field note
first published 2026-10-09
How to rank in Google AI Overviews: what the engine actually needs from your page
AI Overviews cite pages Google can fetch, parse and trust. What an AI Overview is, how the citations are chosen, the five things that decide whether your page is eligible, and how to measure any of it when Search Console won't separate the clicks.
1,017 words · 5 min read · ai search
Every week someone asks me how to "rank in AI Overviews" as if there were a second index with its own rules. There isn't. Google's AI Overview is written from pages Google has already crawled, indexed and judged relevant to the query. If you are not eligible for the normal results, you are not eligible for the summary above them. If you are, the question becomes which of the eligible pages the engine can actually use.
That second question is answerable, and most of the answer is technical.
What an AI Overview is, mechanically
For some queries Google generates a short written answer and places it above the organic results, with a set of citations beside or beneath it. Those citations are links to ordinary web pages. Google has said the Overview is grounded in its search results, and everything I have measured agrees: the cited pages are almost always pages that rank for the query or for a near-variant of it, and they are pages from which a model can lift a clean, self-contained passage.
So there are two filters. The first is the one you already know: rank. The second is new in degree if not in kind: extractability. A page passes the second filter when a machine can find the answer in the HTML, understand what it refers to, and quote it without dragging in navigation, boilerplate or a contradicting sentence two paragraphs down.
The five things that decide eligibility
1. Google can fetch the whole page, from the origin, as Googlebot. Firewalls and bot-management products are the quiet killer here. In the last year I have watched two different sites lose crawler access to most of their catalogue within seconds of a security rule change, one of them 1,647 of 1,752 pages. Check the origin's response to Googlebot's user agent directly, not through the cache, and read the firewall's own event log.
2. The answer is in the raw HTML. Fetch the page with JavaScript disabled. If the paragraph that answers the question only exists after a script runs, Google will render it eventually, but the passage ranking that feeds the Overview is tuned for content it can read on the first pass. Server-render the body copy. This is the single most common failure I find on modern marketing sites.
3. The answer is a passage, not a page. Overviews are built from passages: a heading that states the question in the user's words, followed by a direct answer in the first sentence or two, followed by the detail. Pages that open with a four-paragraph preamble before getting to the point are harder to quote. Put the answer first; earn the right to elaborate after.
4. The page agrees with itself and with the entity behind it. Google cross-checks. A page that says one thing in the body, another in the structured data and a third in the title gives the model a reason to prefer a source that doesn't. Structured data as one connected graph, with the Organization or Person behind the page named and linked, makes the page easier to trust. It is not a ranking trick; it is the removal of a reason to be skipped.
5. Snippet directives are not fighting you. nosnippet, max-snippet and data-nosnippet do exactly what they say, and they also limit what Google may draw into an Overview. I have found them left over from a privacy review on pages that were supposed to be the site's best answers. The same check covers the newer Search Console controls for AI features: read them, decide, and write the decision down.
What does not help as much as the industry says
Writing "for AI" in some special register does nothing. The model is reading the same HTML, and the thing it rewards is clarity, which is also what readers reward. Keyword density, FAQ blocks stuffed with ten near-identical questions, and schema markup for things that are not on the page are at best ignored and at worst a reason for distrust.
Tracking a list of prompts in a dashboard also helps less than its price suggests. The Overview you see for a query today is not the Overview the next searcher sees, and the prompt you track is not the one anyone types twice. Instrument what you control instead.
How to measure it
Search Console reports AI Overview impressions and clicks inside the normal Web data, without a separate filter. That is frustrating and it is the situation. Three things work:
- Keep your own list of the queries that matter and note, weekly, which ones show an Overview and whether you are cited. It is manual and it is the only ground truth you will have.
- Watch position and click-through together for those queries. A page that holds position while its click-through falls is usually a page that has been summarised above the fold.
- Read your server logs. Google's referrer tells you a click came from Google; it does not tell you it came from the Overview, but a step-change in clicks to a page whose position has not moved usually does.
The deeper fix is to run your own instruments for the inputs: crawler access by bot, extractability of your key pages, structured-data validity, and the three gates the Agent-Ready Grader on this site checks for free. Those are the things you can change on a Tuesday and verify on a Wednesday.
A short worked example
A UK dental practice came to me with a site that looked fine in a browser and was invisible to the machines: three H1 headings on the homepage, structured data in unrelated islands, body copy arriving late. On this site's grader it scored 74, a C. We fixed the headings, rebuilt the graph, put the answers in the HTML, published an llms.txt, and the same instrument read 99, an A. Nothing about the practice changed. Everything about its legibility did.
That is the job for AI Overviews, stated plainly: be eligible, be extractable, be consistent, and measure the inputs yourself, because nobody else is going to.
asked straight — answered straight
01What are Google AI Overviews?
The AI-written summary Google shows above the normal results for some queries, with links to the pages it drew on. The links are the part you can influence: they go to pages Google already ranks and can extract a clear answer from.
02Do you need to rank on page one to appear in an AI Overview?
Usually the cited pages already rank well for the query or a close variant, but not always in the top three. Eligibility rests on being indexed, extractable and clearly on-topic. A page that ranks sixth with a crisp, quotable answer is often cited over a page that ranks second and buries it.
03Can you opt out of AI Overviews without leaving Google?
You can limit how much of a page Google may use with the nosnippet and max-snippet directives, and Search Console has gained controls for AI features. Each of those also limits normal snippets, so treat it as a trade, not a free switch.
04How do I measure traffic from AI Overviews?
Search Console folds AI Overview impressions and clicks into the normal Web performance data, so there is no clean split. Watch position and click-through together for the queries where an Overview appears, keep your own log of which queries trigger one, and read your server logs for the referrers Google sends.
next step — ai search optimisation
AI search optimisation as engineering.
Make your site legible to ChatGPT, Perplexity and AI Overviews — then measure it. Send a brief — a few lines about the site, the problem and the evidence you have — and the reply is a straight read on fit and shape, from the person who does the work.
no forms · no funnels · jamie@jamiemckaye.com
written by
Technical SEO consultant and full-stack developer in Hersham, Surrey, in practice since 2007. One person, no handoffs: the audits, the code and the writing come from the same pair of hands. This site is the working proof — it grades itself on the same instrument it runs for clients.