There is no separate thing to buy here. Google’s own documentation says it plainly: to be eligible as a supporting link in AI Overviews or AI Mode a page must be indexed and eligible to be shown with a snippet, and there are no additional technical requirements. So the work is the same work — crawlable pages, clear text, honest structured data. What has changed is the economics on the other side of the crawl, and that is worth understanding before you spend anything on an agency promising “AI visibility”.
What answer engine optimisation actually means
Answer engine optimisation — AEO, or generative engine optimisation if you prefer the other acronym — is the practice of making your pages easy for an AI answer to quote and credit. That is it. It is not a parallel discipline with its own tooling, and anyone selling it as one is selling you SEO with a new invoice.
The reason it feels like a new discipline is that the outcome changed. A ranking used to mean a click. Now a page can be read, summarised and credited in a sentence with a link that most people will not press. Google argues this is fine, and says that when people do click from a results page with an AI Overview, those clicks are higher quality, with users more likely to spend more time on the site. That claim comes from Google’s own measurement and is not independently verifiable, so treat it as a position rather than a finding.
What is verifiable is the direction of travel. So the practical question for a business with a modest site is not “how do I win AEO”. It is “which of the things I was already going to do now matter more, and which matter less”.
Google’s own position: no special optimisation required
Google’s AI features documentation lists the best practices it considers worthwhile, and there is nothing exotic on it: allow crawling in robots.txt and at your CDN or host, make content findable through internal links, provide a good page experience, keep important content in textual form, support text with good images and video, make sure structured data matches the visible text, and keep Merchant Center and Business Profile information current.
Read that list again and notice what is absent. No word count target. No question-and-answer formatting rule. No schema type that unlocks citations. No llms.txt. The one sentence worth pinning above your desk is the requirement that structured data matches the visible text on the page, because that is the rule most AEO advice quietly breaks.
Google Search Central, “AI Features and Your Website”There are no additional technical requirements.
If your site is slow, thin or badly linked internally, fix that. It is the same fix as last year, and we set out the basics in SEO fundamentals every business should know. The uncomfortable part is that there is no shortcut available instead.
The crawl-to-referral gap, in numbers
Here is the part the marketing blogs skip. Cloudflare publishes crawl-to-refer ratios on Radar — how many pages a platform crawls for every visitor it sends back — and the numbers for July 2025 are stark. Anthropic crawled 38,066 pages for every referred visit. OpenAI was at 1,091. Perplexity sat at 195 and had moved the wrong way, up 257% since January. Microsoft was steady around 41. Google, which still sends real traffic, was at 5.4.
| Platform | Jan 2025 | Jul 2025 | Change, Jan to Jul |
|---|---|---|---|
| Anthropic | 286,930 : 1 | 38,066 : 1 | −86.7% |
| OpenAI | 1,217 : 1 | 1,091 : 1 | −10.4% |
| Perplexity | 54.6 : 1 | 194.8 : 1 | +256.7% |
| Microsoft | 38.5 : 1 | 40.7 : 1 | +5.7% |
| Yandex | 15.5 : 1 | 21.4 : 1 | +38.3% |
| 3.8 : 1 | 5.4 : 1 | +43% | |
| ByteDance | 18 : 1 | 0.9 : 1 | −95% |
A falling ratio is good news for publishers, and Anthropic’s fell sharply after Claude gained web search with clickable citations during 2025. But even after an 87% drop it remained the most crawl-heavy platform by a wide margin. The honest summary: these systems consume a great deal and return a little, and the one that returns most is still the search engine you were already optimising for.
That chart is the single most useful thing on this page. Four in five AI crawls are feeding a training set, not answering a live question. Training crawls will never send you a visitor. So when you budget for AEO, you are mostly buying a share of the shrinking 17% that is search-shaped — which is another way of saying you are buying SEO.
Query fan-out, and why one page cannot cover everything
Google describes a “query fan-out” technique behind both AI Overviews and AI Mode: the system issues multiple related searches across subtopics and data sources, and while the response is being generated its models identify more supporting pages. The stated effect is a wider and more diverse set of links than a classic results page.
The practical consequence is counter-intuitive. One enormous page covering a whole topic is now a weaker bet than several focused pages, because fan-out is looking for specific subtopic answers from potentially different sources. A page that answers one question properly and links to its siblings has more surfaces to be found on than a 6,000-word pillar nobody can summarise.
That is not a licence to publish thin pages. It is an argument for topical clusters with real internal links — which is exactly what we argue for on content grounds in content marketing that actually works. The incentive has simply got stronger.
- One clear answer per page, stated in the first paragraph rather than the conclusion.
- Subtopics split into their own URLs with descriptive headings, not anchors on a single monster page.
- Internal links that use the words a person would search, not “read more”.
- Figures with a named source and a date in the sentence, so a quote survives being lifted out of context.
- Text that stands alone without the image or the chart beside it.
Structure a machine can lift cleanly
Structured data is where there is genuine, cheap advantage left, because most of the web has not bothered. The 2025 Web Almanac, built from the HTTP Archive crawl, found JSON-LD on 43% of home pages on both desktop and mobile, up from 41% desktop and 40% mobile in 2024. Microdata is slipping, at 17% on desktop and 16% on mobile. Inner pages do worse than home pages: 39% desktop and 37% mobile for JSON-LD.
