Contents

    When rankings hold but clicks have been falling for months, the cause for Swiss SMEs is often a Google AI Overview. The pattern is easy to spot in Search Console: position and impressions stay stable or rise, click-through rate falls, mainly for questions. For searches for a provider, AI Overviews are rare in Switzerland. Before you act, it pays to rule out other causes. After that, the rule is: if questions no longer bring clicks, you have to appear in the answer itself. That only works when classic SEO and AI visibility are planned together.

    This article is for marketing leads whose SEO report looks good while enquiries decline. It shows how to narrow down the cause in Search Console, what to rule out, what the studies say about lost clicks, what to measure instead and what to do now. How AI Overviews work in Switzerland and whom they cite is covered in our measurement of Google AI Overviews in Switzerland.

    How to spot the AI effect in Search Console

    Clicks from AI Overviews and from Google's AI Mode are counted in Search Console as part of normal web search. There is no filter for «clicks from the AI Overview». You can only recognise the effect by its pattern. Here is how:

    1. Pick the right period. Open «Performance › Search results» and compare the last three months with the same three months a year earlier. A comparison with the previous month mostly shows holidays.
    2. Show all four metrics. Clicks, impressions, average click-through rate and average position. Only together do they tell you something.
    3. Split queries into groups. In the «Query» filter you can choose «Custom (regex)». For German questions, for example ^(was|wie|wann|warum|welche|wer|wo|kann|muss|darf|lohnt) , for English ones ^(what|how|when|why|which|who|where|can|do|does|is) . For provider searches, something like zürich|zurich|winterthur|agentur|agency|anbieter|firma with your own locations and terms. For long, fully written-out queries of eight or more words, ^(\S+\s+){7,}\S+$ helps.
    4. Compare the groups. If click-through rate falls mainly for questions while impressions and position hold, the AI Overview is the most likely explanation.
    5. Look for yourself. Search the ten questions with the biggest click loss on Google. Does an AI Overview appear? Are you cited in it? Who is there instead?

    Why splitting questions from provider searches matters so much is shown by our own measurement: in October 2026 we ran 84 searches from 12 industries from Zurich. For 56 of 60 questions about costs, problems, rules, terms and comparisons, Google showed an AI Overview. For the 24 searches for a provider, it did so only twice. There, Google almost always showed the map with businesses.

    This table helps you classify what you see:

    ImpressionsPositionClick-through rateLikely cause
    stable or higherstablefalls, mainly for questionsAI Overview or AI Mode
    fallstablestableless demand: season, market, brand
    fallworsefallsa real ranking loss
    stablestablefalls for provider searches tooads, the map or a weak snippet
    stablestablestable, but GA4 shows fewer visitstracking or cookie banner

    What to rule out before you blame AI

    Not every lost click comes from AI. A study by Seer Interactive found that click-through rate also fell for searches without an AI Overview, just less sharply. So check these seven points:

    • Seasonality. Many Swiss industries have clear cycles: summer holidays, the weeks around Christmas, ski holidays that fall at different times depending on the canton. The year-on-year comparison shows whether the drop is normal.
    • Ads and the map. When more ads or the map sit above your result, click-through rate falls even though your position stays the same. For provider searches, Google showed the map in 22 of 24 searches in our measurement.
    • Brand demand. Filter the queries that contain your company name. If those clicks fall, fewer people are looking for you specifically. That is a demand or awareness issue, not an AI Overview problem.
    • Tracking and consent. Search Console counts clicks regardless of cookies. GA4 only counts what it may measure after consent. If only organic sessions in GA4 fall but Search Console clicks do not, check the cookie banner, Consent Mode and the Google tag.
    • Cannibalisation. When two of your pages alternate for the same search, position fluctuates and neither page wins. In Search Console, click a query and open the «Pages» tab.
    • Losses hidden by the average. Average position is a mean. If your most important page drops from position 2 to position 6 while many new queries are added, the average can look stable. Check the ten pages and queries with the most clicks individually.
    • Anonymised queries. Google does not show queries that were issued by no more than a few dozen people over two to three months. They are included in the totals, but not in the query table or when a filter is applied. Part of your clicks therefore cannot be traced to any search.

