Why this signal matters for Google
An AI Overview is a summary written by an AI, placed at the very top of Google's results. It answers the question before the visitor ever scrolls down to the usual links. Next to the text, Google shows a few pages as sources, each with a link. On many queries, those links are the only doorway left. Take someone searching « how long to keep tax records ». The summary gives the number right away. The reader only clicks to double-check, and they click the cited pages. So your target moves: you are no longer just chasing position one, you are chasing the citation. And an AI only reuses what it can verify elsewhere, which means pages that show where their numbers come from and cite their sources.
There is no button that puts you inside an AI summary. Google draws from its normal index: the pages already ranking on the query are the ones it has at hand. So the work is the same as for classic search, with a higher bar for proof. The quality criteria are your compass. Human evaluators rate pages using a public manual, the Quality Rater Guidelines (QRG for short). Section 3.2 tells them to research what others say about you, away from your own site. Section 2.5.3 tells them to check who is responsible for the page. A page with no identifiable author and no outside reputation fails both checks. It is also the least attractive page for a summary to cite.
What an AI reuses is a short, complete answer placed where it can find it. If your page takes nine hundred words to get to the point, the usable passage does not exist. The right shape is three lines, right under the heading: the direct answer first, the detail after. On a plumber's page, that reads « report water damage to your insurer within five business days », and only then the exceptions and the steps. On topics about money, health or safety, which Google calls YMYL (Your Money or Your Life), the bar rises again: section 3.4.1 expects real trustworthiness, and Google is more careful about what it summarizes there. An answer with no qualified author and no source will not be picked up.
The same work pays off in ChatGPT, Claude and Perplexity. These tools read the web with their own crawlers and list their sources at the end of each answer. They look for the same things Google does: a clear page, dated, signed, that says where the information comes from. Two technical details matter here. First, blocking the Google-Extended crawler does not remove you from AI Overviews: that crawler serves other purposes, and it is your normal indexing rules, plus the nosnippet tag, that decide what can be reused. Second, a page untouched for four years loses to an equivalent page updated this year, because an AI summary wants the answer that is true today, not the one from 2022.
The NeuroEEAT scoring grid (0-10)
Rate each important page on whether an AI summary could reuse it as is, without the machine having to guess.
| Score | Description | Concrete indicators |
|---|---|---|
| 0-2 | Nothing to reuse | No page answers a question directly. No author, no date, no sources. The text opens with a long introduction before getting to the point. |
| 3-4 | Visible but not quotable | You rank on a few queries, but the answer is buried mid-page. Sources are not given and dates are missing or wrong. |
| 5-6 | Quotable in places | Some pages answer up front, with a named author and an honest update date. Sources are mentioned, but without a link to check them. |
| 7-8 | Picked regularly | Every page targets one precise question and answers it in the first three lines. Identifiable author, linked sources, accurate structured data, fast-loading page. |
| 9-10 | The reference source | Your answers get cited on your key queries. You publish original material nobody else has: your numbers, your tests, your method. And your reputation checks out away from your site. |
How to audit it on your site
Five checks to find out whether an AI summary can help itself to your pages. Start with the ten pages that bring you the most visits.
- 01
Look at what Google already shows
Open a private browsing window. Type the ten questions your customers actually ask. For each one, note whether an AI summary appears and which pages it cites. If the same three competitors keep showing up, read their pages: you will see the exact shape Google likes to reuse.
- 02
Count the words before the answer
On each page, count how many words sit between the heading and the actual answer. Past one hundred, the usable passage is too far down. Move the answer right under the heading and keep the context for later.
- 03
Check that robots can read the page
Three points: your robots.txt does not block the page, no nosnippet or max-snippet tag limits the extract, and the text is present in the source code. If the answer only appears after a script loads, do the « view source » test and search for your sentence. If it is not there, a crawler may never see it.
- 04
Check the proof on the page
Named author with visible credentials, a real update date, sources linked to the original, numbers attributed to someone. An unsourced claim is exactly what an AI summary leaves out, because it cannot cross-check it anywhere.
