Home |
SEO Notes by Ardan Michael Blum
Search engine optimization updates from Ardan Michael Blum, founder and principal consultant at A. Blum Localization Services, a Palo Alto–based SEO and localization consultancy founded in 2016.
Home |
Search engine optimization updates from Ardan Michael Blum, founder and principal consultant at A. Blum Localization Services, a Palo Alto–based SEO and localization consultancy founded in 2016.
September 2, 2026
Google says it completed the worldwide rollout of its dedicated generative AI performance reports in Search Console on August 31, 2026. First announced on June 3, the Search report combines impressions from AI Overviews and AI Mode; a separate report covers generative AI features in Discover. Neither report isolates the clicks or queries behind those appearances.
The word “performance” can suggest more than the reports currently measure. The Search report shows impressions—how often links to the site were shown to a user—and allows site owners to group them by page, country, device, or date. The separate Discover report offers page, country, and date dimensions. Search Console does not include data from Search Labs experiments in either report.
Google’s current documentation exposes no click, click-through-rate, or average-position metric and no query dimension in these dedicated reports. The Search report also does not distinguish between AI Overviews and AI Mode.
That does not mean AI-feature clicks go uncounted. Google says that clicking an external link from either an AI Overview or AI Mode counts as a click in the main Search results Performance report. The dedicated AI report simply does not isolate those clicks. It therefore cannot calculate an AI-specific click-through rate, identify the query behind an impression, or connect a reported AI impression directly to a click or conversion.
Do not add the figures together. The dedicated AI-impression count is already included in the ordinary Web-search total. Adding it again would double-count those impressions.
Using identical property, date, and other applicable filters, a site owner can express AI impressions as a proportion of total Web impressions. That calculation should be described as a visibility share, not a traffic share.
A reported impression means that a link appeared under Google’s counting rules. By itself, it does not establish that Google quoted the page, relied upon a particular passage, sent a visitor, or rewarded a specific optimization.
In the Search report, chart and page-table totals may also differ. The chart is aggregated by property unless a URL filter is applied, while a table grouped by page is aggregated by page. Several linked pages from the same property can therefore produce multiple page rows while contributing only one property-level impression.
For a small or local website, the cautious first use is to establish a baseline. Identify the reported pages and countries—and, for Search, devices—then compare longer periods without attributing every movement to a recent edit. Google says insufficient impressions or exclusion from generative AI features can explain why a property does not display the report, even though the worldwide rollout is complete.
Google also completed the worldwide rollout of its separate Search generative AI control on August 31. Google says inclusion is the default, while child properties inherit a parent property’s configured choice by default. Once processed, exclusion prevents the property’s links and content from appearing in the covered AI features, removing the associated opportunity for impressions and traffic.
Google says this choice is not used as a ranking or inclusion signal elsewhere in Search and does not govern AI training. For separate controls concerning training future Gemini models and certain Gemini grounding uses, Google documents the Google-Extended robots.txt token.
The reports narrow an important measurement gap without closing it. They show where and when reported AI impressions occurred, but they do not isolate AI-feature clicks, click-through rates, queries, or conversions.
Further information: Google Search Central’s announcement | Google’s AI-feature counting rules
August 17, 2026
Keyword research tools place phrases in separate rows. A website must decide whether any of those rows justify separate, long-lived URLs. These are not equivalent operations. Two phrases can have different search-volume estimates while still expressing substantially the same need.
In an August 11, 2026 article for Search Engine Land, Bharath Ravishankar argues that every proposed page should pass an independence test. The keyword should attract a meaningfully distinct set of searches, support genuinely different content, and occupy a logical place within the website.
The first investigation begins with the site’s existing visibility. If a current page already receives impressions for the proposed keyword and closely related searches, the immediate question is whether that page can be improved. Creating another URL may be unnecessary when the additional subject fits naturally within a page that Google already displays for the query.
Search Console provides useful evidence, but its reports are incomplete. Google explains that privacy protections remove some queries and that Search Console stores top data rows rather than every row. A visible connection between a query and a page is meaningful; the absence of a query from the report does not prove that the page has never appeared for it.
