Sessions down, rankings steady: how to tell if AI is taking your clicks

We hear it in almost every conversation with businesses at the moment: “our sessions have dropped”. It is usually followed by “we realised we need to start working on GEO” and, more and more often, by “we’ve started running ads in ChatGPT”.

All three reactions make sense. But one step comes before any of them: confirming what exactly has dropped, and why. A fall in sessions can be AI keeping your clicks, a genuine loss of rankings, or simply a change in how you measure. Each cause calls for a different response, and you can make a useful first diagnosis in about 15 minutes with data you already have.

The pattern: steady impressions, falling clicks

Since AI Overviews and AI Mode began answering informational searches on the results page itself, the most common reason we find for that shape is that the search is answered without a visit. It is not the only one. More ads or other features above your listing, a shift in which queries bring your impressions, or a change in the country or device mix can produce the same pattern, which is why the diagnosis below rules those out before pointing at AI.

The distinction matters. Losing rankings is a classic SEO problem with well-known fixes. Losing clicks while keeping your position is something else: your content is still relevant to Google, but some of the value that used to arrive as a visit now stays inside the generated answer. At that point the goal is no longer only to rank. It is also to be the name that answer gives.

What has changed in Google

Two changes explain most of the drops we see, and both have dates.

AI Overviews, the summary that sits above the results on many informational searches, launched in the US in May 2024 and reached the UK in August 2024. AI Mode followed in the US in May 2025 and reached the UK on 28 July 2025. It is a full conversational search experience, not a box on the results page.

The effect on clicks has been measured in several studies, though its size varies by method and market:

• Pew Research Center analysed 68,879 real searches by 900 US adults in March 2025. When an AI summary appeared, users clicked a traditional result on 8% of visits, against 15% when it did not, and clicked a link inside the summary itself on just 1% of visits. Google disputed the study’s method.
• Ahrefs found in April 2025 that, across 300,000 keywords, the presence of an AI Overview was associated with a 34.5% lower click-through rate for the page ranking first. Its February 2026 update, comparing December 2023 with December 2025 on aggregated Search Console data, put the gap at 58%, modelled against what the click-through rate would have been had AI Overviews not appeared.

Neither study measures the click-through rate of an individual site cited in an AI Overview: Pew measures user behaviour across searches, and Ahrefs models an aggregate gap for position one in the US. Treat both as evidence of the mechanism, not a forecast for your domain. The useful question is not how far click-through rates have fallen in general. It is how far yours has fallen, and on which searches.

A 15-minute first diagnosis in Search Console

Search Console is the right source for this because its figures do not depend on your analytics cookies or consent banner: impressions and clicks are recorded by Google Search itself. Two things to know about how it counts. Google’s documentation says a click on a link inside an AI Overview counts as a click, and the whole AI Overview occupies a single position shared by every link in it. And a page’s position is the topmost spot any of its links occupies. So if a page is cited in an AI Overview at the top while its ordinary listing slips down, its reported position can stay steady. Read a steady average position as “still being shown”, not as proof the blue link has not moved.

1. Compare with the same period last year, not last month. Performance, web search, the last three months against the same quarter a year ago. A month-on-month comparison mixes seasonality into the drop and produces false conclusions.
2. Open the Generative AI performance report, if you have it. Google rolled it out to all websites on 31 August 2026, but it usually does not appear for sites with too few impressions in AI features. Where it exists (Search results, then Generative AI), it shows which of your pages appear in AI Overviews and AI Mode, with impressions by country, device and date, but no clicks or CTR. Note which pages appear there; you will compare them with the pages losing clicks in a moment.
3. Turn on all four metrics at once: clicks, impressions, CTR and average position. The combination of the four is the diagnosis.
4. Isolate informational queries. Filter for queries containing “what”, “how”, “how much” or “why” and repeat the comparison; in the query filter, Custom (regex) with ^(what|how|why|when|where|can|does|is|are|should)\b does it in one go. Treat the filter as a starting point, not a verdict. “How much does a divorce solicitor cost” is a commercial search phrased as a question, and “best employment solicitor in London” is commercial without a question word. Read the intent behind your biggest losers rather than trusting the words.
5. Split brand from non-brand, and check country and device. Searches for your own name behave differently from searches about your services, and a decline confined to one country or to mobile looks like something else in the total. Filter the brand out (query does not contain your name) and compare again. Query filters drop the anonymised queries Google withholds for privacy, so the filtered figures will not add up to the total; treat them as a comparison, not a complete breakdown.
6. Look at pages, not just the total. Sort by click difference. If the biggest losers are blog posts and guides while service pages hold up, and those guides are the pages showing in the Generative AI report, the pattern is consistent with AI. That is supporting evidence, not proof: the report shows your pages appear in AI answers, not how many clicks those appearances replaced, so look for the same pattern across dates, pages and intent before you conclude. Search the biggest losers yourself too; more ads, shopping results or other features above the organic listings produce the same steady-impressions, falling-CTR shape. If service pages are falling, look for another cause.
7. Export the query data and sort by clicks lost, not by CTR change. A 40% CTR drop on a query that used to bring 2,000 clicks matters far more than an 80% drop on one that brought 30, and an average hides the handful of queries doing most of the damage.

