AI Overviews SEO: what changed, and what to do about the clicks

What AI Overviews SEO means

AI Overviews SEO is the work of protecting organic traffic from Google’s AI Overviews, the generated summaries shown above the results, and of getting your pages cited inside them. It uses Google’s normal index and ranking, so it is mostly ordinary SEO, plus measuring the click loss and deciding which queries are still worth competing for.

AI Overviews differ from chat assistants in one way that drives the rest of this guide: they appear on a results page where you may already rank. A chat assistant that does not cite you costs a click you never had. An AI Overview above your result can take clicks you were receiving last quarter, on queries you still rank for.

The feature has been part of Google Search in the US since May 2024, when the earlier Search Generative Experience was renamed and launched at Google I/O, as the Wikipedia history of AI Overviews summarises. Google’s Search help page describes it as a core Search feature that users cannot turn off, though the Web filter shows text links without AI Overviews.

The questions this guide answers are the ones an operator faces once the traffic dip shows up: how much was lost and where, what gets a page cited, what a citation is worth, which queries to defend or concede, and whether opting out is ever sensible. How AI retrieval works at the passage level is covered in the pillar guide, what AI search actually retrieves from your page.

How AI Overviews choose which pages to cite

AI Overviews choose sources from Google’s own index: a page must be indexed and eligible to show with a snippet, and Google says there are no extra technical requirements or special optimisations. The model may run several related searches behind the scenes, so the pages cited are not always the ones ranking for the exact query typed.

Google’s documentation on AI features and your website makes three points that settle most of the folklore:

• Eligibility is the normal kind. A page needs to be indexed and eligible to be shown in Google Search with a snippet. There are “no additional technical requirements”, and no special schema.org markup is needed.
• Query fan-out is real. Both AI Overviews and AI Mode may issue “multiple related searches across subtopics and data sources” to build a response.
• The controls are the normal ones. What appears is limited through nosnippet, data-nosnippet, max-snippet or noindex, and crawling through Googlebot’s robots.txt rules.

1. 1QueryA searcher types a question, often a long, informational one
2. 2Fan-outGoogle may run several related searches across subtopics
3. 3IndexCandidates come from pages indexed and eligible for a snippet
4. 4AnswerA generated summary is shown with links to supporting pages

Google Search Central, AI features and your website

How much ranking still matters

Ranking matters, but less directly than it did. In July 2025 an Ahrefs study of 1.9 million citations found 76.1% of AI Overview-cited pages ranked in the top 10 for the query. By March 2026, a larger Ahrefs analysis of 863,000 keyword SERPs and 4 million AI Overview URLs found only 37.9% did, with 31.2% at positions 11 to 100 and 31.0% beyond the top 100.

Ahrefs notes the two studies are not like-for-like: its parsing improved in between and AI Overviews moved to Gemini 3. Its working explanation is fan-out, with Google citing pages that do well across the cluster of sub-queries rather than only the original one; Google does not publish its fan-out queries, so that remains a hypothesis. The practical reading: rank for the query and for the obvious follow-up questions around it, not only the head term.

The May 2025 Ahrefs study of 75,000 brands adds an off-page factor: branded web mentions had the strongest correlation with AI Overview brand visibility of anything tested there (Spearman 0.664), ahead of Domain Rating (0.326). A later Ahrefs study across three AI surfaces found YouTube mentions correlated slightly more strongly still. It is a correlation, not proof, but it is consistent with the fan-out explanation: brands discussed across many pages show up across many sub-queries.

AI Overviews and AI Mode

AI Mode is Google’s conversational results view, and the same documentation covers both features together: the same eligibility, the same fan-out technique and the same controls. The line between the two is blurring. In December 2025 Google began testing a mobile flow in which a follow-up question on an AI Overview opens in AI Mode, as Search Engine Journal reported. At Google I/O on 19 May 2026, Sundar Pichai said in his keynote that AI Overviews had over 2.5 billion monthly active users and AI Mode had passed one billion, and Google’s Search announcements described asking a follow-up question right from an AI Overview and continuing the conversation in AI Mode. For measurement the difference barely matters, because traffic from both is reported inside the Web search type in Search Console. For planning, AI Mode points the same way: a conversational answer that can take follow-up questions is likely to leave even fewer reasons to click for a simple query.

Which queries trigger AI Overviews

Informational, longer and question-shaped queries trigger AI Overviews far more often than short ones. In Pew Research Center’s analysis of 68,879 searches by 900 US adults in March 2025, 18% of searches produced an AI summary, but only 8% of one- or two-word searches did, against 53% of searches of 10 words or more and 60% of searches starting with a question word. Ahrefs’ April 2025 click study found 99.2% of the keywords triggering AI Overviews in its dataset were informational.