Ravishankar’s next test compares the leading organic results for the proposed keyword with those for the closest term already covered. Substantial overlap suggests that Google may interpret the searches similarly. The types of pages appearing in the results also matter: one query may favor broad guides while another produces specialist services, product categories, or narrowly focused resources.
This comparison is a diagnostic method rather than a documented Google rule. Google has not published a numerical overlap threshold at which two subjects must share a page. Results can also change according to time, location, language, and the composition of the results page. SERP overlap is therefore evidence to interpret, not an automatic instruction.
The framework’s strongest test is editorial rather than technical: prepare the outline before creating the URL. If the proposed page requires different evidence, examples, problems, and a different next step for the reader, it may be capable of standing alone. If its headings merely reproduce an existing page with the target phrase or audience name changed, expanding the existing page will usually produce a more coherent result.
This distinction is consistent with Google’s people-first content guidance. A useful page should serve an intended audience even if that reader reaches it without first searching Google. Search demand may reveal a subject worth addressing, but it does not by itself supply the page’s purpose or value.
A separate page also needs a place within the site. Its parent section, navigation route, and relevant internal links should be identifiable before publication. Google’s documentation confirms that links help it discover pages and understand their relevance. That does not guarantee rankings, but it supports the broader principle that a page should belong to the website rather than exist as an isolated destination for search traffic.
The available choices are not limited to publishing or abandoning the idea. A site can create a new page, expand an existing one, merge genuinely overlapping pages, or test a small group of pages built from a repeatable template. When overlapping material is consolidated, relevant redirects and updated internal links can direct users and search engines to the surviving page. Unrelated URLs should not simply be redirected to a general destination.
Publication remains a hypothesis rather than a verdict. Once a new page has been crawled, indexed, and shown often enough to produce useful evidence, its query pattern and visitor behavior can be reviewed. A successful independent page should develop a recognizable purpose. Several similar URLs receiving impressions for the same query does not automatically prove harmful “cannibalization,” but repeated switching may justify a closer examination of their content and internal linking.
The article is worth reading because it replaces the mechanical question “Does this keyword have search volume?” with the more demanding question “Does this page have a reason to exist?” Its framework is contributor analysis rather than official Google policy, but its lasting lesson is sound: finding another keyword does not remove the need for editorial judgment.
Read: “How to decide whether a keyword deserves its own page”, by Bharath Ravishankar, Search Engine Land, August 11, 2026.
Note: This update is an independent interpretation and summary of the article. Readers should consult the original for the author’s complete analysis and examples.
August 17, 2026
Search visibility inside an AI answer is not a single event. A web page can be retrieved, listed among the available sources, attached to a citation, or opened and read. Those stages are not necessarily equivalent.
In an August 17 article for Search Engine Land, Olivier de Segonzac, co-founder of the digital-marketing consultancy RESONEO, reports on his team’s attempt to reverse-engineer these stages within ChatGPT. The researchers examined approximately 1,200 ChatGPT answers, 88,000 search results, and 26,900 distinct web pages.
Their findings suggest a layered retrieval process involving a discovery index, stored copies of previously fetched pages, and occasional live page openings. Within the measured sessions, numerous pages appeared among the sources or citations without being opened during the current request.
This does not mean that their contents played no role: ChatGPT may have relied on a previously stored copy or the shorter text held in its discovery index. It does mean that visits from ChatGPT-User appearing in a website’s server logs cannot provide a complete account of its visibility.
The study’s most practical SEO finding concerns the beginning of a page. In one observed retrieval path, ChatGPT received the complete page title together with a fixed snippet of approximately 200 characters, generally drawn from the visible text around the H1 heading. Of 534 cited pages examined, 463 contained an H1, and the heading appeared in 387 of those stored snippets, or 83.6 percent.
The meta description was not used in this particular retrieval path, although it remained relevant to other paths using Google-derived results. The evidence therefore suggests that a clear H1 followed promptly by a useful opening sentence may help the system understand a page before opening it. It does not prove that changing these elements will improve citations or recommendations.
The article is worth reading because it replaces some of the vaguer claims surrounding “AI optimization” with measurements and a published methodology. Its conclusions must nevertheless be treated as dated observations rather than permanent rules. The architecture was inferred rather than documented by OpenAI, and parts of the system changed while the investigation was underway.
