AI Search Visibility: Measuring AI Mode & Overview Impressions

AI search visibility is the metric everyone in this cycle reports and almost nobody defines. The phrase usually blends two different things: how often Google’s AI surfaces appear for the queries you care about, and how often they point at you. Neither arrives as a figure you can open a dashboard and read. Impressions from AI Mode and AI Overviews are real, and they are counted, but they sit inside the same Search totals as everything else. Measurement here is something you build rather than something you retrieve. What follows is how to build it so that the numbers still mean something in three months.

What Search Console will and will not give you

Start with the constraint, because it shapes everything after it. Google’s documentation on AI features in Search states that impressions and clicks from those features are included in the Performance report. What it does not offer is a control that isolates them: no AI surface type to filter by, and no subtraction that recovers the AI portion of a total.

So a rise in impressions is compatible with several stories at once: you started appearing in an AI Overview, you moved up in ordinary results, a competitor dropped out, or the query got more popular. Search Console is telling the truth. It is telling you Google showed your page somewhere, and declining to say where.

That limitation is why this measurement needs a second instrument alongside the platform data, and why a vendor reporting a clean AI impression count is reporting an estimate, whatever the interface implies.

What an AI search visibility number is actually made of

A visibility figure is a ratio, and a ratio needs a denominator. Most reporting skips straight to the numerator: we were cited eleven times this month. Eleven out of what?

There are two candidate denominators and you want both. The first is the query panel: the fixed set of searches you check on a schedule. The second is the eligible set: the pages that could be cited at all. Eligibility is not a matter of quality. AI Overviews and AI Mode draw from the same index that serves ordinary results, built by the same process Google sets out in how Search works. A page outside that index has a citation probability of zero, and averaging it into your visibility figure quietly drags the number down for a reason that has nothing to do with content. The dependency is worth reading in full in why AI surfaces only cite indexed pages.

Establishing the eligible set is the one part of this that can be settled with facts rather than inference. Google’s page indexing report gives you a site-level view. For a named list of URLs, AIndexify checks each submitted URL against live Google results on a fixed schedule (24 hours, 48 hours, 96 hours and 7 days on Fast; 1, 4, 12, 24 and 72 hours plus 7 days on Instant) and returns a dated status: Indexed with the date it was found, Not indexed as of a given date, or Couldn’t check when a noindex, 404 or robots rule is detected before any check is billed. No Search Console connection is involved, which matters when the pages belong to a client whose property you do not own.

The panel, and why it has to stay still

Pick twenty to thirty queries your pages genuinely answer. Write them down. Then leave them alone for at least a quarter.

The temptation every month is to swap in fresher queries, and it destroys the series: you cannot tell whether the number moved because your visibility changed or because the panel did. A stale panel that stays fixed is more informative than a perfect panel that is rebuilt each time. Add new queries as a second cohort with its own start date rather than editing the first.

For each run, record five things per query: the date, whether an AI surface appeared at all, whether your domain was cited, which URL was cited, and the index status of that URL at its last check. The practical discipline for running the checks fairly (clean sessions, location, repeat sampling) is set out in how to check whether you appear in AI Overviews and AI Mode.

Four numbers worth reporting each month

• Trigger rate. The share of panel queries that produced an AI surface at all. This moves for reasons entirely outside your control, and tracking it separately stops you reading Google’s product decisions as your own performance.
• Citation share. Of the queries that did trigger a surface, the share where you were cited. This is the closest thing to a visibility rate, and it is only comparable across months if the panel held still.
• Eligible set. How many of the pages behind the panel are indexed, with the date each was last confirmed. A citation share calculated over pages you never verified is a guess wearing a percentage sign.
• Search Console shape. Impressions and clicks for the same page set, read as context. Not as attribution.

Four numbers, one page, one cadence. Monthly is usually right; weekly produces noise you will be tempted to explain.

Reading a divergence without overreading it

The pattern people watch for is impressions holding steady while clicks decline, which is consistent with a surface answering the question before anyone needs to click. It is also consistent with a ranking slip, a new competitor snippet, a seasonal shift in intent, and three other things. A divergence in the platform data is a question, not a finding. Promote it to a finding only when your panel log agrees: the same queries, in the same weeks, showing citations that were not there before.

This is also the honest answer to the AI visibility scores now being sold. A score assembled from a sample of queries is a sample, not a measurement of the whole, and a citation is a judgement made inside Google by systems no vendor operates or inspects. The trustworthy version of that product tells you its panel size, its query list and its dates. Any promise attached to the citation itself is a promise about someone else’s decision.

Measure the link you can actually verify

Nearly every step in this chain is inference: whether a surface fired, whether it fired for a searcher in another city, whether the citation was durable. One link is not. Whether a URL was in Google’s index on a given date is a checkable fact, and it is the floor under every other number on the page. The full sequence from crawl to index to verification is covered in the guide to Google indexing.

If the pages behind your query panel have never been checked, that is the cheapest part of the measurement to fix. The credit packs are one-time purchases starting at $19 for 1,000 URLs, and the credits never expire.