AI Overviews and organic traffic decline: a diagnostic guide for UK websites

A fall in organic traffic does not prove that Google AI Overviews took the clicks. The same graph can be produced by weaker demand, lower rankings, an indexing mistake, a site migration, a reporting anomaly, changed search-result features or a broken conversion journey. Treating all of those as an 'AI traffic problem' leads to the wrong work.

The right response is a diagnosis. Compare impressions, clicks, click-through rate, position, indexed pages and business outcomes at page and query level. If the evidence points to AI-assisted results changing click behaviour, adapt the content and measurement model. If it points elsewhere, fix the actual cause.

Can AI Overviews reduce organic traffic?

Yes, particularly when a generated summary satisfies an informational need without a visit. But the effect is not uniform across every query, page or business model. A short factual article, an original comparison, a local service page and a checkout page do different jobs and should not be expected to respond in the same way.

Pew Research Center observed this behaviour in browsing data from 900 US adults in March 2025. Participants clicked a traditional result on 8% of visits when an AI summary appeared, compared with 15% when it did not; links inside the summary were clicked on 1% of visits. This is useful original research, but it is US evidence from one period—not a UK-wide conversion rate.

A 2026 causal study compared English Wikipedia pages exposed to AI Overviews with matched pages in language editions that were not exposed during the same rollout. It estimated an approximately 15% decline in daily traffic to the exposed English articles, with larger effects for topics that a short summary could more easily satisfy. That is credible evidence of substitution for one major informational publisher; it is not proof that every commercial website will lose 15%.

Five traffic patterns—and what they usually suggest

Start with the shape of the decline in Google Search Console. These patterns are diagnostic clues, not automatic verdicts, but they narrow the investigation quickly.

1. Clicks fall while impressions and position remain broadly stable

This is the pattern most consistent with a click-through-rate change. An AI Overview, featured snippet, local pack, shopping result, stronger competitor snippet or changed title can all take attention while your page continues to appear. Segment the affected queries and inspect the current result pages before assigning the cause.

2. Clicks and impressions fall while position remains stable

Search demand may have declined. Check year-over-year comparisons, seasonality, Google Trends and whether customers now describe the need differently. A stable position cannot produce the same traffic if fewer people search for the topic.

3. Clicks, impressions and position fall together

Investigate ranking loss, competition, changed intent, a core update or a site-wide quality issue. Check Google's ranking-status history, identify whether the decline affects one page group or the entire site, and compare the pages now ranking above yours. An AI Overview may be present, but it does not explain the loss of position.

4. Indexed pages or eligible URLs fall

Treat this as a technical or indexing investigation first. Check status codes, robots directives, canonicals, rendering, internal links, sitemaps and recent releases. A page that has left the index cannot earn a conventional result or an AI-feature impression.

5. Search clicks hold up but leads or revenue fall

The problem may be attribution, tracking, page experience, offer quality or lead intent rather than visibility. Test forms, calls and booking journeys, then compare conversion rate and lead quality by landing page. Traffic recovery will not repair a broken enquiry path.

A 90-minute traffic-decline audit

Step 1: establish the dates and rule out bad reporting

Record the first clear date of the decline and any site release, migration, consent change, analytics change or campaign ending near it. Check Search Console's data-anomalies page. Compare at least 28 days before and after, then use year-over-year data where seasonality matters. Avoid comparing a complete month with a partial one.

Step 2: use Search Console as the search record

Filter the Search results report to the United Kingdom if that is the market being diagnosed. Compare clicks, impressions, CTR and position across pages, queries and devices. Sort pages by click difference to find where the loss is concentrated instead of diagnosing from the property total.

• One page falling usually calls for a page-and-query investigation.
• One directory or template falling suggests a shared intent, content or technical issue.
• A site-wide fall makes migrations, indexing, manual actions, security and broad ranking changes more important.
• Mobile-only loss points towards result layout, device experience or mobile technical behaviour.
• UK-only loss may reflect local demand, competition or country-specific result features.

Step 3: separate branded and non-branded demand

Use Search Console's branded and non-branded query filter where it is available. A fall in non-branded informational clicks with steady brand demand is a different problem from fewer people searching for the company by name. The filter can misclassify queries and is unavailable for low-impression properties, so treat it as a useful segment rather than perfect attribution.

Step 4: inspect the Generative AI performance report

Google is rolling out a separate Generative AI performance report to a subset of Search Console properties. It reports impressions from AI Overviews and AI Mode by page, date, country and device. If you have access, compare the pages gaining AI-feature impressions with the pages losing Web clicks.

The report is valuable but limited: it currently reports impressions rather than a complete AI-specific click-and-query journey, and its data is also included within the normal Web performance report. An AI impression appearing beside a click decline is evidence of exposure, not on its own a causal measurement of lost traffic.

Step 5: sample the real result pages

For the material queries that lost clicks, record what now appears above and around the result: AI Overview, featured snippet, local pack, video, shopping block, forum result, competitor or a changed title. Use a consistent device, UK location and date. Search results vary, so a small documented sample is more honest than claiming a permanent layout from one manual search.