OpenAI’s crawler documentation confirms the separate purposes of OAI-SearchBot, ChatGPT-User, and GPTBot. It does not confirm RESONEO’s proposed retrieval architecture or cache behavior.
The durable lesson is therefore modest: keep important pages crawlable, give them precise titles and headings, and place genuinely useful information near the beginning.
Read: “Inside ChatGPT’s retrieval stack: The index, cache, and pages it actually reads”, by Olivier de Segonzac, Search Engine Land, August 17, 2026.
Note: This is an independent interpretation and summary of the article; readers should consult the original for the author’s complete findings and methodology.
August 9, 2026
A page can pass Google's Rich Results Test and still appear as an ordinary search result. Valid markup can make a page eligible for a supported rich result; Google may choose not to display it.
Structured data gives Google explicit labels for information already present on a page: an event date, an article's author, a product's availability, or an organization's legal name. Google says it uses this markup to understand page content and may also use it to enhance how a page appears in search results.
A passing test is not proof that every requirement has been met. Google's guidelines also require structured data to represent the page's visible content, remain relevant and current, and include the required properties for the feature. Misleading or unrelated markup can make a page ineligible. A test result also does not show that Google has crawled or indexed the live page.
Further reading: Google Search Central: Introduction to Structured Data and General Structured Data Guidelines.
July 27, 2026
Google updated its review snippet documentation on July 24, 2026. The date is new, but the central restriction is not. Google introduced its rule against self-serving review markup for 'LocalBusiness' and 'Organization' in 2019.
The rule addresses a common misunderstanding. A company may publish genuine customer testimonials on its website, but pages controlled by that company are not eligible for Google’s organic review-star feature when the reviews concern the company itself. This applies to both individual Review markup and an 'AggregateRating'.
The testimonials do not become eligible merely because they came from real clients. Google’s test also considers who controls the page on which the rating appears. If a review of entity A appears on entity A’s own website, Google regards the arrangement as self-serving for this search feature.
The same principle applies when the ratings arrive through an embedded third-party widget. Displaying Google Business Profile or Facebook reviews on a company website does not turn that page into an independent review source. Google’s current guidance explicitly includes such widgets within the self-serving-review rule.
This restriction does not make testimonials worthless or improper. Specific, verifiable comments can help prospective customers understand the experience of working with a business. They may explain the type of problem solved, the nature of the service, or the reasons a client valued the result. That information can support trust and conversion even when it produces no special search display.
The policy also does not remove ratings from a company’s Google Business Profile. Google’s original explanation of self-serving reviews states that the change concerns organic Search and does not affect the Business Profile. The familiar ratings shown with a Business Profile or local listing are therefore a separate matter from review snippets generated by structured data on the company’s website.
Nor does the rule prevent an independent review site from marking up reviews about other organizations. Google’s documentation still lists 'LocalBusiness' and 'Organization' as supported review subjects when a site collects reviews about entities other than itself.
Independence is only one part of eligibility. Marked-up review text and its associated rating must be readily visible on the page. If an aggregate score is used, visitors must be able to see that score as well. The review must concern a specific item rather than an undefined category or list, and a publisher should not combine ratings copied from other websites.
Google also prohibits fake reviews and undisclosed incentivized reviews. For local businesses, ratings must come directly from users rather than from editors who create, curate, or compile a score. These requirements help preserve a distinction between evidence supplied by customers and promotional claims assembled by the business.
A page can pass the Rich Results Test without becoming eligible for review stars. The tool can identify supported structured data and many technical errors, but it cannot reliably determine every policy or quality issue. Correct syntax establishes that Google can read the markup; it does not guarantee that Google will display a rich result.
Google’s general structured data guidelines make the distinction explicit. A valid implementation only makes a page eligible for consideration. Search appearance remains dependent on the feature-specific rules, the visible content, and Google’s decision about what will be useful in a particular result.
There is another useful nuance. Google’s 2019 FAQ says that a business does not have to remove self-serving review markup and will not receive a manual action merely for having it. Google simply will not display an organic review snippet for those pages. Other misleading or noncompliant uses of structured data can still result in a manual action, so the broader guidelines remain important.
A Palo Alto consultancy, clinic, restaurant, hotel, law firm, or service provider can keep a testimonial page when it helps visitors make an informed decision. Each quotation should be genuine, presented in context, and attributed with the reviewer’s permission. Any incentive should be disclosed clearly.
