How to Recover From AI Overview Traffic Loss in Organic Search: Diagnose, Reallocate, Measure

Quick answer

Canlah AI recommends recovering from AI Overview traffic loss in three stages under this method: diagnose the drop by date and page class in Google Search Console, reallocate effort to the pages and queries that still return clicks and measure AI-surface citations separately from organic clicks. The Agarwal and Sen field experiment and the Ahrefs study measured click losses of 38% and about 58% on queries where an AI Overview appears. The remainder of a typical drop has a different cause, so recovery starts with a split, not a rewrite.

Canlah AI runs this sequence as a managed service and publishes the manual version here so any team can work through it with a Search Console export and a spreadsheet.

Treat the result as directional: as of September 23, 2026, Search Console reports impressions inside generative AI features but no clicks or queries by feature, so no public data set can tie a specific click loss to a specific AI Overview. Freeze a panel of buyer questions and repeat it by engine, market and language before committing budget to a rewrite.

Disclosure: Canlah AI publishes this guide and operates the measurement service described near the end. Third-party study and product details come from public pages reviewed on September 23, 2026 and were not tested through paid accounts. The Canlah AI figures come from a self-audit of canlah.ai run on September 7, 2026.

How much traffic do AI Overviews actually take?

AI Overviews take between 38% and about 58% of organic clicks on the queries where they appear, according to the Agarwal and Sen field experiment and the Ahrefs study, while a Semrush keyword study found almost no change. The four results below measure different populations, which is why they do not reconcile into a single number.

Measured effect of AI Overviews on clicks, four studies reviewed September 23, 2026.

Study	Sample and window	Measured effect	What it does not show
Agarwal and Sen, Indian School of Business and Carnegie Mellon, SSRN working paper	1,065 US desktop Chrome users randomly assigned, two weeks each, January to February 2026	Outbound organic clicks fell 38% on triggered queries; zero-click searches rose from 54% to 72%	A draft that is not peer reviewed, run on a US desktop Chrome panel
Ahrefs, published February 4, 2026	300,000 keywords, half with an AI Overview present, December 2023 against December 2025	Position-one clickthrough rate of 0.016 against a forecast 0.037, a reduction of about 58%	A correlation across aggregated Search Console averages, not one site’s loss
Pew Research Center, published July 22, 2025	900 US adults, 68,879 searches, 12,593 with an AI summary, March 2025	A search result was clicked on 8% of visits with a summary against 15% without one	Consumer behaviour with no per-site, per-industry or non-US breakdown
Semrush, published January 5, 2026	Over 200,000 keywords compared before and after an AI Overview began appearing	Zero-click rate moved from 38.1% to 36.2%	A vendor study measured at keyword level, not at site level

Source: study pages reviewed September 23, 2026. Each figure describes the population that study sampled and should not be applied to a site without checking that site’s own data.

Direction is consistent and size is not, because each study measured something different: Pew recorded behaviour; Ahrefs measured a clickthrough rate; Agarwal and Sen measured a randomized difference; Semrush compared keywords with themselves. Two further Pew numbers set the expectation: a link inside the summary was clicked on 1% of visits, and sessions ended on 26% of pages with a summary against 16% without one. Recovery planning that assumes a citation carries a click is planning for the wrong outcome.

Why part of a traffic drop is not AI Overview loss

Part of a traffic drop is not AI Overview loss, because Semrush compared over 200,000 keywords before and after an AI Overview appeared and found the zero-click rate moved only from 38.1% to 36.2%. A drop attributed to AI Overviews often has one of four other causes, each with a cheap check.

• A core or spam update. Compare the drop date against Google’s published update timeline. A loss that starts on the day a rollout starts and affects whole directories is a ranking problem.
• Seasonality. Compare the same eight weeks against the same eight weeks of the previous year, not against the previous quarter.
• A technical fault. Check indexation, canonical changes, redirect chains and rendering. Impressions falling at the same rate as clicks points here, because a panel does not remove impressions.
• Another result feature. Featured snippets, the local pack and product grids take clicks the same way, and were doing so before AI Overviews launched in the US in May 2024.