Step 6: connect search data to business outcomes

Review organic landing-page sessions, qualified enquiries, bookings, assisted conversions and revenue. Search Console records Google visibility and clicks; analytics records visits it can observe; the CRM records whether those visits mattered. A traffic decline with stable qualified leads needs a different response from a decline that removes the pages responsible for most revenue.

Crawled – currently not indexed is a separate diagnosis

A rise in 'Crawled – currently not indexed' URLs does not demonstrate an AI Overview traffic loss. It means Google fetched a URL but the Page indexing report does not currently show it as indexed. The aggregate report can lag behind the live URL status, so inspect a sample of commercially important URLs individually before drawing conclusions.

1. 1Use URL Inspection to check the current index status and Google's selected canonical.
2. 2Run the live test and inspect rendered output when JavaScript or blocked resources could hide the main content.
3. 3Confirm the URL returns 200, is not noindexed, has a self-consistent canonical and is internally linked.
4. 4Compare it with pages serving the same intent. Consolidate near-duplicates rather than requesting indexing repeatedly.
5. 5Prioritise important service, product and evidence pages; not every discovered URL needs to be indexed.

Choose the recovery plan that matches the evidence

When AI or other result features reduce CTR

Protect the direct answer, but give the right visitor a reason to continue. Add value that a summary cannot fully deliver: first-hand data, a worked example, original photography, a comparison method, current pricing logic, an interactive tool, availability, a case study or a decision with meaningful trade-offs.

Shift some effort from broad definitions towards evaluation and action. A visitor may not click for 'what is heat-pump servicing', but may still need a detailed cost comparison, eligibility check, local availability or quote. Do not make the basic answer vague merely to manufacture a click.

When rankings decline

Compare the affected page with the pages now satisfying the query. Check whether the intent changed, your evidence became stale, competing pages are more complete or several of your own pages split the same purpose. Improve the useful substance and consolidate overlap; avoid a site-wide rewrite based on a small position fluctuation.

When indexing declines

Fix shared technical causes first, then strengthen page value. Restore correct status codes, canonicals, renderable main content and crawlable internal links. Remove accidental noindex directives. Merge thin or repetitive pages where one stronger URL can answer the need. A sitemap can support discovery, but it cannot force indexing.

When demand declines

Preserving a ranking cannot recreate missing demand. Revisit customer language, adjacent problems and more resilient acquisition channels. Build brand familiarity through partnerships, email, events, useful communities and public expertise so search captures demand created elsewhere rather than carrying the whole growth target.

When conversion declines

Repair the journey before buying more visibility. Match the landing page to the query, make the next step clear, test forms and phone links on mobile, and explain price, scope, proof and response time. Track qualified outcomes rather than celebrating form submissions that never become customers.

What not to do after an AI traffic scare

• Do not rewrite every page into short question-and-answer blocks; a format is not an AI Overview recovery strategy.
• Do not publish dozens of near-identical articles to replace one declining topic cluster.
• Do not add unsupported statistics, artificial first-hand experience or unnecessary schema to appear citable.
• Do not remove useful informational content solely because its last-click traffic fell; it may still build recognition and support later decisions.
• Do not repeatedly request indexing without changing the condition that made a page low-value, duplicative or technically unreliable.
• Do not report a third-party CTR study as the forecast for your UK website.

A 30-60-90 day recovery plan

Days 1–30: establish the cause

Export the affected Search Console data, record releases and external events, inspect the largest losing pages, check index coverage and sample the relevant result pages. Select no more than one or two plausible causes for each page group and preserve the baseline.

Days 31–60: improve the highest-value group

Fix shared technical problems, consolidate overlap and improve the pages closest to a customer decision. Add original evidence where it changes usefulness, update titles only where they misrepresent the page, and strengthen internal routes from informational content to relevant services or tools.

Days 61–90: measure outcomes, then expand

Compare clicks, impressions, CTR, positions, generative-AI impressions where available, conversions and lead quality with the baseline. Allow enough time for recrawling and normal variation. Expand the intervention only when the selected page group shows a useful signal.

Common AI Overview traffic questions

Can Search Console show exactly how many clicks an AI Overview removed?

No. The Generative AI performance report can show AI Overview and AI Mode impressions for eligible properties, but it does not create a counterfactual showing how many clicks the same query would have produced without the feature. Use it alongside Web performance trends and documented result-page samples.

Should we delete articles that lose informational traffic?

Not automatically. Assess whether the article supports customers, links naturally to a decision, earns references, builds brand recognition or provides source material that search systems use. Merge or remove it when it duplicates another page or no longer offers a useful purpose—not merely because sessions fell.

Will adding FAQ or Article schema recover the clicks?

No supported schema guarantees an AI Overview appearance or restores CTR. Use structured data only when it accurately describes visible content and qualifies for a documented search feature. The recovery still depends on access, relevance, value and the customer's reason to continue.

How long should we wait after making improvements?

It depends on crawl frequency, the scale of the change and the cause. Technical fixes can be recognised after recrawling, while broader quality reassessment may take weeks or months. Record the change date and review a stable comparison window instead of changing the page again after a few days.