Over the years you have built up rankings that hold. And now ChatGPT answers your customer’s question before they even reach you. On top of that, in Google’s results: where an AI summary sits above the results, around 58% fewer people click on the top result, and depending on the measurement, such summaries now appear in somewhere between one in six and one in four searches.
Sounds like bad news? Only as long as you are groping in the dark. Because the good news is: AI visibility can be measured. That is exactly what this article is about. You will get a clear idea of what ‘Generative Engine Optimization’ actually is, what is really changing right now, and which built-in features and tools you can use to put your visibility in ChatGPT, Perplexity and the like into figures from this week on.
Generative Engine Optimization explained briefly, without the buzzword fog
Generative Engine Optimization, or GEO for short, is simple at its core: you no longer optimise your content only to rank at the top of a list of links, but so that an AI mentions and cites your brand in its answer.
By the way, the term is not marketing speak from an agency. It comes from a research paper by Princeton, Georgia Tech and the Allen Institute, which coined it at the end of 2023 and presented it at the KDD conference in 2024. The same paper shows that visibility in AI answers can be increased by up to 40% with targeted GEO measures.
The decisive difference from classic search engine optimisation: SEO thinks in rankings and clicks, GEO thinks in citations and brand mentions. So it is about whether and how an AI builds you into its answer at all. Some also speak of ‘GEO SEO’, ‘AI Overview SEO’, ‘ChatGPT SEO’ or simply of AI visibility. What they always mean is the same: making sure you appear in the generated answer, and tracking whether that works.
How you become worth citing in the first place, from crawler access to structured content, is broken down in Can AI even find your website?. Here we deal with the other half: measuring whether it works.
Why the topic is so pressing right now
Let us take a quick look at the figures, because they explain why so many marketing teams are nervously watching their traffic right now. Google AI Overviews, the AI summaries right at the top of the search results, appeared in around 6.5% of all search queries at the start of 2025. By the spring the figure was already above 13%, in the summer it climbed towards 25% at times, and by the end of the year it settled at around 15%.
In 2026 the curve is rising further: Conductor measured around 25% across more than 21.9 million search queries in the first quarter of 2026, and BrightEdge even sees 48% in commercial sectors in March 2026. The trend is clear: AI Overviews have gone from being the exception to being the norm.
And where such a summary appears, considerably fewer people click through. At the end of 2025, Ahrefs measured that the click-through rate of the top result is around 58% lower when an AI Overview is present. A study by Seer Interactive even puts the drop in organic click-through rate for such search queries at around 61%. The values vary by study and sector, but the direction is the same everywhere.
At the same time, a second source is growing: traffic that comes directly from AI assistants. It is still small, measured against the web as a whole it is only in the low per-mille range, but it is growing at triple-digit rates. And it is high quality. Several analyses show that visitors from ChatGPT and Perplexity convert noticeably better than anonymous organic traffic, because they arrive with a concrete recommendation in mind. That is exactly why it is worth looking at, and doing so now.
Incidentally, there is also some really good news in the data. The same Seer analysis found around 35% higher organic click-through rates for pages that are cited within an AI Overview. Visibility in the AI answer is therefore not a consolation prize but a real lever.
How to make AI visibility measurable in four layers
Now for the practical part. ‘Measuring AI visibility’ sounds like rocket science, but it breaks down into four manageable layers. You do not have to set up all four at once. Start at the top and work your way down.
Layer 1: make AI traffic visible in your analytics
This is the quickest lever, because you do not need a new tool for it, only your existing analytics. By default, a click from ChatGPT or Perplexity ends up either in the ‘Referral’ bucket or, worse, under ‘Direct’, because the AI platforms often strip the referrer. Without adjustments, you therefore do not see your AI traffic for what it is.
Since May 2026, Google Analytics 4 has automatically recognised part of this traffic via a dedicated channel for AI assistants, but only the part with an intact referrer. Depending on the platform, a considerable share of AI visits arrives without this information and seeps away into ‘Direct’. The solution is a custom channel group that sorts these sources properly.
This gives you a figure from day one that was invisible before: how many people land on your site via AI assistants, on which pages, and how well they convert. That is your baseline.
Layer 2: check whether and how AI names your brand
Analytics shows you the people who have already clicked. The more interesting question is often the one before: does your brand appear in the answers at all, even if nobody clicks? This is where the concept of ‘share of voice’ comes into play. How often does an AI name you for the questions that matter to your business, compared with the competition?
You start manually, without any tool at all. Write down 15 to 20 questions that a customer would ask before buying from you, and put them to ChatGPT, Perplexity and Gemini. For each answer, note: are you named? Linked? In what context?
That is laborious and only a snapshot, but within an hour it gives you an honest picture of your starting position. Before you start, check your AI visibility at the technical level: our free check shows in 30 seconds whether ChatGPT, Claude and Perplexity can read and cite your website at all.
Layer 3: track citations and sources
An AI answer is built from sources. For GEO it is therefore decisive which pages the AI draws its statements from and whether you are one of them. Worth knowing: where AI answers disclose their sources openly, large platforms such as Wikipedia and Reddit dominate.
That is a strategic pointer that many underestimate. Your visibility depends not only on your own website, but also on whether you appear in the places the AI trusts, for example in a well-maintained Wikipedia entry or in strong specialist industry forums.
Layer 4: monitor AI Overview presence in Google Search
The fourth layer is the bridge between classic SEO and GEO. For your most important search terms, you want to know whether an AI Overview appears there and whether you are cited in it. Part of this is now answered by Search Console itself: its AI report counts how often your pages appear in AI Overviews and AI Mode, but without the search terms behind them.