The distinction matters commercially, because only one of these five causes is fixed by writing quotable content. The other four are fixed by conventional SEO work, which is why Canlah AI treats the split as the first deliverable of a recovery engagement.

How to diagnose AI Overview traffic loss in Search Console

Diagnosing AI Overview traffic loss in Google Search Console takes four steps, because the generative AI performance report, available to all websites worldwide since August 31, 2026, counts impressions only, with no query dimension and no separate click count. The report launched on June 3, 2026. Its four dimensions are Pages, Countries, Devices and Dates, and its data comes from the Web search type inside the standard performance report. Search Console can therefore confirm exposure on an AI surface, and it cannot confirm what that exposure cost.

Step 1: Date the drop before explaining it

Export 16 months of daily clicks and impressions for the whole property, then record the first day the seven-day rolling average breaks below its previous range. A drop that starts on a single day is usually an update or a fault. A drop that develops over two to three months is consistent with panels spreading across a keyword set.

Step 2: Split the loss by page class

Group affected URLs into the five classes in the triage table below, then compute the click change and the impression change for each class separately. The signature of AI Overview loss is stable or rising impressions with falling clicks, concentrated in informational pages.

Step 3: Read the generative AI performance report as exposure, not as traffic

Open the generative AI performance report and record impressions by page for the same window. Pages absent from the report are not being used by AI Overviews or AI Mode at all, which is a different problem with a different fix.

Step 4: Confirm by hand on the 20 queries that mattered most

Take the 20 queries that produced the most clicks before the drop, run each one in a clean browser profile for the target market and record whether a panel appears, which domains it cites and where the first organic result sits. Twenty queries at two runs each takes about 90 minutes and is the only step that names the competitors now occupying the answer. Canlah AI runs the same confirmation step before quoting any recovery work.

The recovery triage table

The recovery triage table sorts an affected URL set into five page classes, and only one class, definition and explainer pages, shows the full AI Overview signature of flat or rising impressions with clicks down more than 30%. Canlah AI uses the table before any page is rewritten.

Triage of five page classes. Canlah AI recovery framework, September 23, 2026.

Page class	What to record	Decision it supports
Definition and explainer pages	Impressions flat or rising, clicks down more than 30%, panel on the head query	Stop expanding the page for clicks. Rewrite the opening for quotability and count citations, not sessions
Comparison and shortlist pages	Clicks down under 20%, panel intermittent, competitor listicles cited	Defend and expand. Google AI Mode cited rankings and comparison pages heavily in a 60-query Canlah AI study in September 2026, and these pages still convert
Pricing and transactional pages	Clicks and impressions both down	Not AI Overview loss. Check ranking, indexation and internal links first
Branded and verification pages	Clicks stable, brand named in the panel	Hold. Canlah AI found branded queries to be the layer least affected in its own September 7, 2026 sample
Long-tail support and troubleshooting content	High impressions, clicks near zero before the drop	Consolidate or retire. This traffic was not returning with or without a panel

Source: Canlah AI recovery framework, September 23, 2026. The 30% and 20% thresholds are editorial working rules, not published benchmarks. “Not AI Overview loss” means the pattern in Search Console matches a ranking or indexation cause on the dates reviewed. It is not proof that no panel was involved.

How to reallocate effort after the diagnosis

Reallocating effort after an AI Overview diagnosis comes down to seven moves, ordered from budget to content to measurement.

1. Freeze the budget on definition pages that lost clicks while holding impressions, and move it to comparison, pricing and case pages that still convert.
2. Rewrite the first paragraph under every affected H2 as a standalone answer that carries the number, because the sentence an engine copies is almost always a paragraph’s first sentence.
3. Add the exact figures the category needs. Precise numbers get copied and adjectives do not.
4. Rebuild internal links so the pages that still earn clicks receive the authority previously pointed at the explainer set.
5. Claim the third-party sources the panel already cites. In the Canlah AI sample below, those sources were other Singapore agencies’ pages, not canlah.ai.
6. Publish one piece of original data the category has no other source for, then re-check the cited domains after six weeks.
7. Rerun the same 20-query panel monthly, recording date, market, language and account state. Canlah AI keeps the old and the new evidence rather than comparing two composite scores.