- 05
Score and prioritize
Give each page a mark out of ten. Fix the high-traffic, low-score pages first: that is where the shift to AI summaries costs you the most. Redo the round every three months, because summaries change their sources often.
How to activate or improve it
Three levels, from a page rewritten by hand to content nobody else can produce.
Make your pages reusable
- Put the answer to the question in the first three lines, as one sentence that stands on its own.
- Use the question itself as the heading, in the words your customers use.
- Sign the page with a name, a role and a genuine update date.
- Link every number you quote to its source.
Structure it for the machine
- Break the page into short questions, one per subheading, each followed by its direct answer.
- Add the structured data that genuinely matches the page (Article, FAQPage, Organization) and nothing more.
- Make sure the useful text is in the source code, not only rendered by a script.
- Rework your old pages one by one: answer moved up, numbers rechecked, date corrected.
- Track who gets cited instead of you on your key queries, and note what that page does better.
Become the source that gets cited
- Publish data only you have: your measurements, your survey, your test, with the method explained.
- Exist somewhere other than your own site, so an AI finds several sources telling the same story.
- Link your pages together around one topic, so the answer and its proof sit one click apart.
- Keep a per-page review calendar, with an owner and a date, instead of a full redesign every three years.
- Watch ChatGPT, Claude and Perplexity on your key questions too: their cited sources are not always the same as Google's.
5 common mistakes to avoid
Sector specifics
The highest bar
On these topics, section 3.4.1 expects real trustworthiness and Google is more cautious before reusing an answer. You need a qualified, named author with their role shown, and every claim tied to an official source. A generic answer about a treatment or an investment will not be picked, however well written.
Practical questions first
AI summaries love concrete questions: deadlines, prices, paperwork, what to bring. A page that answers « who files the claim for water damage » in one sentence has a better shot at being cited than a brochure page. Keep your name, address and phone identical everywhere, so the machine ties your pages to the right business.
Compare instead of praising
On a product query, the summary compares criteria: size, material, price, warranty. A product page that recycles the manufacturer's pitch adds nothing. A page that gives real measurements, the product's limits and the result of an in-house test gives the AI something to cite.
Know-how put in writing
Your queries are rare but highly targeted. That is where original content pays: a method explained step by step, a document template, a project outcome with numbers. Those are answers your competitors do not have and an AI can only find at your place.
Signals to activate in parallel
An AI summary never picks a page for a single reason. These signals decide whether yours is reusable:
FAQ
How to appear in AI Overviews when starting from zero?
First get ranked on the query: Google picks its sources from pages it already knows. Then move the answer to the top of the page, as one self-contained sentence, and sign and date the page. Those three moves cost an hour per page and change everything, because they hand the machine a passage it can reuse without rewriting it.
Which queries trigger an AI Overview?
Mostly questions: how, why, how much, what to do. Transactional and navigational queries trigger them less. A summary can also appear one day and vanish the next, because Google keeps adjusting. The practical move: check your own ten key queries in a private window instead of trusting a market average.
Is there special markup that gets you into an AI summary?
No. No tag, no code and no registration opens that door. Structured data still helps Google understand your page, but it does not replace a clear answer, a source and an author. Be wary of anything sold as a technical shortcut.
Should you block AI crawlers to protect your content?
It is a trade-off, not an obvious call. Blocking AI tools' crawlers removes you from their answers, and so from their citations. On Google's side, blocking Google-Extended does not pull you out of AI Overviews: the nosnippet or max-snippet tag is what limits what can be reused, at the price of a shorter snippet in classic results.
Do AI Overviews cost you traffic?
On simple questions, yes: the reader gets the answer and stops clicking. On questions that lead to a decision, the click survives, because people want to verify with the cited source. That is why the battle moves: better to be the cited page on a hundred queries than the tenth result on a thousand.
Anonymized case study
Before · a page the machine cannot cite
After · an answer that can be reused as is
Illustrative figures meant to show the typical progression of this signal. They do not come from a real client case and must be replaced with your own measurements before any external use.