You therefore still read the citation status per search term from the answers themselves or from specialised tools, which we come to further down. Keep one peculiarity in mind: AI answers are volatile. Analyses show that, depending on the platform, 40 to 60% of the cited domains change from month to month. A single measurement is therefore a snapshot; the trend over weeks is what counts. For this trend, a look at the AI report is worthwhile.
The AI report in Search Console: measuring AI Overviews and AI Mode
Google has extended Search Console with two things: a dedicated figure for how often your pages appear in AI Overviews and AI Mode, and a setting that lets you opt out of exactly these answers. You should use the figure from now on. With the setting, it is better to take your time.
Since 31 August 2026, the generative AI performance reports have been available for all websites worldwide, including your German domain. Google had announced them on 3 June 2026, at the same time as new requirements from the UK competition authority, the CMA, and has been rolling them out gradually to more and more websites since then.
AI impressions continue to be included in your normal performance report. The new report also shows them separately, in two variants: one for Google Search with AI Overviews and AI Mode, and one for the AI features in Google Discover. You will find both among your property’s performance reports.
What the AI report shows and what it does not
The only metric is impressions, i.e. how often a link to your website was shown in an AI feature. You can break them down by page, country and date, and in the Search report additionally by device.
What is missing is unfortunately the most interesting part: clicks, click-through rate and the search queries for which you appeared. So you can see that a page appears in AI answers, but not for which question and whether anyone comes to you afterwards.
If you cannot see the report at all yet, nothing is broken on your side: it only appears once your website collects enough AI impressions, and according to Google not all properties have access even now. Also check quickly whether your website is excluded via the AI setting described further down. Incidentally, it does not count experiments from Search Labs.
How to use the AI report for your GEO measurement
The report gives you an official figure for layer 4. Two analyses are worthwhile straight away:
- Find your AI source pages: Sort the Search report by pages. The pages with the most AI impressions are the ones Google links to most often in its AI answers. You should keep these particularly up to date and cleanly structured.
- Spot visibility without visits: Put your top pages next to the normal performance report. If a page’s AI impressions rise while its clicks fall, that may be a sign that users are already getting their answer in the summary. That is the typical pattern of a zero-click search.
Ideally, export the AI report once a month, so that a time series builds up which shows real trends after a few months. Note: values that appear in the report as ‘~’ or ‘-’ end up as zero in the export.
The report does not replace the other layers. Which questions might be behind the impressions is something you continue to narrow down with your prompt list from layer 2, by also asking the questions directly in Google Search and in AI Mode. And GA4 counts visits from AI Overviews and AI Mode under ‘Organic Search’, not in the channel for AI assistants and not in your AI channel group from layer 1 either. These clicks cannot currently be measured separately with built-in tools.
The AI setting: measure first, then decide
Together with the report, Google has introduced a setting: in Search Console you decide whether links and content from your website may appear in AI Overviews, AI Mode and the AI features of Discover and serve as a basis for AI answers. By default, this is allowed.
If you switch it off, according to Google’s help page on the Search generative AI control this usually takes effect after a few days. For the normal search results, the setting is not a ranking signal. It has nothing to do with the training of Google’s AI models; you continue to control that via Google-Extended in your robots.txt.
If you want to become more visible with GEO, leave the setting on, because by switching it off you disappear from exactly the answers you want to be in. The case is different if your business depends on people reading your content on your own site, for example a specialist publisher. Then switching it off is a legitimate option.
Make the decision on the basis of your own figures: for a few months, put the AI impressions of your most important pages next to their clicks in the normal performance report. If the zero-click pattern described above shows up there, you have a basis for the decision.
Which tools take the measuring off your hands
Apart from the AI report, you have so far measured a lot by hand. That is good for understanding the logic. In the long run you will want to automate it, and in 2026 a whole toolbox has emerged around GEO for that purpose.
Three names come up again and again in practice. Peec AI is a good starting point if you mainly want to visualise visibility trends and your share of voice over time. Profound is the enterprise heavyweight for larger teams with a budget and analyses behaviour across ten or more AI engines. AthenaHQ plays in the same field.
At their core, they all answer the same questions: how often are you named, in what tone, compared with whom, and where is it heading? The technical preliminary stage (whether AI crawlers are allowed to fetch your pages and whether your markup is correct) is covered by the free AI visibility check, without any tool subscription.
To be honest, the market is still sorting itself out, and which tool is the right one for you depends on size and budget. For most SMEs, the more important decision is not the tool anyway, but that you start measuring regularly at all, even if only with a simple spreadsheet and a weekly prompt run at first. The tool comes once the routine is in place.
This very routine of measuring, finding gaps and sharpening content is also at the heart of our work in digital marketing: AI visibility is not a one-off project but an ongoing process. And if you want to hand over measurement and implementation, our GEO agency for AI visibility will take care of it.
Where to start this week
Theory is all very well, but you want to make progress. If you only do three things this week, make it these: set up the AI channel group in GA4 so that your AI traffic becomes visible straight away. Write down your 15 to 20 most important customer questions and run them once through ChatGPT, Perplexity and Gemini; that is your baseline. And for each answer, note who is named instead, because that is your competitive map for the coming months.
If you still have two minutes after that: check in Search Console whether your property already shows the AI report, and export the current state as a starting point.
Generative Engine Optimization is not a new religion that replaces the old SEO. It is an extension of your visibility to the places where your customers now ask their questions. The reflex to see this as a threat is understandable, but the real risk is not AI search, it is flying blind.
So start small: open GA4 and create the channel group, which takes 15 minutes. After that, ‘the AI is doing something with our traffic’ is no longer a vague worry but a figure you can work on.