Google AI Overviews and AI Mode
Rankings steady, but clicks are falling?
When your Google positions hold and traffic still drops, there is often an AI answer sitting above your result. Google answers the question right in the search page, and some users never click through. This page explains how to recognise it, what does not help, and what gets you into those answers yourself.
Why clicks disappear while rankings hold
AI Overviews place a summarised answer above the organic results. AI Mode goes a step further: Google runs the search as a conversation, and the classic results list fades into the background. Either way, the user gets the answer before they ever see your page.
The clicks that remain go mostly to the classic results, and hardly to the sources inside the AI answer. That is what a Pew Research Center study found, based on the browsing data of 900 US adults in March 2025: on searches with an AI summary, participants clicked a classic result in 8% of visits, on searches without one in 15% (sample: 900 people). They clicked a link inside the summary in just 1% of visits with a summary, measured in the same sample. This is not a measurement for German-speaking markets, but it shows the direction: a link in the AI answer is an extra chance of a visit, not a replacement for the lost clicks.
Informational and comparison questions are hit hardest: "What is …", "How does … work", "Which provider suits …". Searches for your brand name usually stay stable.
How to check it in Search Console
Since 31 August 2026, Google has offered every website a dedicated report on generative AI features in Search Console. Start there. Then add the click pattern from the regular performance report:
- Open the report on generative AI features. It shows how often pages from your site appeared in AI Overviews and AI Mode, broken down by page, country, device and date. It shows no clicks and no queries, and it does not separate AI Overviews from AI Mode.
- In the performance report, compare two periods of equal length, for example the last three months against the same months a year earlier. Show clicks, impressions, CTR and position.
- Sort queries by impressions. Suspicious are queries whose impressions hold or rise and whose position holds, while CTR drops clearly.
- Filter for question phrasing, for example with a regex filter on "how|what|why|which". Long queries phrased as questions trigger an AI answer especially often.
- Spot-check the flagged queries on Google yourself, ideally logged out. Is there an AI answer above your result? Is your page linked in it, and which sources appear there instead?
A made-up example for illustration: the query "how does a heat pump work" had 4,000 impressions and 200 clicks at an average position of 2.3 in last year's period. In the current period it has 4,300 impressions and 90 clicks at position 2.2. Demand is there, the position holds, but fewer than half as many people click. If the ranking page also shows up in the AI report, the connection is plausible.
The click pattern remains a well-founded suspicion, not proof. CTR also moves for other reasons, such as new ad slots, featured snippets or seasonality. The AI report, in turn, only confirms that a page appeared in AI answers, not for which query. Only the two together give a reliable picture.
What does not help
Plenty of shortcuts are being sold around AI Overviews. Google's guide to generative AI features addresses most of them itself:
- llms.txt and other special files: according to Google, you do not need new machine-readable files, AI text files, special markup or Markdown to appear in Google Search.
- Special schema markup: Google states that structured data is not required for generative AI search and that there is no special schema.org markup for it. It stays useful for rich results. Our own first analysis finds no reliable link between schema markup and AI citations, see our analysis of schema markup.
- Rewriting text for machines or chopping it into small fragments: according to Google, neither is necessary. Write for people.
- Artificially placed mentions: according to Google, seeking inauthentic mentions across the web is not as helpful as it might seem. Genuine press and PR work is a different matter; more on that in the next section.
A summary of these points is also in our methodology.
What helps: getting cited, not just ranking
AI answers in Google Search are built from a live search of the index. To appear in them, you need to be among the sources Google draws on for the answer. The fundamentals are the same as in classic search, weighted differently:
- Answer early and clearly: answer the question in the first paragraph in your own words, with concrete details instead of marketing language. A visible FAQ block on the page helps.
- Publish facts of your own: prices, terms, experience and data that exist only on your site. An AI answer can take a sentence that ten pages phrase the same way from any of them; your own number only from you.
- Show up beyond your own website: trade media, comparison portals, directories and forums are frequently cited as sources. Press work, interviews and expert articles with a real reason behind them are the way in; bought mentions are not.
- Keep brand details consistent: name, offering, location and responsibilities should match across your website, profiles and third-party pages so the brand is identified unambiguously.
- Stay technically findable: pages that are not indexed, or whose content sits behind a login or cookie wall, cannot serve as a source.
This is not limited to Google. ChatGPT, Gemini and Perplexity also search the web live for many questions and cite the sources they find. What makes you a cited source on Google usually improves your chances there too. More on this in our page on Generative Engine Optimization.
What can be measured, and what Achtung.app measures
Clicks from AI Overviews and AI Mode cannot be cleanly separated from ordinary Google clicks in your analytics; they arrive with the same referrer. The AI report in Search Console shows impressions only. How many visits an AI answer costs you can therefore be narrowed down, not counted exactly.
For a shop, a publisher or a business that takes enquiries online, lost visits remain the figure that matters. Whether your brand appears in AI answers is an additional perspective, not a replacement. A mention is a visibility signal; its business value has to be checked against visits, enquiries and revenue. On top of that, an AI answer can link your page as a source without naming your brand, or recommend you without linking.
Achtung.app answers a different question from Search Console: not how much Google traffic you lose to AI Overviews, but whether AI assistants name your brand when someone asks about your topic without a brand name. To find out, Achtung.app regularly asks ChatGPT, Gemini and Perplexity such questions and records whether your brand is named, who is named instead and which sources are cited.
Achtung.app does not measure AI Overviews or AI Mode directly, because Google offers no official interface for them. Gemini does use Google Search as a source in the measurement, but it is a separate surface with its own logic and does not replace looking at Google's results. With Search Console connected, Achtung.app also flags queries that show the click pattern described above. What share of your traffic loss comes from AI Overviews is something Achtung.app cannot quantify either.
FAQ
Often an AI answer (AI Overview) sits above the organic results for the affected queries. Users get the answer right in the search page and click through less often. Your position stays the same, your click-through rate drops. Rule out other causes such as new ad slots or seasonality all the same.
AI Overviews are a summarised AI answer above the normal results list. AI Mode is a separate search mode in which Google runs the search as a conversation and largely replaces the classic results list. Both assemble their answers from sources in the Google index and link some of them.
Only with side effects. The nosnippet meta tag removes your page's text snippet from the search results and, according to Google, also keeps the content from being used as a direct input for AI Overviews and AI Mode. The data-nosnippet attribute excludes only the marked passages; the rest of the snippet stays. The AI answer itself still appears in both cases, just without the excluded content. For most websites, nosnippet costs more than it gains.
No. Google states explicitly in its guide to its AI search features that neither special files nor AI-specific markup are required. The same fundamentals apply as in normal search: indexable pages, helpful content and authority on the topic.
By spot-checking: ask typical customer questions in ChatGPT, Gemini or Perplexity yourself. Because answers vary from one query to the next, a single sample says little. Achtung.app asks these questions regularly and measures how often and in what context your brand is named. The free scan gives you a first reading.