How to rank in AI Overviews: what is required and what is not

Six of the first results for this query, read on 5 September 2026, promise a procedure: twelve tips, eight steps, seven tactics, six ranking factors, eight pillars, eight more steps. One of the six quotes the Google page saying the procedure does not exist. The query carries 1,000 monthly US searches at a cost per click near 5 USD, and five dollars a click is what a term costs when the person typing it is about to pay somebody.

How do you rank in AI Overviews?

AI Overviews are an extraction rather than a ranked list, so there is no position to win. Google documents no additional requirements beyond being indexed and eligible to be shown with a snippet. A new site waits on indexing, a mid-authority site gains most from rewriting passages, and a consolidated site defends citations it already holds.

Quick answer

• There is one documented requirement, and it is the same one that governs ordinary search results: indexed, and eligible to be shown with a snippet.
• The controls Google publishes only take you out. nosnippet, data-nosnippet, max-snippet and noindex are named as the way to limit what is shown.
• One family of edits has a measurement behind it, from the paper that named generative engine optimization, and one of the nine methods it tested performed worse than doing nothing.
• How often a summary even appears is disputed, from 25.11 % of queries to 60.32 % depending on which study you open, so any plan built on one of those numbers is built on a choice.

What kind of question this is

The question assumes a list with a door in it.

That assumption is reasonable, because for twenty years the answer to "how do I appear in X" was a ranking problem: there was an ordered set of results, a position in it, and work that moved you up. A summary is not that shape. It is composed at answer time out of passages, and the sources underneath it are supporting links rather than places on a ladder. Nothing gets promoted into it, so nothing can be optimized toward a rank that does not exist.

The consequence is uncomfortable and it is the reason the guides avoid it. If there is no position, there is no procedure to sell for entering one, and what remains is editing prose, which nobody can package as a deliverable with a number next to it. The market fills that gap with a checklist, and the checklist is where the unsourced instructions live.

The one requirement, in Google's own words

Google's documentation on AI features in Search, opened on 5 September 2026 and last updated on 10 December 2025, is short enough to read in full and specific enough to settle the argument. It states that "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary". On eligibility it is equally plain: "to be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements."

Indexing is therefore not a step in this work, it is the gate in front of it, and a page missing from the index is not a candidate for a summary, a snippet or anything else. When that is the situation, the diagnosis order is a separate job and it is in why a page is not indexed.

What the guides add, sorted by what backs it

Sorting the advice by provenance is more useful than sorting it by effort, because the three tiers behave differently when they fail.

Tier one is contradicted by the source. The same Google page says "You don't need to create new machine readable files, AI text files, or markup to appear in these features" and "There's also no special schema.org structured data that you need to add". One of the six results ranks an AI text file as its second tip, ahead of anything about writing. What that file is actually for, and who proposed it for what, is a separate question with a separate answer in what llms.txt is.

Tier two has a measurement. The paper that named the discipline, by Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, tested nine optimization methods across roughly 10,000 queries. Adding quotations, adding statistics and citing sources were the methods that moved visibility inside generated answers; statistics on their own produced a 30 to 40 % relative improvement on the paper's Position-Adjusted Word Count metric, and its best-performing methods reached 41 % on that same metric. Keyword stuffing, the tactic imported from search habit, came in below baseline. The full reading of what the paper measured and what it did not is in what generative engine optimization is.

Tier three is convention. Conversational tone, topical authority, freshness and the rest are plausible, widely repeated and unaccompanied by anything anyone can open. They may well help. The distinction worth keeping is that a convention cannot be defended to whoever is paying, and a plan that mixes all three tiers in one bulleted list hands the client no way to tell them apart. Where the boundary sits between this work and ordinary ranking work is set out in what answer engine optimization is.

The only lever Google publishes, and it removes

Every documented control in this area points the same way, which is out.

The AI features page names nosnippet, data-nosnippet, max-snippet and noindex as the controls "to limit the information shown from your pages in Search". There is no matching control that puts a page in. That asymmetry is the structural fact of the whole subject: eligibility is defined by exclusion, so the only setting with a documented effect on your appearances is one that can only reduce them.

Which makes the cheapest useful check the one almost nobody runs. A short max-snippet inherited from a theme, a plugin default or a staging configuration keeps a page ranking normally and quietly withdraws it from summaries, and the two symptoms never get connected because one of them is invisible. In the portfolio I operate the check that has changed an outcome is not a content audit, it is reading the served HTML of a template for a robots directive nobody remembers adding. It costs minutes and it is the only part of this article where the effect is certain.

