be seen as a leader and trusted resource. 

Upper funnel campaigns attract leads in the early stages of the buyer’s journey or even those who are problem-unaware. Advertisers showcasing engaging content differentiate themselves and make a memorable impression.

PPC expert Julie F. Bacchini highlights the importance of upper-funnel strategies despite their less immediate returns. 

On LinkedIn, there was a discussion about Solo Stove’s partnership with Snoop Dogg – a campaign that some considered a “marketing miss.” 

While it increased brand awareness, the lack of immediate sales uplift led to the CEO’s dismissal, reflecting a culture focused on short-term metrics.

Balancing efficiency and growth

To strike a balance between efficiency and growth, advertisers should:

Diversify campaigns

Incorporate upper-funnel, mid-funnel and lower-funnel campaigns to create a comprehensive strategy that supports both immediate and long-term goals. Also, recognize that humans want to talk to humans when evaluating which campaigns are important in a holistic strategy.

Leverage data and analytics

Use data to track key performance indicators (KPIs) across all stages of the funnel and adjust strategies accordingly. Additionally, understand that we can’t always measure every action perfectly every single time.

Take a step back and consider what metrics align with long-term business goals. While efficiency is important, it should not be the sole focus. Define other success metrics like brand awareness, customer retention and lifetime value.

Invest in creative

Allocate resources toward creating engaging and memorable ads that differentiate your brand and build trust with potential customers. Don’t rely solely on performance-based ads for short-term gains.

Invest in SEO and content marketing

Alongside paid search, invest in SEO and content marketing to build a strong foundation for future growth. As paid search platforms are increasingly saturated, it is important to have a marketing strategy that leverages all customer moments.

Focus on the big picture

Take a step back and examine the overall business goals. How can paid search help achieve those goals in the short term and the long term?

Dollar Shave Club is a brand that exemplifies this balanced approach. They disrupted the traditional razor market by focusing on a clear value proposition, humorous advertising and a subscription-based model and achieved significant growth.

When marketers consider growth in the traditional way, they focus on singular metrics. For brands like Dollar Shave Club, advertising creates consumers who return and interact with them because of a greater value than the product itself. Their marketing strategy includes an entertainment factor.

Dollar Shave Club might have faced what some would consider wasted ad spend with this strategy. However, the campaign led to significant business growth, making the investment worthwhile in the long run.

Leveraging data and analytics for balanced campaigns

Aligning metrics is important to achieving a balanced approach in paid search. Here are some key strategies to help align metrics to what matters most.

Set clear objectives and KPIs 

Define specific goals for each stage of the funnel, from awareness to conversion. This will help you track performance accurately and make informed decisions. 

For instance, setting a 10x ROAS goal for a YouTube campaign may cause it to be paused. However, by utilizing the brand lift studies available on the Google Ads platform, advertisers can justify the campaign’s effectiveness and the associated creative costs.

Utilize descriptive analytics

Analyze past performance to understand what has worked and what hasn’t, providing valuable context for the metrics. 

Rather than viewing metrics in isolation after a campaign, this approach allows advertisers to consider the results over time and gain deeper insights into the campaign’s overall impact.

Implement predictive analytics 

Use historical data to forecast future trends and outcomes, anticipate market changes and proactively adjust campaigns.

Google Ads offers a Performance Planner tool for Search campaigns and Performance Max, allowing users to forecast how their campaigns will react to adjustments in budget and objectives. 

A spreadsheet provides a comprehensive overview. Below is an example designed to forecast the impact of campaigns on platform revenue.

This tool allows decision-makers to assess their investment and impressions across various stages of the marketing funnel, helping to establish a strong presence in awareness campaigns.

While focusing solely on impressions is often discouraged, a brand aiming for growth must ensure its presence at every stage of the marketing funnel.

Using impression-based metrics helps determine the allocation of resources and avoid the tendency to overallocate resources to conversion campaigns. 

This example is adaptable to include additional channels, such as Facebook and programmatic display. A consistent forecasting approach for all paid media channels gives advertisers a holistic view of their marketing efforts and helps with informed budget planning.

Adopt prescriptive analytics

Go beyond predictions by using data to recommend the best course of action and optimize your campaigns for better results. 

Prescriptive analytics is when the paid search expert goes the extra step to recommend a specific course of action. 

Paid search is both an art and a science. Prescriptive recommendations, often framed as “do this” statements, are based on data and help advertisers focus on achieving their business goals.