The company should not expect testimonials on its own domain to generate review stars for its 'LocalBusiness' or 'Organization' result. Adding more markup, moving the comments to another page, or importing them through a widget does not change who controls the publication.
Organization structured data can still describe factual information such as the company’s name, address, telephone number, logo, and website. The purpose of that markup is to help search engines understand the page, not to convert a company’s presentation of itself into independent evaluation.
The practical lesson is simple: publish testimonials for people, and use structured data only where the content and the search feature genuinely match. Review markup can describe eligible evidence, but it cannot manufacture independence.
Further information: Google review snippet documentation | Google’s 2019 explanation and FAQ | General structured data guidelines | Organization structured data | Rich Results Test
July 22, 2026
Google explicitly treats back-button hijacking as a malicious practice under its web-search spam policies. The behavior occurs when a website manipulates browser history so that a visitor cannot immediately return to the page they came from.
Instead of going back as expected, the person may be sent to an advertisement, an unsolicited recommendation, or an address never knowingly visited. Pages responsible for this interference may receive a manual spam action or an automated demotion in search results.
The cause may not be part of the website’s main template. Google warns that imported libraries, advertising platforms, plugins, and other third-party code can alter browser navigation without the publisher deliberately installing such behavior.
A practical audit should begin outside the site. Open a page from Google or another external source, press the Back button once, and repeat the test on desktop and mobile devices. Keep advertisements, consent tools, and other integrations active. If the browser does not return directly to the preceding page, determine which component changed its history.
The broader lesson is that technical SEO also concerns user control. A page should be easy to enter, understand, and leave. For Palo Alto businesses, an aggressive conversion tactic that traps a prospective customer can damage trust before it damages search visibility.
Read more: Google Search Central announcement | Google’s web-search spam policies
July 21, 2026
On July 16, Google announced that people can connect selected external services directly to AI Mode. Early examples include Instacart, Canva, and YouTube Music. Someone can assemble a shopping list, begin a design, or prepare a playlist without following the familiar path through several conventional search results.
For SEO, the important development is the shortening of the journey. A question that once produced a row of links can now lead directly into an activity. This may reduce some casual visits while making the clicks that remain more deliberate and valuable.
Organic visibility has not become irrelevant. AI Mode still needs accurate information, recognizable entities, useful pages, and trustworthy sources. However, rankings and traffic totals may reveal less on their own when part of the research or decision-making process occurs inside the search experience.
For businesses, the practical response is to describe services, locations, availability, expertise, and distinguishing details clearly. A well-built website can still support discovery, comparison, verification, and the final choice, even when some intermediate steps happen elsewhere.
The larger lesson is that search is moving from “help me find a page” toward “help me accomplish something.” Good SEO must understand the task behind the query and provide the facts and evidence people need before acting.
Read more: Google Introduces Connected Apps in AI Mode
July 20, 2026
Google has clarified how long it may take to reassess URLs after a website owner corrects a canonicalization problem. Even when the underlying issue has been resolved, related pages may remain in a duplicate cluster for up to two weeks.
Search systems use canonicalization to choose one representative URL when several locations contain identical or closely similar material. Publishers can express a preference through redirects, canonical elements, internal links, and sitemaps. These signals provide guidance, but they do not guarantee the final selection.
The clarification distinguishes between making a correction and seeing its effect. Crawlers must revisit the affected material before the relationship can be reconsidered and the index updated. An immediate change in search results should therefore not be expected.
Clear, substantial differences can help separate content more quickly. Minor wording adjustments may be insufficient when two documents are supposed to serve genuinely distinct purposes.
Search Console’s URL Inspection tool shows both the declared canonical and the version selected by Google. After resolving the cause, publishers may request indexing, although this option is subject to quotas and should be reserved for important URLs.
For Palo Alto businesses, the guidance is particularly relevant when service, location, campaign, or printer-friendly pages overlap. Verify the technical signals, strengthen any necessary distinctions, and then allow time for re-evaluation. Repeated revisions during that period may make the original problem harder to diagnose.
July 17, 2026
Google has introduced “platform properties,” a new type of Search Console property that measures how content published on Instagram, TikTok, X and YouTube performs in Google Search and Discover.