    What the studies say about lost clicks

    Several studies show that AI Overviews cost clicks. They do not agree on how many. Most of the data comes from the US and from English-language searches.

    StudyDataResult
    Pew Research (July 2025)68,879 searches by 900 US adults, March 2025A click on a traditional result in 8 percent of cases with an AI summary, 15 percent without. 1 percent clicked a link inside the summary.
    Ahrefs (February 2026)300,000 keywords, desktop, December 2023 versus December 2025Click-through rate of the top result 58 percent lower when an AI Overview appears. Around 51 percent lower at position 2 and around 19 percent lower at position 10.
    Seer Interactive (November 2025)3,119 informational queries from 42 organisations, June 2024 to September 2025Organic click-through rate for searches with an AI Overview fell from 1.76 to 0.61 percent, minus 61 percent. Without an AI Overview, minus 41 percent.
    Amsive (April 2025)around 700,000 keywords from 10 websitesMinus 15.5 percent on average, minus 20 percent for searches without a brand name, plus 18.7 percent for brand searches with an AI Overview.

    Google sees it differently. In August 2025, Head of Search Liz Reid wrote that total organic click volume from Google to websites had been relatively stable year on year. Clicks had become higher quality, but traffic was shifting between websites: some lose, others gain.

    The figures are far apart because the studies measure different things: different periods, different industries, desktop only or actual browsing behaviour. What they share is the direction. For informational questions click-through rate falls clearly, for brand searches hardly at all. Pew also shows when Google displays a summary: for 53 percent of searches with ten or more words and for 60 percent of searches that start with a question word, but for only 8 percent of searches with one or two words.

    Seer also offers a hint about what helps. There, brands cited in the AI Overview had an organic click-through rate of 0.70 percent, brands not cited 0.52 percent. Seer itself stresses that this does not prove cause and effect. Strong brands may simply be cited more and clicked more. It is still a pointer: being in the answer beats being absent from it.

    Our own Search Console: the AI reads, the human doesn't click

    What this looks like in practice is shown by the Search Console of hierarchy.ch. On 16 November 2026 we analysed the 3,186 queries it had collected over the past 16 months.

    622 of them had eight or more words. Many do not read like a Google search but like an instruction to an assistant: «unsere seo-berichte sehen gut aus, aber die anfragen gehen zurück. wer in der schweiz misst, ob wir in ki-antworten überhaupt vorkommen?» («our SEO reports look good, but enquiries are falling. who in Switzerland measures whether we appear in AI answers at all?»). An AI assistant sends queries like this to Google when it searches on someone's behalf. 73 percent of these long queries appeared in the top 10. Together they produced 11 clicks.

    That is not a reporting error. The assistant reads the page, summarises it and passes on the answer. The human often never sees our page and so does not click. On top of that, 45 percent of our impressions come from anonymised queries we cannot see at all.

    The long queries are also a gift. They show precisely in which situation customers look for you and which criteria matter to them. Those are exactly the questions that belong in your measurement set.

    What to measure instead

    Clicks still matter. But when a growing share of research happens inside AI answers, the report needs additional metrics:

    MetricWhereWhat it shows
    Impressions by questions and provider searchesSearch Console, regex filterWhether demand and your visibility on Google are holding
    Impressions in AI featuresSearch Console, report on generative AI featuresHow often your pages appeared in AI Overviews and AI Mode
    Mentions and citations in AI answersFixed question set, asked regularly to ChatGPT, Gemini, Claude, Perplexity, AI Overviews and AI ModeWhether you appear in the answers and who is named instead
    Brand searchesSearch Console, filter on your company nameWhether more people are looking for you specifically
    Visits from AI assistantsGA4, custom channel groupWho comes to your website from ChatGPT and the like
    Enquiries and their originGA4 key events, CRM, «How did you find us?» form fieldWhether visibility turns into business

    Google introduced the report on generative AI features in June 2026, and since the end of August 2026 it has been open to all websites. It shows impressions by page, country, device and date, but no queries and no clicks. So it tells you which pages show up in AI answers, not for which questions and not whether anyone clicked.