The same chapter found that an H1 appears on 70% of pages but only 66% are non-empty. A four-point gap between “has a heading” and “has a heading with words in it” is a fair summary of how much of the web’s markup is decorative.
Pick the schema types that describe what you really are. For most service businesses that is Organization, LocalBusiness or Service, plus Article on blog posts, FAQPage only where you genuinely have a question list on the page, and BreadcrumbList. Google’s structured data gallery lists the types that produce rich results, which is the right place to check before you add anything. Then obey the one rule: the markup must match the visible text.
Measuring it: the generative AI reports in Search Console
Until mid-2026 there was no honest way to measure any of this. Google’s Search Console blog announced dedicated Search Generative AI performance reports in June 2026, giving separate views of impressions within generative AI features on Search — AI Overviews and AI Mode — as well as generative AI features in Discover. Google began the rollout with a subset of site owners while it gathered feedback, so your property may not have it yet.
Read it for direction, not precision. An impression means a link to your site was shown inside a generative feature. It does not mean anyone read your name, and it certainly does not mean a visit. Pair it with the ordinary Performance report: if AI impressions climb while clicks stay flat, you are being used rather than visited, and that is information worth having before you commission more content.
Resist third-party “AI visibility” trackers that prompt a chatbot repeatedly and count your mentions. The answers are non-deterministic and personalised, the sample is tiny, and the vendor controls both the questions and the scoring. Search Console is at least measuring the real surface.
Controlling what gets used, and what that costs you
You do have controls, and they have costs. Google is explicit that because AI is built into Search, robots.txt directives for Googlebot are the control for how your site is crawled for Search — there is no separate AI Overviews crawler to block. To limit what is shown from your pages you use nosnippet, data-nosnippet, max-snippet or noindex. Google-Extended is the control for AI training and grounding in some of Google’s other systems.
Search Console also exposes a Search generative AI control, which lets you decide whether your site’s links and content can appear in Search generative AI features and receive impressions and traffic from them. That is a real choice, not a theoretical one.
Blocking Googlebot removes you from Search, not just from AI Overviews. Using nosnippet suppresses your snippet everywhere, including ordinary results, which usually costs more clicks than the AI feature was taking. For almost every small business the right answer is to stay fully crawlable and compete on substance. Opting out is a decision for publishers whose entire product is the text itself.
A six-week plan that does not wreck your existing SEO
Six weeks, a few hours a week, no new tools. Week one: verify Search Console, check whether the generative AI reports are available on your property, and record a baseline of impressions and clicks. Week two: fix crawlability — robots.txt, the XML sitemap, and any CDN rule or bot-protection setting that is quietly blocking Googlebot. Week three: add or correct JSON-LD for your organisation and your articles, and make sure every claim in the markup appears on the page.
Week four: rewrite the opening paragraph of your ten most important pages so the answer comes first. Week five: split one oversized page into three focused ones and link them properly. Week six: go back to Search Console, compare against your baseline, and write down what moved. Then stop and let it settle for a quarter.
If you are also running paid campaigns, do not shift budget on the strength of six weeks of AI impression data. The comparison between paid and organic is slower and messier than that, and we work through it in SEO vs paid ads: where should you invest?
Frequently asked questions
Do I need an llms.txt file?
No major search or AI provider has committed to reading one, and Google’s AI features documentation does not mention it. It costs nothing to publish, so treat it as a lottery ticket rather than a task. Your robots.txt, sitemap and internal linking are the files that demonstrably affect what gets crawled and cited, and they are where the time should go.
Should I block AI crawlers to protect my content?
Rarely, if your business is selling services rather than selling the text itself. Blocking Googlebot removes you from Search entirely, and Google treats robots.txt for Googlebot as the single control for Search crawling. Google-Extended limits training and grounding in some other Google systems without touching Search. Weigh the lost visibility honestly before opting out of anything.
Will adding an FAQ section get me cited more often?
Only if the questions are real. Google’s guidance is that structured data must match the visible text, and an invented question list is both a markup violation and a bad page. A short FAQ answering what clients actually ask is good practice for human readers first; any citation benefit is a side effect, not the reason to do it.
How long before AI impressions show up in Search Console?
Google rolled the generative AI reports out to a subset of site owners from June 2026, so availability varies by property. Once the report appears, it draws on the same pipeline as the rest of Search Console, which typically lags by a couple of days. Content changes take longer: recrawling can take days or months depending on how often Google refreshes your pages.
Is AEO worth a separate budget line?
Not as a separate retainer. Fold it into the search and content work you already fund, and judge it on the same measures: impressions, clicks, enquiries. If a proposal promises guaranteed AI citations or a visibility score, ask which measurable surface it comes from. The honest answer is Search Console, and you can read that yourself.
We work on the unglamorous half of this remotely: crawlability, structured data, internal linking and a content plan that answers one question per page. Our digital marketing services cover search and AI visibility together, because Google treats them as the same job. Get in touch with Eudora Technology to talk about your project.
.png){kind=link}
.jpg){kind=link}