Pew’s data also shows where the summaries send people. Wikipedia, YouTube and Reddit were the most frequently linked sources, making up 15% of the sources in AI summaries and 17% in standard results. For a B2B brand that is a reminder that the pages cited above your result are often not competitors’ sites but encyclopaedia entries, videos and forum threads, and that presence on those is part of the same job.

How much traffic AI Overviews take

AI Overviews reduce clicks to the organic results beneath them, most of all on the top-ranking page for informational queries. The best public estimates put the loss at roughly a third to well over half of position-one clicks, depending on the period and the method.

Study	Data	Finding
Ahrefs, Feb 2026	300,000 keywords, desktop CTR from aggregated Search Console data, Dec 2023 vs Dec 2025	AI Overview presence correlated with 58% lower CTR for the position-one page (0.016 actual vs 0.037 forecast)
Ahrefs, Apr 2025	300,000 keywords, Mar 2024 vs Mar 2025	34.5% lower position-one CTR (0.026 actual vs 0.040 forecast)
Pew Research Center, Jul 2025	Browsing data from 900 US adults, March 2025	Users clicked a traditional result on 8% of visits with an AI summary, 15% without one
Pew Research Center, Jul 2025	Same panel	A link inside the AI summary was clicked on 1% of visits; 26% of summary pages ended the browsing session, against 16% without

Ahrefs, 2025 and 2026; Pew Research Center, 2025

The February 2026 Ahrefs update also estimated smaller effects lower on the page, from about minus 51% at position 2 to about minus 19% at position 10. Its caveats are worth repeating to anyone who wants to quote the 58%: it is a correlation, it uses desktop data only, the forecast assumes AI Overview keywords would otherwise have declined like other informational keywords, and featured snippets and other SERP features have been eroding CTR for years.

Google’s position is different. The same AI features page says clicks from results pages with AI Overviews are “higher quality”, meaning users are more likely to spend more time on the site. Both things can be true: fewer clicks, and the ones that remain from people who wanted more than the summary. Your own analytics, not either claim, should decide which describes your site.

Which queries lose clicks to AI Overviews

Queries lose clicks to AI Overviews in proportion to how completely a paragraph can answer them. Definitions and single-step how-tos lose the most, comparisons lose some, and commercial and navigational queries lose least, because the searcher still needs to see options, prices or a specific site.

Query type	Exposure	Why
Definition: what is X	Highest	A complete answer fits in the Overview. Nothing left to click for
Single-step how-to	Highest	Same reason
Multi-step or situational how-to	Moderate	The summary gives the outline; the reader clicks for detail that depends on their case
Comparison: X vs Y	Moderate	Summarised, but buyers want detail, pricing and a source they trust
Commercial: best X for Y	Lower	People want to see the options and vendors themselves
Navigational or branded	Lowest	The searcher already knows where they are going

Consistent with Ahrefs’ finding that 99.2% of AI Overview keywords were informational

The strategic reading: informational content at the top of the funnel has become less valuable as a traffic source, and commercial-intent content relatively more so. A programme built on high-volume definitional guides was always weaker than its traffic reports suggested, because those visitors rarely converted; AI Overviews have removed the traffic that disguised it.

How to measure AI Overview impact in Search Console

Measuring AI Overview impact in Search Console means comparing click-through rate before and after, at a stable average position, for queries that now show an AI Overview against queries that do not. Search Console counts AI Overview clicks inside the ordinary Web search type, so the click comparison has to be built by hand.

Google’s AI features documentation confirms the reporting: sites in AI Overviews and AI Mode are included in the overall search traffic in Search Console, under the Web search type. There is no AI Overview filter on that report. In June 2026 Google added a separate beta Generative AI performance report which, as Search Engine Roundtable reported, shows impressions in AI Overviews, AI Mode and AI features in Discover, broken down by page, country, device and date. It has no clicks and no queries, and Search Engine Journal reported that it reached all websites worldwide on 31 August 2026. Use it to see which pages appear in AI features and how often; use the cohort method below to estimate what that costs in clicks.

The procedure

1. Export the Performance report at query level (query, clicks, impressions, CTR, position) for two comparable periods, such as the same quarter a year apart, to cancel out seasonality.
2. Keep queries with enough impressions to give a stable CTR in both periods, and whose average position moved by less than one place. A query that dropped from 2 to 6 lost clicks for a ranking reason, not an AI Overview reason.
3. Tag each remaining query: AI Overview present now or not. Check by hand for your top queries in a signed-out private window, and use the SERP feature filter in a rank tracker such as Ahrefs or Semrush for the long tail. The Generative AI report’s page list shows which landing pages to check first.
4. Tag each query by type (definition, how-to, comparison, commercial, branded).
5. Compare the CTR change for queries with an AI Overview against queries without one, inside each type. The difference between the two is your estimate of the AI Overview effect.
6. Convert the click difference into conversions and pipeline using your landing pages’ own conversion rates, so the conversation is about revenue rather than sessions.