For example, the prescriptive recommendation “increase budget by 20% for consideration campaigns” lets an advertiser achieve their goal of driving more conversions long term and ultimately increasing revenue over time.

Navigating the trade-offs between PPC efficiency and sustainable growth

The future of paid search lies in striking a balance between efficiency and growth. Advertisers need to adapt their strategies to keep up with changing trends. 

While immediate returns are important, investing in long-term strategies for brand awareness and customer loyalty is equally crucial.

Diversifying campaigns, leveraging data and investing in YouTube, paid social, programmatic, SEO, and content marketing helps businesses create a sustainable strategy that drives both immediate results and future growth.

Courtesy of Search Engine Land: News & Info About SEO, PPC, SEM, Search Engines & Search Marketing

How to overcome bias in PPC for better ad performance

Monday, August 12th, 2024

One reason pay-per-click (PPC) marketing is so challenging is there isn’t one right answer to any problem. 

When considering account structure, there’s an argument to be made for single keyword ad groups (SKAGs), single theme ad groups (STAGs), dynamic search ads (DSA) and Performance Max. This applies to bidding and creative as well.

However, if you rely too heavily on Google Ads strategies that seemed successful early in your PPC career, you might develop biases that limit your ability to tailor your approach to each client’s unique needs.

This article will break down:

• Major biases a strategist might have.
• When these biases are justified and when to re-examine them.
• How to safely test your assumptions without compromising your client’s profitability.

Bias 1: Smart vs. manual bidding

There are two major schools of thought on bidding, and both have pros and cons. 

Some believe manual bidding is outdated and all bidding should be automatic or smart (conversion based). Others believe smart and auto bidding perform poorly due to data limitations and potential for disproportionate bid to budget ratios. These biases often come from when someone started in digital marketing.

For example, people who started in digital marketing between the early 2000s and 2010s often prefer manual bidding because it was the main approach back then. It requires a deep understanding of the auction process and taking full responsibility for identifying which signals are important to the brand.

Conversely, those who entered the field closer to 2020 may lean toward auto and smart bidding. Ad platforms heavily promote them and don’t require as much manual intervention.

Manual bidding is typically favored by those who distrust ad platforms and prefer control, while smart bidding appeals to those who dislike micromanaging accounts and prefer efficiency. Both methods can be valid, depending on the context:

Low-volume accounts 

• Manual bidding may be necessary due to insufficient data to support smart bidding. However, bidding strategies like max clicks with a bid cap can help unlock volume. 
• Based on Optmyzr data (25,000 accounts reviewed), Max clicks with and without a bid cap beat manual on CPC, conversion rate and CPA, though manual did better on ROAS. 

High-volume accounts

• Relying exclusively on manual bidding can be unwise, as it ignores the benefits of smart bidding signals. When you can consistently get at least 60 conversions in a 30-day period, smart bidding does outperform manual bidding. 
• Optmyzr data found that across 25,000 accounts, Max Conversion Value beat Manual bidding by 400%. Manual beat out max conversions pretty handily, which is why many might be biased against smart bidding (max conversions have 30% higher adoption than max conversion value). 

Testing manual vs. smart bidding

To test these bidding strategies, you need to be able to control variables and have a risk-tolerant campaign. 

For local businesses, this can be straightforward. Simply target different locations and compare performance. For single service or product-focused accounts, choose a part of the market where some fluctuation is acceptable.

Remember that automatic or smart bidding requires a learning period of at least five days, potentially extending to 14 days. During this time, avoid making significant changes to prevent fluctuations. However, you can adjust bid floors and caps without triggering a new learning period.

If you are testing manual bidding, be prepared to make precise bid adjustments, considering audiences, devices, locations and times of day. Decide whether to adopt aggressive or conservative cost-per-click (CPC) bids and adjust accordingly.

For example, if you decide to go conservative on the bid, you might have a bid of $3 and bid adjustments of ~50%. An aggressive bid of $5 might warrant 10%–15% bid adjustments. Remember that bid adjustments are cumulative and can be positive (direct budget toward something) or negative (direct budget away from something). 

There is no definitive answer to the manual vs. smart bidding debate. The key is supporting your chosen strategy and communicating your decisions clearly with your client.

---

Bias 2: Performance Max as a branded cannibal

Performance Max campaigns have garnered mixed reactions due to their focus on visual content and initial lack of control over certain elements. 