Until now, Search Console has primarily been associated with websites. The new feature gives creators and publishers a consolidated view of how their public social and video posts are being discovered through Google—even if they do not operate a separate website.
The Performance report shows clicks, impressions and other measurements. Users can filter the information to identify which posts and search queries are producing traffic. An Insights report provides a simpler view of recent trends, leading posts and the ways people are finding an account. The data can also be exported for further analysis.
Adding a platform property requires the account owner to authorize the connection through Search Console. Google currently supports Instagram, TikTok, X and YouTube. The feature is being introduced gradually, so it may not yet appear in every Search Console account.
This is a reporting change, not a new ranking shortcut. Connecting an account does not make its posts more likely to rank. What it does provide is evidence: creators can now see whether Google is finding their social content, which subjects attract search traffic and which posts continue to receive attention after their initial publication.
For Palo Alto businesses and independent publishers, the new reports may reveal searches that are difficult to see through native social-media statistics alone. A locally focused video, photograph or short post may appear in Google for a useful regional query. That information can then help shape future posts and more substantial website content. Social platforms remain rented space, but Search Console now makes their relationship with Google Search considerably easier to measure.
Source: Google Search Central: See How Content From Social and Video Platforms Performs on Google Search (published July 7, 2026).
July 15, 2026
Google has published expanded guidance on optimizing websites for generative AI features such as AI Overviews and AI Mode. Its central message is refreshingly direct: visibility in Google’s AI-generated results still begins with conventional SEO.
Google explains that these features draw information from its existing Search index, using techniques such as retrieval-augmented generation and “query fan-out.” A page therefore needs to be crawlable, indexed, and eligible to appear with a search snippet before it can be considered for these AI experiences.
The guidance also pushes back against several widely promoted “GEO” and “AEO” tactics. Google says website owners do not need special AI markup, unusually fragmented content, or a separate writing style designed for machines. It specifically states that llms.txt files neither improve nor damage visibility in Google Search because Google does not use them. This statement applies to Google Search; other AI services may handle such files differently.
What does Google recommend instead? Create original, non-commodity material based on real knowledge or experience. Maintain a clear technical structure, reduce unnecessary duplication, use relevant images and video, and keep important content publicly accessible to crawlers. Structured data remains useful for established search features such as rich results, but Google says there is no special schema required for generative AI visibility.
The practical lesson is that AI search has not replaced SEO. It has increased the importance of being retrievable, understandable, and genuinely worth citing. A first-hand case study, local observation, original photograph, or expert explanation offers more potential value than another generic article assembled from information already available elsewhere.
For Palo Alto businesses, this means that accurate Google Business Profile information, clearly described services, locally grounded content, and evidence of genuine experience may be more useful than chasing an entirely new collection of “AI optimization” tricks.
Source: Google Search Central: Optimizing Your Website for Generative AI Features (updated July 10, 2026).
July 09, 2026
“Palo Alto” does not always mean the city in search results. It may refer to Palo Alto Networks, Stanford-related places, restaurants, films, books, or other businesses that use the name.
A genuinely local page therefore needs more than repeated mentions of “Palo Alto.” Streets, neighborhoods, institutions, landmarks, transportation, and civic details help establish which Palo Alto the page actually describes. Local relevance is not a keyword count. It is a web of recognizable relationships.
Read further → Local Relevance Is a Proof Problem
This page was created independently of the individuals and organizations discussed, none of whom had editorial control over its contents. It contains no affiliate links, sponsored content, paid placements, or compensated endorsements. Neither the author nor this website received any payment, free or discounted product or service, preferential access, travel, hospitality, gift, or other material benefit connected with this page. Unless expressly disclosed otherwise, neither the author nor this website is affiliated with, sponsored by, endorsed by, or officially connected with any individual or organization mentioned. Names and trademarks are used only to identify the subjects discussed. A mention does not, by itself, constitute a recommendation or endorsement.
Please review the website's full Privacy Policy, Disclaimer, and Further Terms.
For project inquiries, corrections, accessibility assistance, or general questions:
Ardan Michael Blum
345 Forest Avenue
Palo Alto, 94301, California, United States
Telephone: +1 (650) 427-9358
Online: Contact Form