If your average position in Google Search Console suddenly improved while clicks flatlined or dropped, don't celebrate. Google now counts appearances inside AI Overviews as regular impressions and positions, and that quietly pulls your rankings up while the traffic those rankings used to send you disappears.
The short version: GSC's default view blends AI-surface data with classic blue-link data, so a page that lost real clicks can look like it climbed to position one. To see what's actually happening, you have to split AI appearances out from everything else. Here's how to do that without lying to yourself.
This matters right now because most accounts only have a couple of months of blended data. That's just enough to form a misleading trend and not enough to notice the traffic quietly leaking away underneath it.
What actually changed in your Search Console data
Google folded AI Overviews and AI Mode into the same Performance report you already use, with no separate toggle by default and no way to fully exclude them. Your impressions, average position and CTR now mix two very different things: links people clicked and citations that sat inside an AI answer nobody had to click.
The practical effect is a distortion in three of the four headline metrics:
- Impressions go up. Being cited in an AI Overview counts as an impression even when your link is collapsed behind a "show more" panel.
- Average position looks better. AI Overview citations often register at or near position one, dragging your blended average up.
- CTR falls. More impressions with fewer clicks mechanically lowers your click-through rate, and it's easy to blame yourself for a drop that's really a measurement artifact.
- Clicks stay honest. This is the one number that didn't change meaning. A click is still a real person landing on your site.
So when someone tells you rankings are up, ask which number they're looking at. If it's average position, that's now the least trustworthy metric in the report.
The specific numbers that will fool you
Average position and CTR are the two metrics most likely to mislead you, because both are now averages across two behaviours that shouldn't be averaged together. Treat any big swing in either as a question, not an answer.
Picture a page that ranked #4 for a buying-intent query and pulled 200 clicks a month. Google starts showing an AI Overview for that query and cites your page inside it. Now you also register 5,000 AI impressions at position #1. Your blended average position jumps from 4 to maybe 1.6. Your clicks? They drop to 120 because the AI answer resolved the question on the results page.
Every default chart in that account tells a success story. Rankings up, impressions up. The only honest signal is the click count sliding down, and it's buried in an aggregate that looks great.
The trap is emotional as much as technical. Improving numbers make you stop investigating. That's exactly when a real decline gets a head start.
How to segment AI impressions out of your reports
Segment your data by applying a Search Appearance filter for AI Overviews, then compare that slice against everything else so the two behaviours never sit in the same average again. This is the single most useful thing you can do in GSC this month.
Here's the exact workflow:
- Open Performance → Search results.
- Click + New and choose Search Appearance.
- Select AI Overviews (and AI Mode if it appears). This isolates only AI-surface data.
- Note the clicks, impressions and position for that slice.
- Now flip your thinking: your "real" web-link performance is roughly total minus this AI slice. GSC won't hand you a clean inverse, so track the AI slice as its own trend and watch the totals separately.
Then rebuild your monitoring around clicks and query-level detail, not average position. Add the Queries and Pages tabs, sort by clicks, and compare a rolling 28-day window against the previous 28 days. If clicks fall while impressions climb, you've found an AI-cannibalised query. Those are the ones to study.
For anything commercial, pull the data into a spreadsheet or Looker Studio monthly. GSC's built-in comparison only goes so far, and export lets you keep a real history before Google's 16-month window rolls old data off.
How to tell AI cannibalisation from a real ranking drop
The two look identical in a blended chart, so you separate them by checking whether impressions rose or fell alongside the click loss. Rising impressions plus falling clicks points to AI eating your traffic; falling impressions plus falling clicks means you genuinely lost visibility.
Run each declining page through this quick diagnostic:
| What you see | Most likely cause | What to do |
|---|---|---|
| Clicks down, impressions up, position "improved" | AI Overview cannibalisation | Rework the page for questions AI can't fully answer; add depth, tools, opinion |
| Clicks down, impressions down, position down | Real ranking loss | Check for lost links, thin content, a competitor move, or a core update |
| Clicks flat, impressions up sharply | You're being cited but not clicked | Strengthen title and meta to earn the click over the AI summary |
| Clicks up, everything up | Genuine growth | Do more of whatever you just did |
Take a position on this: if a query is now dominated by an AI Overview that answers it completely, chasing that ranking harder is wasted effort. Redirect that energy toward queries with buying intent, comparisons, and content that needs your actual expertise, judgement or first-hand testing. Those don't summarise cleanly, so people still click through.
A monthly check that keeps you honest
Build a five-minute monthly routine that leads with clicks and treats every other metric as context, so no single flattering number can hijack your read of the data. Consistency beats depth here; the point is catching a trend early, not producing a report.
Run this checklist on the first working day of each month:
- Clicks first. Compare the last full month to the previous one. This is your ground truth.
- Segment AI Overviews using the Search Appearance filter and log that slice separately.
- Sort queries by biggest click loss. Ignore position entirely at this step.
- Cross-check impressions on each dropping query to classify it using the table above.
- Export the raw data so you own a history longer than 16 months.
None of this requires a paid tool. It does require the discipline to distrust a number that makes you feel good. If clicks are your north star and everything else is context, the AI data stops being a trap and starts being useful, because a growing AI slice tells you which topics Google is answering for you.
One infrastructure note that's easy to overlook: when AI crawlers and citation checks hammer your pages, slow or flaky hosting quietly costs you those citations. If your site is on shared hosting that buckles under crawl spikes, the real engineers on our support team at TPC Hosting are on hand 24/7, and moving over is free — we handle the migration so you don't lose data or downtime while you sort out your analytics.
Where this is heading
Expect Google to keep blending AI surfaces into your metrics, so the marketers who win are the ones who already measure clicks and outcomes instead of positions. Rankings as a headline number were always a proxy; AI Overviews just broke the proxy in a visible way.
Practically, shift the questions you ask of your data. Instead of "did we rank higher," ask "did more of the right people reach a page and do something useful." Track assisted conversions, email signups, calls and sales — the things GSC never measured anyway. Those numbers don't lie when an AI answer sits between you and the reader.
The accounts that get burned over the next year are the ones celebrating position-one averages while their real traffic thins out underneath. Segment your reports, lead with clicks, and you won't be one of them.
FAQ
Can I turn off AI Overview data in Search Console?
No, Google doesn't let you exclude AI Overview data from the Performance report entirely. The best you can do is isolate it with the Search Appearance filter for AI Overviews and analyse that slice separately from your other search performance.
Why did my average position improve but my traffic drop?
Because AI Overview citations often register near position one and get blended into your average, pulling it up even as real clicks fall. Check whether impressions rose while clicks dropped — that pattern means an AI answer is resolving the query without sending you the visit.
Which Search Console metric can I still trust?
Clicks remain the most honest metric, because a click still means a real person reached your site. Average position and CTR are now distorted by AI-surface data and should be treated as context, not headline numbers.
How often should I check for AI cannibalisation?
Once a month is enough for most small sites. Compare the last full month's clicks to the previous month, segment the AI Overviews slice, and flag any query where impressions climbed while clicks fell.
Should I stop trying to rank for queries dominated by AI Overviews?
If an AI Overview fully answers a query, chasing that ranking harder rarely pays off. Redirect that effort toward buying-intent, comparison and expertise-driven content that AI can't summarise cleanly, since those still earn real clicks.