What a passage has to be before anything quotes it

The unit that competes is not the page. It is the block that survives being cut out of it.

The mechanical test is to copy any fifty words from your page, paste them where there is no context, and see whether anything remains. Most business writing fails immediately, because its meaning lives in the buildup rather than in the sentence. What follows is the same claim written twice, and the difference is not tone.

Before: AI Overviews have dramatically reduced clicks for publishers, and the trend is accelerating. After: Ahrefs, comparing December 2023 with December 2025 across 300,000 keywords, reports that the presence of an AI Overview correlates with a 58 % lower average clickthrough rate for the top-ranking page.

Two things changed. The claim now carries who measured it and over what, so it can be checked by somebody who disagrees, and the verb stopped asserting cause where the study reports an association. The second change is the one that gets skipped, and it is the difference between a sentence a careful writer can borrow and one they have to rewrite before using. The same discipline applied to a model that names sources rather than extracting them is worked through in how to get cited by ChatGPT.

Where the hours pay, and where they do not

The decision is not whether to do this. It is which ten hours.

The middle row is where the paper's own numbers point. With its Cite Sources method, sites ranked around fifth gained 115.1 % in visibility inside generated answers while the top-ranked site lost 30.3 % on average, so the measured upside sits with pages that are competitive without being dominant. That is the opposite of where most teams spend, and it is also why a first-place page has something to defend here rather than something to win.

There is a second reason not to send every hour into the summary, and it comes from the click data. Seer Interactive, across 3,119 informational queries and 42 organizations between June 2024 and September 2025, measured organic clickthrough falling from 1.76 % to 0.61 % where an AI Overview appears, which Seer reports as a 61 % decline. Where no AI Overview appears it also fell, from 2.72 % to 1.62 %, a 41 % decline. Clickthrough is dropping in both conditions, so only the gap between the two declines belongs to the summary, and the queries in your own account that never trigger one are a real place to spend. Separating those two piles before writing anything is what free keyword research is for.

When ranking in AI Overviews is not the goal

When the page is not indexed. Everything above assumes a candidate, and there is no shortcut past that.

When the demand is transactional. Summaries concentrate on informational and comparative queries, and a page selling a product competes for a different surface entirely.

When the business lives on an address. Local demand barely touches the queries that produce summaries, and how they behave in that context specifically is in AI Overviews and local search.

When the whole answer fits in forty words. Winning there hands the reader everything and keeps nothing, which is a donation rather than a win.

How you check that it happened

This is where most plans quietly stop, and the reason is that the honest answer has two halves and neither is a dashboard.

Google's Search Console documentation, opened on 5 September 2026, describes how the counting works: "clicking a link to an external page in the AI Overview counts as a click", standard impression rules apply, and "an AI Overview occupies a single position in search results, and all links in the AI Overview are assigned that same position". AI Mode follows the same methodology as a results page. What the page does not give is any way to separate those rows from the rest, so the evidence arrives mixed into ordinary search data. Getting the property that holds it is an afternoon of work, described in how to set up Search Console.

The second half is worse and it is the part experience teaches. A screenshot of your page inside a summary proves almost nothing, because the same query can return a different composition tomorrow, or no summary at all. Even the population question, how often a summary appears at all, has four incompatible published answers: 25.11 % of queries in Conductor's count of 21.9 million queries in the first quarter of 2026, 43 % in Similarweb's Generative AI Landscape 2026, 48 % in BrightEdge's commercial verticals in March 2026 and 60.32 % in Xponent21's US sample in April

1. Those four differ because each study chose its own keyword sample, which is a separate problem from one query changing its mind overnight. Together they are the reason a single look is not a measurement.

What survives that is a fixed cohort. Ten to twenty of your own queries, checked on a schedule, with a written record of whether a summary appeared and who was cited, is slow and imprecise and still the only instrument that matches the thing being measured. When the motive behind all of this is traffic that already fell, the diagnosis belongs before the tactic, and the order for it is in why organic traffic drops.