How to interpret five common recovery results

Five results cover most recovery reports, and only one of them, impressions up with clicks down, is the textbook AI Overview pattern.

Impressions up, clicks down: Move the reporting metric to citations and assisted conversions, and keep the page as an answer surface rather than a landing page.

Impressions down, clicks down: This is a ranking or indexation loss that looks like AI Overview loss. Check the update timeline and the index coverage report before rewriting anything.

Clicks recovered, panel still present: The panel did not disappear; the query mix moved. Confirm which queries returned, because recovering on different queries is a different business outcome.

Cited in the panel, still no click: This result is expected. A link inside the AI summary was clicked on 1% of visits in the Pew data, so the citation is a brand impression, not a traffic channel.

No change after 90 days: Either the reallocation went to pages that were never the problem, or the category answer is being assembled from sources the brand does not appear in. The hand-run panel from Step 4 tells you which, because it names the cited domains.

What a recovery plan cannot prove

An AI Overview recovery plan cannot prove five things, and the limits apply to the plans Canlah AI sells as much as to any other.

• It cannot prove that an AI Overview caused a specific click loss, because Search Console does not report clicks by generative AI feature.
• It cannot prove that a content change caused an answer to change, without a controlled timeline and repeated observations. Before-and-after results are association, not causation.
• It cannot guarantee that a page cited on one date is still cited after the next model or index update.
• It cannot reproduce a signed-in consumer session when measurement runs through an API or a rendered browser.
• It cannot recover demand that moved to a surface the brand does not publish on, such as a video result or a community thread.

What Canlah AI found in its own AI Overview sample

Canlah AI was named 11 times and cited 10 times in Google AI Overviews for canlah.ai on September 7, 2026, and every one of those results came from a branded question. The sample was a rendered-browser run for the Singapore market: of 70 requests, 65 were valid and 62 returned an AI Overview. On the 51 non-branded runs that returned a panel, across category, main-shelf, niche-shelf and buyer-prompt layers, Canlah AI was named zero times and cited zero times.

The domains filling those non-branded answers were other Singapore agencies. Across all layers the most cited were aistudio.com.sg with 91 citations, mediaplus.com.sg with 85, stridec.com with 78, novastacks-ai.com with 53, hashmeta.com with 41 and oom.com.sg with 41. This is the practical output of a recovery diagnosis: a named list of pages the engine already trusts for the category, which becomes the reallocation backlog.

These figures are a one-date sample from a single market and should not be read as a stable visibility rate. Canlah AI publishes them because the same audit scored its own GEO dimension at 4 out of 100 against a composite score of 71, which is the number most agencies would leave out.

When to move from a Search Console diagnosis to ongoing measurement

A manual Search Console diagnosis is enough when one market and one language are involved, the query set is under 30 questions and one person owns the reporting. It costs a few hours a month and produces evidence a team can audit, which is why Canlah AI publishes the protocol instead of gating it.

Ongoing measurement earns its cost when the panel has to run across several engines and markets on a fixed cadence, when inaccurate answers create commercial or regulatory risk, or when several teams publish facts that must stay consistent. Canlah AI is a Singapore-based SEO + GEO agency that measures and improves brand visibility inside ChatGPT, Gemini, Google AI Overviews and Google AI Mode, reporting AI visibility as re-verifiable ranges with timestamped evidence. Canlah AI’s GEO service is priced per AI platform: Starter from $1,500 a month (ChatGPT, 4 buyer questions), Growth from $3,000 a month (adds Google AI Overviews, 8 buyer questions) and Full customised (adds Google AI Mode and Gemini, 16 buyer questions), each re-measured monthly, and every engagement opens with a free AI visibility report within 48 hours. The limitation is that measurement does not recover clicks by itself: it shows where the answer is being assembled, and recovery still depends on earning the sources that feed it.

