Guide 11 · Evidence review
What Google’s AI optimization guide confirms, and what our data adds
Google now has an official page on optimizing for AI Overviews and AI Mode, last updated 10 July 2026. Its main message is that this is still SEO. It lists five things you do not need. Two of them, llms.txt and schema, match what the evidence reviews on this site had already found, and our citation data agrees with the other three. What it cannot tell you is how the other engines behave, and in our data they read very different sources.
What does Google say you do not need?
Google’s guide to optimizing for generative AI features lists five things in plain terms. They are worth reading in Google’s own words, because each one is still sold as a GEO service:
| Google says | What we measured | Verdict |
|---|---|---|
| No new machine-readable files, AI text files or Markdown. Google Search “ignores” llms.txt. | Ahrefs found 97% of 137,000 published llms.txt files had never been fetched by any crawler. Our retrievability check scores the file at zero. | Agrees |
| “Structured data isn’t required for generative AI search.” No special schema.org markup. | Ahrefs tracked 1,885 pages that added schema against 4,000 that did not: no major lift on any engine (AI Overviews −4.6%, AI Mode +2.4%, ChatGPT +2.2%). | Agrees |
| “There’s no requirement to break your content into tiny pieces.” | The most-cited vendor in our leaderboard run publishes long comparison articles, not fragments. | Agrees |
| “You don’t need to write in a specific way just for generative AI search.” | A one-character change to a query changed a fifth of Perplexity’s sources. You cannot write for one exact phrasing. | Agrees |
| Seeking inauthentic mentions “isn’t as helpful as it might seem.” | Mentions correlate with visibility (0.66), but the vendor that shows up most earned it with its own content, not placements. | Agrees, with a nuance |
Each row has its own page with the full evidence: llms.txt, schema, chunking, writing for AI and manufactured mentions.
What does Google say you do need?
Less than most checklists. Google’s AI features page says a page must be “indexed and eligible to be shown in Google Search with a snippet” to appear as a supporting link, and that there are “no additional technical requirements.”
That makes three technical things decisive: the page is crawlable, it is not marked noindex, and it is not marked nosnippet or max-snippet:0. A stray directive from a staging site or an SEO plugin setting removes a page from AI Overviews while every crawler still gets a 200. Since 23 September the retrievability check reads both directives on your homepage, in the meta tags and the X-Robots-Tag header.
On content, the guide asks for pages that are not “commodity content”. Its example of commodity content is a generic tips list. Its example of the opposite is a first-hand review. That is the one piece of the guide you cannot fix with a setting.
On JavaScript it says Google “is able to process content within JavaScript as long as it isn’t blocked.” That is true of Googlebot. It is not true of the retrieval crawlers used by ChatGPT, Perplexity and Claude, which is why the JavaScript guide still matters.
Where does our data go further than the guide?
Google’s guide covers Google. The buyer asking an AI which plugin to install is also asking ChatGPT and Perplexity, and across three engines we found three largely different reading lists:
| Measurement | Result | Source |
|---|---|---|
| Source overlap between engines, median | Perplexity–Google 15.5%, Google–ChatGPT 12%, Perplexity–ChatGPT 8% | Run 05 |
| Domains cited by all three engines | 17 of 250 (7%) | Run 05 |
| youtube.com as a source | Google AI Mode 13 of 15 answers; the other two engines 2 of 30 | Leaderboard |
| Sources ChatGPT retrieved vs showed | 217 retrieved, 36 displayed (17%) | Run 05 |
| Same query string, asked again | 100% identical Perplexity source list | Run 04 |
Two practical consequences. First, doing everything in Google’s guide gets you eligible for Google’s answers, and says nothing about whether ChatGPT retrieves you. Second, the YouTube gap is Google-specific. If your category’s Google answers lean on video and you have none, that is a concrete gap, and the guide’s mention of “high-quality images and video” points the same way.
Google says AI Overviews and AI Mode may use “query fan-out”, running several related searches to build one answer. That fits what we see: the pages cited are often about adjacent questions, which is how one vendor publishing comparison articles next to its own category ended up a source in 18 of 45 answers.
How do you measure it?
Search Console now has a Generative AI performance report, which the guide names as the way to measure it. AI Overview and AI Mode traffic is also still counted inside the ordinary Web performance report. For this site the new report showed its first two impressions around 14 September 2026, on the homepage and the results page. That is small, and it is the only first-party number any engine gives you.
For ChatGPT and Perplexity there is no equivalent report. You have to run real prompts and read the sources, and re-running the same prompt text returns a cached answer, so each re-run needs a new phrasing.
What this page cannot tell you
Google’s guide is a statement of how Google says its systems work. It is not a measurement, and it can change without notice. Our own data comes from WordPress plugin categories only, from runs in September 2026 on three engines. Neither source tells you what any other engine does.
What would change this page: Google revising the guide, which I will note here with the date, or a future run where the engines’ reading lists converge. So far the overlap has gone down as engines are added, not up.
How this was made: figures come from our own runs in the engines’ public interfaces or from the studies named above. Drafted with an AI assistant (Claude) and reviewed by Alauddin Chowdhury. Method and AI use.
Want to know which of this applies to your plugin?
Send me your wordpress.org slug. I will check the indexing and snippet directives, what each retrieval crawler gets from your site, and which domains the engines actually read for your category. Free, no call, usually within a day.