Estimate your own click loss

The calculator applies the same method to your numbers. Use the CTR of comparable queries without an AI Overview, at stable position over the same period, as the baseline. It computes expected clicks = impressions * baseline CTR / 100, actual clicks = impressions * current CTR / 100, clicks lost = expected - actual, conversions lost = clicks lost * conversion rate / 100 and value lost = conversions lost * value per conversion. A negative result means the query did better than its control cohort.

Enter one query or one cohort of tagged queries for the current period.

Impressions this periodBaseline CTR: similar queries without an AI Overview%Current CTR on the query with an AI Overview%Landing page conversion rate%Value per conversion$

Expected clicks without the AI Overview

2,268

Actual clicks

1,386

Clicks lost to the AI Overview

882

Share of expected clicks lost

38.9%

Conversions lost

4.4

Value lost

$1,323

Per period, at your conversion value

Hypothetical defaults, matching the worked example above; the value per conversion is invented for illustration. Change any number.

How to get cited in AI Overviews

Getting cited in AI Overviews takes a page Google already trusts for the query or its sub-queries, with a passage that answers the question directly enough to be lifted out. There is no separate markup or submission route; the work is conventional ranking plus passage-level editing.

The page-level procedure

1. Confirm eligibility: the page is indexed, not blocked by robots.txt, and not carrying nosnippet or a restrictive max-snippet that would keep it out of snippets.
2. List the follow-up questions around the target query from People Also Ask, related searches and your own Search Console long tail. These approximate the fan-out sub-queries Google does not publish.
3. Give each important question its own heading, phrased the way searchers phrase it, with a direct answer in the first one or two sentences beneath.
4. Make the answer specific: a number with its source, a named method, a concrete constraint. Generic phrasing is interchangeable with every other page and gets swapped out.
5. Name the subject in each section’s opening sentence instead of writing “this” or “it”, so the passage still makes sense when read alone.
6. Strengthen ranking for the sub-queries with internal links from related pages, using anchors that match the sub-question.
7. Recheck the SERP after the page is recrawled, and log whether you are cited, in the way described in the guide to tracking brand mentions in AI search.

How to find the sub-queries around a target

Google does not publish its fan-out queries, so the practical substitute is to map the questions a searcher would ask next. Four sources cover most of them:

• People Also Ask and related searches on the target query, opened two levels deep.
• Search Console, filtered to the page: the long-tail queries it already earns impressions for are sub-questions Google already associates with it.
• Rank tracker SERP data: Ahrefs Keywords Explorer and Semrush both show related questions and the pages ranking for each.
• The current AI Overview itself: the sub-topics it covers, and the pages it cites for each, show the cluster Google built for that query.

Then decide, for each sub-question, whether it belongs as a section on the target page or as a separate page linked from it. If the answer needs more than a few paragraphs or has its own search demand, give it its own page; otherwise make it a heading on the main one. Do not turn every variation into a page. Google’s AI optimization guide warns that creating separate pages for each fan-out variation, primarily to manipulate rankings or AI responses, violates its scaled content abuse policy.

Before and after (hypothetical)

Take a hypothetical project-management SaaS page targeting “how to calculate project float”. The original section opened: “Project managers have many tools at their disposal, and understanding schedule flexibility is something every team benefits from. In this section we will look at it.” Nothing in it can be quoted, and the subject is never named.

The rewrite opens: “Total float is the latest finish date of a task minus its earliest finish date. A task that can finish on day 12 at the earliest and day 15 at the latest has 3 days of float; a task with zero float is on the critical path.” The heading matches the query, the first sentence answers it, the subject is named, and the worked figure gives a summary something specific to lift. The longer explanation, with exceptions and a diagram, follows underneath for the reader who clicks.

A passage checklist

• Heading matches a question searchers ask.
• First sentence answers it without needing the paragraph before.
• The subject is named, not referred to.
• The claim comes before its qualifications.
• Any number has its source in the same sentence.
• The passage is under about 80 words before the detail starts.
• The page states something a generic summary could not: data, a judgement or a situation-specific rule.

To check a passage before publishing, paste it into the AI Citability Checker, which scores each passage against seven rules, shows it without its heading and previews the effect of snippet controls such as max-snippet.

