Search Console Now Shows AI Answers – but No Clicks

For two years the most important question in SEO was not measurable: does my page appear in the AI answers? You could sample it manually, you could infer things from traffic drops, and you could read studies that reached different conclusions depending on method. Since June 2026 there is a report for it in Search Console: impressions in AI Overviews and AI Mode, with pages, countries, devices and a time series. What is missing says as much as what is there – the report does not output clicks. That gap determines how the data may be read.

Key takeaways

• New since June 2026: a dedicated report for impressions in the generative AI features of Google Search; Discover has a separate one, Search Labs experiments are excluded.
• Impressions only, no clicks – and no queries. An AI click-through rate therefore cannot be calculated.
• Dimensions: pages (grouped by canonical), countries, devices, date.
• Mind the counting: if two results from the same property appear in one AI answer, that counts as one impression.
• Do not blend with organic impressions: same name, different user experience.
• Not everywhere yet: staged rollout, and the property needs sufficient impressions.
• Valid use: a diagnostic tool for how pages are built, not a success metric for reporting.

What the report can do

The report answers exactly one question, and it answers it reliably: which of my pages appear in AI answers? That used to be the biggest blind spot. You knew answer surfaces pull clicks away, but not whether your own page featured in them at all – whether you were passed over or cited.

The counting method deserves attention because it shapes the result. Impressions are aggregated at property level: if two results from the same website appear in one AI answer, that is one impression, not two. For assessing individual pages this means the sum across pages need not match the property total. On top of that come the usual Search Console limits: the most recent days are preliminary, tables stop at a thousand rows.

What it cannot do – and why that matters

There is no click data. That is not a detail, it is the heart of the matter. Without clicks you can neither form a click-through rate for AI answers nor quantify the value of a citation. You see visibility, not effect.

So the obvious mistake is also the most dangerous one: adding AI impressions to organic ones and computing a combined click-through rate. Both metrics are called impressions and mean different things. In the results list a hit sits at a position, with a title that can be clicked. In an AI answer the same link may be one of eight citations, behind a disclosure element, far below the answer itself. Summing the two builds a metric that exists in no reality – and budget decisions then rest on it.

The report is equally unsuited as a target. A curve that shows only visibility invites you to optimise visibility – even where it earns nothing. That is the same thinking error that put rankings above revenue for years, just on a new axis.

How to read it properly

The report becomes useful as soon as you place it beside the organic data rather than inside it. A workflow that holds up:

1. Export the AI-visible pages over a period long enough to smooth weekly rhythms – not single days.
2. Put the organic figures for the same URLs next to them: impressions, clicks, position.
3. Group by construction rather than topic: explainers with definitions, comparison tables, step-by-step guides, overview pages. The question is not which topic gets cited but which form does.
4. Look at outliers in both directions. High AI visibility at a mediocre position: those pages supply answer material – good pattern, do more of it. Good position, barely any AI visibility: the substance is probably buried in prose a model cannot cleanly take apart.
5. Follow trends, not daily values – and look again after every major Google update.

What follows from that pattern is not a new discipline. It is the same set of things that carry LLM optimisation for stores and the pillar on the state of the art in SEO: unambiguous statements instead of hints, clean structure, evidenced numbers, machine-readable markup. To check the technical side, our Technical SEO Check walks the fundamentals, and how strongly load time affects both is calculated in page performance.

The numbers on the situation – handle with care

On the question of how much traffic answer surfaces actually cost, there are now many surveys, and they contradict each other. The range runs from moderate declines to shares beyond half of all searches ending without a click to any website – depending on whether brand and navigational searches are counted, which countries are considered and whether mobile is included. We deliberately do not present a single percentage as truth here: those very numbers are the reason your own data is becoming more valuable. Your Search Console report is imprecise, but it is imprecise about your pages – which is more than any external study can offer.

Bottom line

The report closes the biggest measurement gap of the last two years and opens a smaller one in the process: we now know where we appear, but not what it earns. That makes it a diagnostic instrument and not a success metric – and treating it that way yields something concrete: evidence about which form of content Google finds usable for answers. Anything beyond that would be precision the data does not support.