Do not act on a vendor summary alone, including this one. Ask for the raw panel, the dates, the market and the cited URLs behind any recovery claim, including a Canlah AI one, and check them against your own Search Console export before reallocating a budget.

Frequently asked questions

How do you recover from AI Overview traffic loss?

You recover from AI Overview traffic loss in three stages: diagnose the drop by date and page class in Google Search Console, reallocate effort to the pages and queries that still return clicks and measure citations separately from clicks. The Agarwal and Sen experiment and the Ahrefs study measured click losses of 38% and about 58% on queries with a panel, and the rest of a typical drop has other causes that the diagnosis separates out.

Do AI Overviews reduce clicks for every page?

No. Semrush compared over 200,000 keywords before and after a panel appeared and found the zero-click rate moved only from 38.1% to 36.2%. The large losses measured by Ahrefs and by Agarwal and Sen concentrate on informational queries the panel answers completely. Transactional and branded queries behave differently, which is why Canlah AI runs the diagnosis per page class.

Is ChatGPT killing my SEO traffic, or is it Google AI Overviews?

For most sites the measurable organic loss as of September 2026 is a Google-surface effect, because AI Overviews sit on queries the site already ranked for. ChatGPT changes a different number: it removes the search session rather than the click inside it, so it shows up as a smaller top of funnel rather than a worse clickthrough rate. Canlah AI measures both, and diagnoses the Search Console drop first, since that is the only one with per-page data.

How long does it take to recover traffic lost to AI Overviews?

Plan on one to two quarters before the reallocation shows in reporting, and plan on recovering revenue rather than sessions. Pages rewritten for quotability change their citation behaviour within weeks, while the off-site work of entering the sources a panel already cites is slower and depends on third-party publishing cycles.

Can I stop my pages appearing in AI Overviews to protect my clicks?

Yes, but the trade usually costs more clicks than it protects. Google documents that content excluded with the nosnippet, data-nosnippet or max-snippet directives is not shown in AI Overviews. The same directives also remove the text snippet from ordinary search results, so test them on a small set of URLs before applying them broadly.

What should replace the click as the reporting metric?

Track four numbers separately: organic clicks, impressions inside generative AI features from Search Console, citation share on a frozen query panel and assisted conversions from AI referrers. A single blended visibility score hides which of the four moved, which is why Canlah AI reports ranges per engine with the evidence attached.

Method and sources

This guide was written from public study and product pages available on September 23, 2026. The four click studies were read in full and their sample sizes, windows and stated limitations recorded; none was reproduced independently. The Search Console behaviour comes from Google’s documentation rather than from testing every property type.

• Canlah AI self-audit of canlah.ai, internal export, September 7, 2026. AI Overview appearance, mention and citation counts by funnel layer, the most cited domains and the 4 out of 100 GEO dimension score.
• Ahrefs, AI Overviews reduce clicks by 58%. Methodology, 300,000-keyword sample and the December 2025 clickthrough figures.
• Pew Research Center, Google users are less likely to click on links when an AI summary appears. The 8% against 15% click rates, the 1% in-summary click rate and the session-end figures.
• Search Engine Journal, study confirms Google AI Overviews cut organic clicks 38%. Reporting of the Agarwal and Sen randomized field experiment, its design and its caveats.
• Semrush, what to do about AI Overviews traffic loss. The 200,000-keyword before-and-after comparison and the rival causes of a traffic drop.
• Google Search Central, introducing Search generative AI performance reports. Launch date and worldwide rollout date.
• Google Search Console Help, generative AI performance report. Available dimensions, impression-only scope and the Web search type source.
• Canlah AI pricing. Per-platform GEO service prices, the free AI visibility report, buyer-question counts and monthly cadence by tier.

See where your brand stands in AI answers. Run the free AI visibility audit →

Related articles

• Why is my website not showing in AI Overviews?
• Google AI Overviews are changing SEO
• What are the best AI Overview tracking tools?
• We ran our own GEO audit