Updated September 17, 2026: Google has now completed the worldwide rollout of both its Generative AI performance report and its Search generative AI control in Search Console. That changes an important part of the measurement problem. SEO teams no longer have to treat Google AI visibility as entirely hidden, although the dedicated report still measures impressions rather than the full path from exposure to revenue.
The useful question is not whether Google has created an entirely new discipline called GEO or AEO. Google explicitly says that optimisation for its generative AI search features is still SEO. The operational change is narrower, and more important: teams need to separate eligibility and visibility, website visits, and business value. Those three layers can move in different directions.
That distinction matters most for informational content. A page can remain indexable, appear in AI Overviews or AI Mode, and still attract fewer visits if the generated answer satisfies the first part of the user’s need. The response should not be to rewrite every page for an AI system. It should be to decide which pages still have a defensible role once a generic summary is easy to produce.
The August 31 Update Changes the Measurement Question
Google announced its dedicated Generative AI performance reports on June 3, 2026 and initially rolled them out to a subset of sites. As of August 31, Google says the Search report has been rolled out to websites worldwide. The report covers impressions in AI Overviews and AI Mode and allows analysis by page, country, device and date. It does not replace the normal Performance report, and it does not turn AI visibility into a complete attribution model.
The practical limitation is easy to miss. The dedicated report is primarily an impression report. Google still counts clicks from AI Overviews and AI Mode in the normal Search Console performance data, but the generative AI report itself is designed to show where AI-feature visibility occurred. If a property does not show meaningful data, low impression volume can be one reason.
Google also completed the worldwide rollout of the Search generative AI control on August 31. The default is to include a site’s links and content in supported generative AI features. A site owner can instead exclude the site, which prevents its links and content from appearing in those features and removes the related opportunity for traffic and impressions.
This turns crawler and AI policy into a business decision rather than a slogan. A publisher concerned about reuse may weigh the trade-off differently from a service business that depends on discovery. The first question is therefore not “Should we block AI?” It is “What do we gain or give up if this site is excluded from Google’s generative AI surfaces?”
Google’s AI Guidance Narrows the Content Strategy
Google’s current generative AI search optimisation guide is unusually useful because it removes several distractions. Google says AI Overviews and AI Mode are rooted in its core Search ranking and quality systems, and that a page must be indexed, eligible to appear with a snippet and included in Search generative AI features to be eligible for these experiences. Meeting those conditions still does not guarantee crawling, indexing or appearance.
The same guide puts strong emphasis on valuable, non-commodity content. Google’s examples contrast first-hand or expert-led material with summaries that simply restate information already available elsewhere. This is not a requirement for every article to contain proprietary research. It is a reason to stop treating word count, keyword coverage or another rewritten definition as sufficient differentiation.
What Google does not require
- No special AI-specific schema is required for AI Overviews or AI Mode.
- Google Search does not require an
llms.txtfile for visibility, and says such a file neither helps nor harms Google Search visibility. - There is no requirement to split pages into tiny “chunks” so Google’s AI systems can understand them.
- Pages do not need to be rewritten in a special style for generative AI search.
- Creating many pages around fan-out query variants primarily to manipulate rankings or generated responses can conflict with Google’s scaled content abuse policy.
That removes a fair amount of busywork. A content team is better served by checking whether the page adds a point of view, evidence, utility or decision support that a generic answer does not provide. For Google Search, an AI-labelled tactic is not automatically a new tactic.
The practical test is whether the page changes a decision
Non-commodity value can take several forms: original data, a documented workflow, a comparison model, a useful tool, an expert explanation of a constraint, a real screenshot, or a market-specific example supported by evidence. The common feature is not novelty for its own sake. The information changes what the reader understands, verifies or does next.
This is where many content audits go wrong. Teams often ask whether an article is comprehensive when the more useful question is whether it is replaceable. A 3,000-word page can still be commodity content. A shorter page can be defensible if it contains the evidence or judgement the user actually needs.
Zero-Click Evidence Matters, but It Does Not Justify Site-Wide Assumptions
Pew Research Center analysed March 2025 browsing behaviour from 900 US adults. In that sample, users clicked a traditional Google result in 8% of visits when an AI summary appeared, compared with 15% when no AI summary appeared. A link inside the AI summary itself was clicked in 1% of visits where a summary was present.
Those numbers are useful, but they are not a universal CTR forecast. The research covers one US panel, one month and an earlier stage of Google’s AI search product. Query type, country, device, brand strength, SERP layout and user intent can all change the result. The study is best used as evidence that click behaviour can change, not as a percentage to apply across every site.
MOCOBIN’s separate analysis of AI Overviews and publisher traffic goes deeper into that traffic question. For this page, the more important consequence is operational: do not diagnose an AI problem from a traffic decline alone. A ranking change, seasonality, competitor movement, SERP redesign or a shift in demand can create a similar pattern.