What an AI Overview citation is worth

An AI Overview citation is worth brand exposure at the moment of the question and comparatively few clicks. Pew found a link inside the summary was clicked on 1% of visits, so a citation should be valued as visibility and credibility, not as replacement traffic.

That has an awkward consequence. Optimising hard to be cited on a query that used to send you clicks rarely brings those clicks back, because the mechanism that would cite you is the same one that removed the need to click. The citation is still worth having where the brand impression matters, such as a category or comparison query where buyers are forming a shortlist.

Situation	Value of being cited	Reason
Category or comparison query, buyer forming a shortlist	High	Your name appears as an option at the decision point
Problem query in your product’s domain	Medium	Associates the brand with the problem before the buyer knows the category
Definition far from your product	Low	Read and forgotten; no commercial follow-through
Query about your own brand	Critical, but for accuracy	What the Overview says about you is what every searcher reads

The bottom row is worth checking monthly regardless of everything else. Search your own brand with pricing, integration and comparison terms, signed out, and read what the Overview says.

Which queries to concede and which to defend

Concede a query when the AI Overview answers it completely and the traffic never converted; defend it when the answer depends on the reader’s situation, changes over time or sits next to a purchase decision. Conceding means stopping investment, not deleting the page.

Signal	Reading	Action
Answer is one uncontested paragraph	Absorbed by the Overview	Concede. Keep the page if it supports a cluster; stop updating and promoting it
Answer depends on the reader’s situation	Not fully answerable in a summary	Defend. Add the situational detail the summary cannot give
Answer changes over time	Freshness favours a maintained page	Defend, and show the update date and what changed
Query sits next to a purchase	Less exposed	Invest here. This is where freed budget goes
Page lost clicks but kept conversions	The lost visitors were not buyers	Leave it alone

A scoring template

Score each affected query from 0 to 2 on four questions and add them up. Defend anything scoring 5 or more, review 3 or 4, and concede 2 or less.

• Commercial value: 0 = no conversions from the landing page, 1 = some assisted conversions, 2 = direct conversions.
• Answer completeness in the Overview: 0 = fully answered, 1 = partly, 2 = the Overview cannot answer it without the reader’s details.
• Freshness: 0 = stable for years, 1 = changes yearly, 2 = changes monthly or faster.
• Cluster role: 0 = standalone, 1 = linked from a few pages, 2 = the page other cluster pages depend on for a term or concept.

Applied to the hypothetical glossary page above: commercial value 0, completeness 0, freshness 0, cluster role 2. Total 2: concede, keep the page live for the cluster, and move the effort.

Commercial value0: no conversions1: assisted conversions2: direct conversionsAnswer completeness in the Overview0: fully answered1: partly answered2: needs the reader’s detailsFreshness0: stable for years1: changes yearly2: changes monthly or fasterCluster role0: standalone1: linked from a few pages2: other pages depend on it

Score out of 8

2

5 or more: defend. 3 or 4: review. 2 or less: concede, keep the page live if its cluster role is 1 or 2.

Points short of defend

3

Defaults are the hypothetical glossary page above. The thresholds are this guide’s rule of thumb, not a measured cut-off.

Removing a conceded page is usually a mistake. A definitional page may be the one that establishes a term the rest of a cluster depends on, and deleting it removes the internal links it was passing on. The method for that decision is in the content audit guide: judge a page on what it earns now, not on what it used to earn.

Should you opt out of AI Overviews?

Opting out of AI Overviews can be done for a whole site with a Search Console setting, or page by page with Google’s snippet controls, which also limit how the page appears in ordinary results. For most marketing sites either route costs more than it saves, because it removes the citation along with the summary.

The site-level route is new. In June 2026 Google added a property-level control in Search Console that, as Search Engine Roundtable reported, stops a site’s content being used in AI Overviews, AI Mode and AI Overviews in Discover, both as links and for grounding. Opted-out sites get no impressions or traffic from those features, and Google said blocking would not be used as a ranking signal for results outside them. It began with a subset of UK sites; on 31 August 2026, Search Engine Journal reported, Google noted that the AI reports had rolled out to all websites worldwide, and the control went global on the same date. It does not govern training, which stays with Google-Extended, and it does not affect other assistants.

