The Position-1 Paradox
For two decades the goal of SEO was clear: get to position 1. The top result captured the large majority of available clicks and everything below declined predictably. That model still describes plenty of queries, but it no longer describes all of them.
When Google places an AI Overview above the organic results, the answer arrives before the blue links do. Some users are satisfied there and never scroll. You still rank first, and the keyword still reports the same position, but it can deliver less traffic than it used to.
You will find no reliable universal number for how large that effect is, and you should be suspicious of anyone who quotes one. The impact varies enormously by query intent, by vertical, by whether the AI Overview cites you, and by how much detail the answer contains. The useful question is not "what is the industry average" but "what is it for my queries", and that is measurable.
What Rankings Miss
A traditional rank tracker reports your position among the blue links. It does not tell you:
- Whether an AI Overview or AI Mode answer is appearing above those links
- Whether your content is cited there, which partially offsets the visibility loss
- Whether the affected query previously converted well
- Whether a competitor is cited despite ranking below you
That last one is the most instructive. Citation and ranking are related but distinct outcomes, and a competitor can lose the SERP while winning the answer. We cover how AI Overview tracking differs from rank tracking in more detail separately.
A Better Measurement Stack
1. Search Console generative AI performance reports. Since June 2026, Search Console reports your site's visibility within generative AI features on Search, for both Search and Discover. This is first-party data from Google about your own site, which makes it the starting point rather than an afterthought. Start here before inferring anything.
2. Classic Search Console performance, read as a divergence check. If impressions hold steady or grow while clicks flatten or fall, and your average position is unchanged, something above the blue links is absorbing the difference. That pattern is a prompt to investigate a specific query, not a conclusion in itself.
3. Direct SERP observation. Check what the result page actually looks like for your priority queries. AIORadar's AI Overview Tracker records whether an AI Overview appeared, whether you were cited, and which sources were, keeping each observation tied to its market, device, and check time so that results from different contexts are not compared as though they were equivalent.
4. Branded search and AI referral traffic. Citations create brand exposure that generates no click event, so click-based analytics will understate them. Watch branded query volume in Search Console, and connect GA4 to separate out recognized referral visits from AI assistants, which otherwise sit scattered across individual source rows.
Responding to the Gap
Once you have identified queries where an AI answer is suppressing clicks, there are broadly two responses.
Win the citation. Improve the page targeting that query so it is a better candidate for the answer. Being cited recovers some exposure in the most prominent position on the page, even though it will not fully restore the click.
Shift intent targeting. Some queries are now fully satisfied in the answer box. If a query's informational intent is being absorbed, the page behind it may be better re-oriented toward commercial or comparison intent, or the effort redirected to a related query that still rewards a click.
Neither is universally correct. The decision depends on what the query is worth to you and how realistic a citation looks given who is currently being cited.
The Strategic Takeaway
Rankings remain a useful signal and are not going away. They are simply no longer sufficient on their own. Add AI presence and citation status to your reporting, reconcile position-versus-traffic divergences regularly, and be honest in client reporting about which figures are measured and which are inferred. Reporting a modelled estimate as though it were an observation is the fastest way to lose the credibility that makes the rest of the report useful.