    For mentions in ChatGPT, Gemini or Claude there is no Search Console. There you measure with a fixed set of 20 to 50 real customer questions that stays the same every month. Each question is asked several times because AI answers fluctuate, and for each answer you record: named, cited or absent. How to set this up and report it to management is covered in Measuring AI visibility.

    The «How did you find us?» form field is the simplest measurement of all. Many people type the company name from an AI answer instead of clicking. Those visits end up in GA4 as «Direct» or as a brand search on Google. Without the question on the form, they stay invisible.

    What to do now: SEO and GEO together

    The good news: the pages that rank on Google are the same pages AI Overviews and assistants read. You do not have to start over. You have to write the same pages for both.

    1. Answer questions so they get cited. Every important customer question gets its own page or its own section that answers it in the first paragraph. Specific rather than general: prices in Swiss francs with a worked example, deadlines, responsibilities, Swiss rules and standards. Headings may sound like the question. In our measurement, half of the websites cited in AI Overviews were not in the top 10 for that search. A clear answer sometimes beats the better ranking.
    2. Strengthen provider and comparison pages. For provider searches an AI Overview rarely appears, and people still click. One page per service and region, pages for comparisons and clear statements about whom you are the right choice for. This applies to B2B in particular: one page per application, technical data as text rather than in a PDF. More in AI visibility in B2B.
    3. Appear in third-party sources. AI answers do not rely only on your website. In our measurement, 28 percent of AI Overview citations went to portals, comparison sites and media. Trade media, industry directories, associations, reviews and a well-kept Google Business Profile are part of the work.
    4. Measure with a fixed question set. Take the long queries from your Search Console and the questions your sales team hears every day. Ask them the same way every month and report the trend, not the single answer.
    5. One plan, one report. When one agency optimises for the ranking and a second one optimises the same page for citability, you get two reports that know nothing about each other. A shared plan with rankings, impressions, AI mentions and enquiries in one report shows whether the work is paying off. How the two disciplines differ and where they are the same is explained in our article What GEO is and how it differs from SEO.

    What you should not do: exclude your website from AI Overviews. Since summer 2026 that is possible via a setting in Search Console. But for questions the overview sits at the very top of the page. If you are missing there, you are hardly seen for questions any more.

    How Hierarchy helps

    Hierarchy is a full-funnel agency in Zurich specialising in SEO, GEO and paid ads. We treat search engine optimisation and generative engine optimization as one job: the same pages, one plan, one report. We narrow down in your Search Console where clicks are being lost, measure with our own tracking whether ChatGPT, Gemini, Claude, Perplexity and Google name your company, and build the pages and third-party sources this requires.

    Conclusion

    Good rankings and falling clicks are no longer a contradiction. For questions, Google increasingly answers on its own, and AI assistants read your page without a human clicking. First narrow down in Search Console whether the drop comes from questions or from provider searches, and rule out seasonality, ads, brand, tracking and real ranking losses. Then measure whether you appear in the answers, not just where you rank. Whether your company is named in AI answers today is shown by the free AI visibility check.

    Sources

    Tenzin Langdun

    About the author

    Co-Founder of Hierarchy · SEO, GEO & AI Expert

    Tenzin Langdun is a co-founder of Hierarchy, where he leads SEO and GEO: how visible companies are in Google, in Google's AI answers and in assistants such as ChatGPT, Gemini and Claude. He also leads the agency's own data studies, including the Zurich AI Visibility Study of 1,515 AI answers about 5,110 companies. He brings over 10 years of marketing experience across strategy, paid acquisition and SEO, and has worked at KPMG, EY, Siemens, Adnovum and LGT, in AI, cybersecurity, audit and consulting, and in software engineering for the insurance sector and for banking and finance. He holds a BSc from ZHAW and an MSc in Artificial Intelligence from the University of Bath. His research on AI-based thyroid cancer detection was published in Scientific Reports (Nature Portfolio) in 2024 and won the Siemens Excellence Award 2024 and the ZHAW Lab Sciences Award 2024.