The page-level route is the older one. Google’s AI features documentation lists nosnippet, data-nosnippet, max-snippet and noindex as the ways to limit what is shown in Search, and points to Google-Extended only for training and grounding in some other Google systems. The trade-offs differ by control:

Control	What it limits	Cost
Search Console generative AI control (site)	AI Overviews, AI Mode and Discover AI features, for the whole property	All AI-feature impressions and citations; Google says classic rankings are unaffected
nosnippet (page)	Any text snippet from the page	The ordinary result loses its description too, which usually lowers CTR
data-nosnippet (element)	Snippets from the marked section only	Precise, but the marked text can no longer be quoted anywhere in Search
max-snippet:[n]	Snippet length	A blunt instrument; short limits also shorten normal snippets
noindex	The page appears nowhere in Search	Removes all organic traffic to the page

Is the content itself what you sell, such as journalism, paid research or licensed data?

Yes

Would a summary of it remove the reason to visit or pay?

Yes, for the whole site

Consider the site-level Search Console control. You lose all AI-feature impressions; Google says classic rankings are unaffected.

Only for some sections

Use data-nosnippet on those elements and leave the rest of the page eligible.

No, it is marketing content

Stay in. Opting out removes the citation along with the summary, and the page-level controls also weaken the ordinary snippet.

Controls as described in Google’s documentation and the Search Console reports linked above.

A sensible use of data-nosnippet is narrow: a paid or gated asset, such as the key figures from proprietary research, where being summarised would remove the reason to visit. For ordinary marketing pages, opting out usually means losing the citation and, with the page-level controls, some of the ordinary click-through at the same time. The site-level control makes sense mainly for publishers whose business is the content itself, and who would rather lose AI impressions than be summarised.

What AI Overviews change in a content programme

AI Overviews shift the return on content away from definitional and synthesis pieces and toward material a summary cannot reproduce: original data, situation-specific guidance, judgement with reasoning, and commercial pages that help a buyer decide. The volume playbook of many guides on high-volume questions is the part that no longer pays.

Content type	Holds up	Why
Original data or research	Well	The summary has to cite you to use the number
Situation-dependent guidance	Well	The right answer depends on details only the page can work through
Comparison and pricing pages	Well	Buyers click to verify and decide
Opinion with reasoning	Moderately	Distinctive enough to be named, rarely enough to earn a click
Synthesis of existing sources	Poorly	The Overview is itself a synthesis
Definitional explainers	Poorly	Answered in a sentence above your result

If most of a library sits in the bottom two rows, a large part of it has lost its traffic rationale. The response is to move new investment up the table, not to rewrite the old pages. For a B2B team that usually means more comparison, alternatives and pricing-explainer pages, and fewer top-of-funnel guides, which is the same direction argued in the B2B content marketing guide for distribution reasons.

The constructive reading is that this pushes content toward being worth reading, because generic content no longer earns traffic. That is a harder standard and a better one.

How to monitor AI Overviews over time

Monitoring AI Overviews works best at the level of query cohorts and quarters, not individual queries and days, because AI Overviews appear inconsistently from one search to the next. Track CTR at stable position for tagged cohorts, and AI Overview presence on your priority queries.

• Watch: CTR at stable position, by cohort, quarterly. A steady decline on informational queries while impressions hold is the signature.
• Watch: AI Overview presence on commercial queries. If they start appearing on your money terms, the exposure table above changes for you.
• Watch: conversions from organic landing pages. The number that decides whether the click loss matters.
• Ignore: daily presence counts. A tool reporting that Overviews appeared on a given share of your terms this week is mostly measuring how often it checked.
• Ignore: total organic sessions as the headline. It mixes affected and unaffected queries.

Keep a change log of known shifts, such as the move to Gemini 3 that Ahrefs noted between its 2025 and 2026 studies. Any measurement describes the current behaviour of a feature that keeps changing, so respond to the shape of the effect, which has been consistent, rather than to a single quarter’s magnitude. If you want the cohort analysis and the concede-or-defend scoring done as part of a wider programme, that is part of the SEO systems service.

A first-quarter plan for AI Overviews

A first-quarter plan for AI Overviews starts with measuring the damage by cohort in week one, makes the concede-or-defend calls in month one, and by the end of quarter one has moved budget toward the queries that still convert. The order matters: measuring first stops a team rewriting pages that were never the problem.

Week 1

Measure

Export query data for the same quarter a year apart, filter to stable positions, tag AI Overview presence and query type, and estimate the click and conversion effect per cohort.

Month 1

Decide

Score affected queries with the concede-or-defend template. Rewrite the defended pages’ key passages with the checklist. Check what the Overview says about your own brand.

Quarter 1

Reallocate

Move planned content spend from conceded informational topics to comparison, alternatives and pricing pages. Re-measure the cohorts and compare against the week-one baseline.

What not to do in that quarter: rewrite the whole library, delete pages because they lost clicks, or buy a monitoring tool before you know which queries matter. A page that lost half its traffic and still converts is not a failed page.