Initially, these campaigns often drove branded queries, sparking debates about their true value. However, Performance Max has evolved, offering tools such as:

• Asset and asset group level data
• YouTube placements for potential account-level exclusions
• Campaign-level exclusions for placements, topics and negative keywords
• Generative AI tools for brand safety guidelines
• Portfolio bidding with bid caps and floors via Search Ads 360

Those who struggle with Performance Max often excel in search-first marketing, while Performance Max is designed to allocate budgets based on customer presence and budget availability. 

If visual content dominates your budget distribution, it may indicate a visual preference among your audience or budget constraints affecting search bids.

Key considerations for Performance Max

• Conversion volume: Can you achieve at least 60 conversions in a 30-day period? If not, either avoid Performance Max or allow branded traffic within the campaign and turn off stand-alone branded campaigns.
• Account structure: Decide whether multiple Performance Max campaigns with different location targets and budgets or a single campaign with multiple asset groups better suits your needs.
• Objective alignment: Ensure Performance Max campaigns focus on driving leads and sales, not top-of-funnel awareness or remarketing.

When testing Performance Max, ensuring you have enough budget for the campaign (minimum 10% of the budget) is important. If you are borrowing a budget from existing campaigns, make sure you still honor bid-to-budget ratios. 

Bias 3: Keyword structure and the future of keywords

Keywords have evolved from rigid syntax-oriented elements to signals guiding the system.

Despite this, biases around keyword structures persist, whether favoring single keyword ad groups, dynamic search ads or theme-oriented structures.

Single keyword ad groups

The basic premise of a SKAG is that you have one keyword in each ad group so you can benefit from a “perfect” keyword-to-ad-to-landing page relationship. This can either mean lots of ad groups per campaign or lots of campaigns with a single ad group with one keyword.  

These are powerful when used moderately and supported by sufficient budgets and aggressive negative keywords. However, they may struggle with low volumes. Also, if you’re not able to be surgical with your negatives, it’s very easy to make accidental duplicates.

SKAGs do best when you know exactly how your people will search and want to allocate a very specific budget to those ideas. However, be careful not to include too many ad groups in the same campaign or too many campaigns. 

The former will cause some ad groups to miss out on impressions due to which get initial impressions/conversions, while the latter will cause data threshold issues.  

Keyword match types

Broad match keywords have long since transitioned from syntax matching to intent matching. Yet even phrase and exact have close variants baked in, leading to divided opinions on how best to utilize keywords. 

Testing broad match keywords in single keyword ad group can be effective, so long as you add all other keywords as a negative. Conversely, match-type specific campaigns can lead to accidental duplicates due to how close variants work (namely, that broad and phrase can lead to an exact match or exact close variant). 

While exact match might consistently “perform better” than broad, it’s not really fair to say they have the same job. Consider the roles assigned to each entity within your account. 

• For transactional goals, you might lean toward non-broad keywords (minimum 3+ words in the keyword phrase), exact match single theme ad groups, or dynamic search ads with extensive negatives. 
• For data acquisition or ramping up an account, broader keywords and concepts may be more effective.

Addressing platform intent bias

We couldn’t discuss bias in PPC without addressing many practitioners’ bias against advice and updates from the ad platforms themselves. 

Between brands not wanting to part with profit data (even though it will improve results and reporting) and taking any action as an overstep (even if it’s as innocent as pausing keywords with no data over the past 13 months), it’s hard to see a way for brands and ad platforms to rebuild trust. 

A big source of this mistrust is if someone learns one network, they may struggle to adapt to the rules of another network. 

For example, most paid search networks function at the campaign level, while most paid social function at the ad set level. Ad networks like Google favor old entities, while Meta favors newer ones. 

All of these mechanics end up creating biases around which channels are best suited for a brand and whether the channel will actually be a good partner. 

While this bias isn’t conquerable like the other biases, we all must remember that humans work on the product teams at all ad networks, and they thrive on specific constructive feedback. 

If you’re going to test a network, make sure that you budget enough for a realistic test (time and money) and that you’re upfront with your stakeholders on what kind of reporting you can expect. 

Conclusion

Biases are an inherent part of human nature, and while we can’t eliminate them, we can identify and counteract them through objective testing. Choose one test to run as you approach the fourth quarter to challenge your biases and validate or refine your strategies. 

Embrace the flexibility of both smart and manual bidding, understand the potential of Performance Max campaigns and structure your keywords to maximize relevance and performance. 

This approach will help you keep your PPC campaigns adaptable and effective, benefiting your clients and helping them achieve their business goals.

Courtesy of Search Engine Land: News & Info About SEO, PPC, SEM, Search Engines & Search Marketing