Mistakes that repeat

• Buying markup as a requirement. Google documents that no special schema.org data is needed for these features, so the purchase is for something else.
• Never reading the served HTML. A short max-snippet keeps a page ranking and removes it from summaries, and nothing in any report connects the two.
• Quoting a clickthrough figure without its date. The same firm reported 34.5 % in April 2025 and 58 % in February 2026, over different windows.
• Spending the budget on the page already ranking first. That is where the paper's Cite Sources method measured a loss rather than a gain.
• Treating one screenshot as proof. The same query can return a different summary, or none, the following day.
• Counting sessions as the result. With 93 % of AI Mode queries ending without an external click, a session report undercounts exactly what was won.

Data and transparency

That there are no additional requirements to appear in AI Overviews or AI Mode and no special optimizations necessary, that eligibility is being indexed and eligible to be shown with a snippet while fulfilling the Search technical requirements, that no new machine readable files, AI text files or markup are needed and no special schema.org structured data has to be added, and that nosnippet, data-nosnippet, max-snippet and noindex are the controls that limit what is shown, all come from Google's documentation on AI features in Search, opened on 5 September 2026 and carrying a last-updated date of 10 December 2025. The click, impression and position rules quoted for AI Overviews and AI Mode come from Google's Search Console performance report reference, opened the same day; that page describes how each one counts and gives no method for separating them, which is the basis for calling the data mixed rather than absent.

The clickthrough figure of 58 % comes from Ahrefs, published 4 February 2026, over 300,000 keywords split evenly between queries with and without an AI Overview, comparing December 2023 with December 2025, on aggregated Search Console data, using average desktop clickthrough per month calculated as the sum of clicks divided by the sum of impressions. Its own wording is that the presence of an AI Overview correlates with a 58 % lower average clickthrough rate for the top-ranking page, so it is an association between summary presence and average clickthrough and not a loss followed on a fixed set of pages. The same update supersedes the 34.5 % that the firm published in April 2025 for a March 2024 against March 2025 window; the two windows are different widths, so the pair is not evidence that the effect doubled.

The paired decline figures come from Seer Interactive's study published 4 November 2025, covering 3,119 informational queries across 42 organizations and 25.1 million organic impressions from June 2024 to September 2025: 1.76 % to 0.61 % with an AI Overview, reported by Seer as a 61 % decline, and 2.72 % to 1.62 % without one, a 41 % decline, so only the gap between them is attributable to the summary. The visibility figures come from the GEO paper by Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, arXiv 2311.09735, presented at KDD 2024: nine methods over roughly 10,000 queries, a 30 to 40 % relative improvement on Position-Adjusted Word Count from adding statistics, 41 % on that metric for the best-performing methods rather than for statistics alone, a 115.1 % gain for fifth-ranked sites and a 30.3 % average drop for the top result under the Cite Sources method, and keyword stuffing below baseline. The 93 % of AI Mode queries ending without an external click has two independent measurements behind it: Semrush, over 69 million US desktop sessions between 1 May and 5 July 2025, and Seer Interactive, over 25.1 million impressions.

The four presence figures are separate measurements that are not comparable and are never averaged here: Conductor 25.11 % (21.9 million queries, first quarter of 2026), Similarweb 43 % (Generative AI Landscape 2026), BrightEdge 48 % (commercial verticals, March 2026) and Xponent21 60.32 % (United States, April 2026); those four are named without links because their source pages were not opened for this article. That six of the first results promise a procedure and that one of them quotes Google's eligibility line is my own reading of that results page on 5 September 2026. The search demand for the query, 1,000 monthly searches in the United States with a keyword difficulty of 39 and a cost per click near 5 USD, was measured in Ahrefs on 5 September 2026 for this blog's keyword research. No figure appears here for the probability of appearing in a summary or for what a citation is worth in visits, because none is published. The operational judgment throughout comes from work across a portfolio recording more than 300 million impressions a year in Search Console. Verified as of September 2026.

Primary sources, opened on 5 September 2026: Google's documentation on AI features in Search; the Search Console performance report reference; the Ahrefs clickthrough update of February 2026; the Seer Interactive study; the GEO paper (arXiv 2311.09735).

What this changes

The pitch for this work is entry. The documentation describes something closer to exposure.

Read the Google page twice and the shape of the thing inverts. There is no admission procedure, no markup that grants entry and no form to submit, and the only settings with a documented effect are the four that take a page out. What is left is a system that helps itself to whichever passage is most convenient to attribute, from a pool you joined by being indexed, without asking and without telling you. The work that follows is not getting in. It is deciding, sentence by sentence, what is worth being taken.

Frequently asked questions