With that done, the reading is almost always one of these:

What you see in Search Console	What it usually means	Where to look first
Impressions and position steady or rising, clicks and CTR falling	Most often absorbed by AI: you appear as before, but the search is answered without a click. Ads and other features above you can do the same	Whether the losers are informational queries, and whether those pages now appear in the Generative AI report
Impressions falling, position steady	A change in demand or query mix: fewer people searching, or different searches bringing your impressions	The query list year on year, and search demand for the biggest losers
Impressions and position falling, CTR steady	A genuine loss of rankings: a Google update, competition or a technical problem	Dates against known updates, then the losing pages and who outranks them
Everything steady in Search Console, sessions falling in Analytics	A measurement problem, or the drop is in another channel rather than organic	Consent set-up, tags, channel definitions
A drop concentrated on a few URLs from a specific date	A change on your side: a migration, a redirect, a noindex or a block	Redirects, indexing, canonicals and internal links from that date

A note on the data: the Performance report itself has no filter that isolates AI traffic. The Generative AI performance report, announced in June 2026 and rolled out to all websites on 31 August 2026, fills part of that gap where it appears: it shows which pages are shown in AI Overviews and AI Mode, with impressions by country, device and date, drawn from the same Web search type. It does not show clicks or click-through rate, so you can see where you appear in Google’s AI answers but not how many visits those appearances produce. Read it alongside the pattern in the table, not instead of it.

Before you blame AI, rule out measurement

The “sessions” causing concern usually come from Google Analytics, and Analytics does not see what Search Console sees. Depending on how Consent Mode is implemented, visitors who decline analytics cookies may not generate conventional sessions at all; with the advanced implementation, Google uses cookieless signals and modelling to estimate some of that missing activity. Either way, a change to the cookie banner, the consent set-up or a Tag Manager container can lower measured sessions without a single real visit being lost. Search Console clicks and Analytics sessions are not interchangeable numbers, and the gap between them has been widening.

Three quick checks:

• Does the drop coincide with a technical change? A new cookie banner, a version published in Google Tag Manager, a redesign. If the date fits, start there.
• Is the tag still firing? A GA4 tag that fails to load, a container published with an error or a changed measurement ID will cut reported sessions overnight. Check real-time reporting and the tag in debug view before looking anywhere else.
• Do Search Console clicks track Analytics organic sessions? They never match exactly, but they should move together. If Search Console is flat and Analytics is falling, the problem is measurement.
• Which channel is the drop in? If direct or referral traffic is falling rather than organic, Google’s AI is not the explanation.

While you are in Analytics, look separately at traffic arriving from AI assistants. This is a different thing from AI Overview visibility: it shows that an assistant sent someone to you, not how often you are named. Visits from ChatGPT show up as referrals from chatgpt.com or carry the parameter utm_source=chatgpt.com; Perplexity, Gemini and Copilot appear as referrals too. Attribution is incomplete, and a fair share arrives as direct traffic, so the referral report understates it. The volume is usually small; compare its engagement and conversion rate with your other sources to see what it is actually worth, and treat it as a clue to whether you are already being recommended.

What to do if AI is the cause

Measure what now matters. If some of your searches are answered inside a generated response, the missing metric is how often that response names you rather than a competitor. That is what we explain in what an AI Visibility Score is and how to measure yours. Without that figure, working on GEO is working blind. If you already track it and the figure itself has fallen, that is a different question, covered in why your AI visibility dropped. Three measures get confused here. AI visibility is whether an assistant names or recommends you. AI citations are whether your pages are used as a source. AI referral traffic is whether anyone clicks through. You can be recommended without a click, cited without being recommended, and sent traffic from one prompt while absent from most others, so track all three. Watch branded search too: someone who meets your name in an AI answer often searches for you later rather than clicking, so rising branded impressions can be a sign that recommendations are working without a link.

Check which pages really matter. Informational guides lose clicks, but many of them never converted. Service, pricing and case-study pages do convert, and they tend to hold up better because they answer searches where the user wants to choose a provider, not read a summary. Protect those first. Fewer clicks are not automatically fewer opportunities: many of the visits lost to AI answers came from people who were never going to instruct anyone. Organic conversion rate can rise even as sessions fall, particularly when the traffic lost was mostly informational. Compare organic enquiries, conversion rate and the pages producing them, not sessions on their own.

Make your content citable. AI engines cite what they can read, understand and attribute. Ranking in Google does not mean AI will name you: we see it often, and explain it in the ghost citations problem. Look too at what AI actually reads on your website and the website mistakes that most often hold back a recommendation.

Work on GEO with a method, not a hunch. Generative engine optimisation does not replace SEO; it builds on it. Our page on generative engine optimisation sets out what that work involves.

Treat advertising as a complement. OpenAI began testing ads in ChatGPT in the US in February 2026, and by August had launched them in the UK, Mexico, Brazil, Japan and South Korea, with more markets planned. An ad buys a position labelled as sponsored, and it can build awareness. OpenAI states that ads do not influence the answers ChatGPT gives, so an ad cannot make the assistant recommend you when someone asks who is best at what you do. One does not replace the other.

What not to do

Do not rewrite the whole blog. If the cause is that the search is answered without a click, rewriting the articles will not bring the clicks back. First measure whether you appear in those answers; then decide what to change.

Do not block AI crawlers in robots.txt on reflex, and know which one does what. OpenAI runs separate crawlers with independent settings: GPTBot gathers content for training its models, while OAI-SearchBot surfaces sites in ChatGPT search. Blocking GPTBot alone keeps your content out of training without removing you from ChatGPT search; blocking OAI-SearchBot means your pages will not be shown in ChatGPT search answers. Other assistants publish their own crawlers, so check what each does before blocking it. None of this affects Google’s AI Overviews, which rely on ordinary Googlebot indexing.

Do not compare one month with the last. Seasonality explains many drops that look alarming. Always compare with the same period last year.

Do not assume you have lost rankings. It is the easiest conclusion to jump to, and when impressions and average position are holding it is rarely the main explanation. Acting on a wrong diagnosis means spending the budget on the problem you do not have.

Do not measure success in sessions or rankings alone. If part of your visibility now happens inside AI answers, sessions tell only part of the story. Add AI visibility and, above all, the enquiries and opportunities that arrive, which is what really matters.

FAQ