Google’s AI Search Console data is a metric trap

Eight weeks in, the verdict on Google's AI Overview reporting inside Search Console is starting to firm up. And it isn't the win the industry celebrated in June.

The new dashboard shows impressions inside AI Overviews and AI Mode. That's it. No click data, no query context, and — critically — a position-reporting model that treats a link buried three levels deep in an accordion the same as a top citation. Enterprise teams are building budgets around these figures. They shouldn't be.

Dan Taylor has laid the mechanics out clearly in Search Engine Journal, and once you see them, you can't unsee them. This isn't measurement. It's a comfort blanket with a Google logo on it.

The presence-without-engagement problem

Search Console now tells you when your URL appeared inside an AI Overview. It does not tell you whether anyone clicked. That distinction is the entire ballgame.

A dashboard that grows impressions while your traffic, leads, and pipeline shrink isn't a reporting tool. It's cover.

Similarweb's 2026 data, covered by Rand Fishkin last week, puts the ChatGPT external-citation click rate at 6.8% — up from around 1% a year ago, but still meaning 93 out of every 100 AI answers send nobody anywhere. Google's AI Overviews behave similarly. The impression exists. The commercial value largely does not.

A dashboard that grows impressions while your traffic, leads, and pipeline shrink isn't a reporting tool. It's cover. And it's the exact kind of cover that gets marketing directors through Q3 board meetings and into Q4 redundancy conversations.

The position-one lie

Here's the part that should genuinely concern anyone reporting to a stakeholder who takes SERP position seriously.

Google's own documentation confirms that every URL cited inside an AI Overview block gets recorded as position one in Search Console when the block sits at the top of the results. Every URL. The featured citation, the secondary reference, the link tucked inside an expandable menu the average searcher never opens — same position metric.

Which means an enterprise site that used to sit at organic rank six and now appears as a buried tertiary reference inside an AIO gets a *rank improvement* in the dashboard. From six to one. Nothing about the user's actual exposure to that link has improved. In most cases it's got worse. But the metric moved the right way, so the quarterly report writes itself.

I've watched clients over the last few weeks get genuinely excited about position gains that, on closer inspection, are entirely composed of these ghost placements. The link is technically there. Nobody is technically clicking it.

Average position was already broken. Now it's dangerous.

Average position has been a bad KPI for as long as it's existed. It's a mean applied to a distribution that isn't remotely normal. But the AIO data has taken it from unhelpful to actively misleading.

Consider the maths Taylor lays out. Your URL appears at "position one" inside an AI Overview block and at organic rank four on the same SERP. Search Console records the average as 2.5. That number looks like a page-one triumph. In practice, the organic rank-four listing is doing all the real work of driving visitors, and the AIO citation is doing effectively nothing.

The average obscures the collapse. The traditional rank-four position might be losing clicks month over month as the AIO absorbs more of the SERP real estate above it. The dashboard shows a stable 2.5. The business shows a bleeding funnel. These two things are supposed to be connected. They no longer are.

What the honest measurement stack looks like now

The GSC data isn't useless. It's just not what it looks like. Treated as a *presence indicator* — did we appear in this AI surface at all — it has some diagnostic value. Treated as a *performance metric*, it's a trap.

The measurement stack that actually reflects reality in mid-2026 looks something like this. First-party analytics as the source of truth for traffic, leads, and revenue by channel — not GSC clicks, which are increasingly a subset of a subset. Median position, not average, if you insist on tracking rank at all. Brand-search volume and direct traffic as leading indicators of AI visibility, because AI systems overwhelmingly cite brands they've heard of. Server-log analysis for AI crawler activity, which tells you which systems are actually training on and referencing your content. And a genuine acceptance that citation-share inside AI answers is currently unmeasurable at scale without paid tooling that most SMBs won't buy.

That last point matters. Google reports its own revenue to the penny and reports web traffic in adjectives. The GSC AI dashboard is a continuation of that pattern, not a break from it. It's the appearance of measurement without the substance.

What this actually means

If you're an in-house marketer and your last board slide leaned on GSC impression growth from the AI reports — quietly rebuild that slide before your next one. Impressions inside AIOs are not clicks, are not leads, are not revenue, and are increasingly not even a proxy for any of those things. Anyone who tells you otherwise is either selling something or hasn't done the reading.

If you're an agency reporting to clients, the honest thing to do is add a disclaimer to any AI-search reporting slide that names the three limits: no click data, distorted position logic, and average-based aggregation. Clients don't punish you for honesty about measurement gaps. They punish you when the honesty arrives six months late and the pipeline has already collapsed.

And if you're a business owner watching your marketing team celebrate improved GSC numbers while your leads flatten — trust the pipeline, not the dashboard. The pipeline has always been the truth. It just used to correlate more neatly with what Google was willing to show you.

The measurement layer marketing runs on is quietly broken. The GSC AI reports aren't a fix. They're a very well-designed distraction from the fact that the fix hasn't arrived yet.