Measure Eligibility, Visibility and Value Separately
A single “AI visibility score” compresses too many different questions. Google now provides enough first-party data to build a cleaner measurement sequence.
| Layer | What to check | What it tells you | What it does not prove |
|---|---|---|---|
| Eligibility | Indexability, snippet eligibility, Search generative AI control | Whether the site can be considered for supported generative AI features | That Google will show the page |
| AI visibility | Generative AI performance report impressions by page, country, device and date | Where URLs from the site appeared in AI Overviews or AI Mode | Clicks, conversions or causal impact |
| Search acquisition | Normal Search Console clicks, impressions, CTR and position by relevant query and landing-page groups | Whether organic visits and search behaviour changed | That AI features caused the change |
| Business value | Analytics, leads, sales, subscriptions, assisted journeys and other first-party outcomes | Whether search exposure and visits contribute to the business | Perfect attribution to one search feature |
I would not start by counting AI citations across dozens of third-party tools and calling the resulting number a KPI. Manual observation and external tools can be useful for research, but Google warns that third parties do not have access to its internal ranking or AI systems. First establish what first-party Search Console and analytics data can tell you, then use outside observations to explain gaps rather than manufacture precision.
A Page-Level Content Portfolio Is More Useful Than an “AI Content” Rule
The content strategy question should be decided URL by URL. Broad informational libraries are especially easy to mismanage because teams can mistake historical traffic for continuing strategic value.
Retain a page when it still performs a clear role even if immediate click demand becomes weaker. That role might be navigation, brand discovery, support for a commercial journey, a reference point for another article, or a durable explanation that users genuinely need.
Improve a page when the intent is valuable but the information is replaceable. Add evidence, expert reasoning, decision criteria, first-party material or a clearer next step. More words are not the objective.
Consolidate competing pages when several URLs answer the same question with little difference in audience or outcome. AI search does not create this problem, but it makes weak duplication harder to justify because generic summaries already reduce the value of publishing another near-identical explanation.
Retire or noindex a page when it no longer serves users, has no useful structural role and cannot be improved without recreating content that already exists elsewhere. That decision should follow a URL, redirect, canonical, sitemap and internal-link review rather than a blanket cleanup.
Trust belongs inside the same audit. Clear authorship, accurate dates, direct sourcing and editorial accountability make the page easier for readers to assess. They should describe real responsibility, not act as decorative E-E-A-T signals. MOCOBIN’s E-E-A-T guide provides broader background, but the operating rule is simple: do not invent experience to make commodity content look expert.
What Should Stay Human-Owned in an AI-Assisted Workflow
AI can reduce production time without taking ownership of the publishing decision. The boundary should follow risk and accountability rather than a fixed rule about which tool wrote which sentence.
- Suitable for assistance: source organisation, formatting variants, keyword clustering, transcription, low-risk draft preparation and repetitive data cleanup.
- Suitable for drafting with verification: research synthesis, outlines, comparison drafts, localisation drafts and first-pass editorial revisions.
- Keep human-owned: search-intent mapping, page creation versus consolidation decisions, factual sign-off, legal or commercial claims, internal-link architecture, technical prioritisation and final publication approval.
The point is not to protect manual work for its own sake. It is to keep responsibility attached to the decisions that can damage trust, create duplicate URLs or send a team in the wrong direction. A faster draft is useful only if someone still knows why the page exists and who is accountable for its claims.
A Practical Operating Sequence for SEO Teams
- Check eligibility first. Confirm important pages are indexable, eligible for snippets and included in Search generative AI features.
- Open the Generative AI performance report. Establish an impression baseline by page, country and device where data is available.
- Segment the content library by intent and business role. Separate informational discovery pages from commercial, local, product, support and decision pages.
- Find replaceable content. Look for pages that mainly restate common knowledge, duplicate another URL or add little evidence, utility or judgement.
- Choose the URL action before rewriting. Retain, improve, consolidate, redirect or retire according to the page’s actual role.
- Add non-commodity value where the role justifies investment. Prioritise original evidence, operating knowledge, useful comparisons, real media and decision criteria over cosmetic AI formatting.
- Measure the three layers again. Recheck AI impressions, normal search acquisition and business outcomes. Do not treat movement in one layer as proof of movement in another.
For international sites, run the same sequence by market rather than assuming one English-language result represents every audience. Query language, terminology, SERP composition, local competitors, trust elements and conversion paths can differ enough to change which page deserves investment.
The Next Check Is the Page’s Reason to Exist
Google’s 2026 guidance makes the priority clearer than the surrounding industry terminology. Keep the technical foundation stable. Use the new Search Console reporting and control as first-party evidence. Then spend editorial effort on pages that offer something a generic answer cannot replace.
If a URL loses some click opportunity but still contributes original information, a valuable next step, a conversion path, a trusted brand relationship or a coherent site structure, it can remain a useful asset. If none of those roles is present, the problem is not that the page needs more “AI optimisation”. The page may simply no longer deserve to exist in its current form.
