There is a pattern showing up in GSC accounts right now that looks like an SEO problem but isn’t one.
Rankings have improved. Average position is up. Clicks have dropped — sometimes significantly. The natural conclusion is that something went wrong. It didn’t. What happened is that Google inserted an AI Overview above your organic result, answered the query before the user ever reached the link list, and kept the click.
Your SEO is working. The SERP changed around it.
This post walks you through how to find exactly which queries this is happening on — step by step, the same way I work through it with client accounts — and what to do about it once you’ve found them. I’ll use real data from Coviant Software’s GSC account to show what each step reveals in practice.
What you’ll learn:
– How to detect AI Overview displacement in GSC using a three-filter method
– How to isolate affected queries using regex
– How to read the data and why query-level detail changes what you do next
– Content, technical, and entity fixes in the right sequence
– How to measure success differently for the queries you can’t fully recover
Step 1: Enable Compare Mode in GSC
Open Google Search Console → Performance → Search Results. Click the date range selector, select the Compare tab, choose Compare to same period last year, and hit Apply. The column headers update to show pairs — ‘Clicks Last 3 months’ alongside ‘Clicks Difference.’ The chart shows two lines: current period and comparison.
What to look for: Impressions holding while clicks fall. That divergence is the first signal. Both falling = rankings problem. Impressions flat, clicks falling = AI Overview problem. In Coviant’s account: 375K impressions (vs 388K last year — flat) but CTR dropped from 0.5% to 0.3%.
Step 2: Apply the Three Filters
Filter 1: Clicks Difference — Smaller than — 0. Gives you everything where clicks fell year-on-year.
Filter 2: CTR — Smaller than — 0.1. Removes branded and navigational queries where other factors dominate.
Filter 3: Position Difference — Smaller than — 0. The key filter. Negative position difference = position improved. You’re looking for queries where position got better and clicks got worse simultaneously.
All three together show pages where you ranked higher than last year, were seen by roughly the same number of people, but received fewer clicks. What I found in Coviant’s account: coviantsoftware.com/technology-briefs/filezilla-alternatives/ — 102 clicks vs 199 last year (-97 clicks), impressions flat at 30,703 vs 30,716, position improved from ~30 to 6.7.
| Metric | Last year | Last 3 months | Difference |
|---|---|---|---|
| Impressions | 30,716 | 30,703 | -13 (flat) |
| Position | ~30 | 6.7 | +23 positions |
| Clicks | 199 | 102 | -97 clicks |
| CTR | 0.6% | 0.3% | -0.3% |
Rankings up. Impressions flat. Clicks nearly halved. One primary cause.
Step 3: Refine to Informational Queries Using Regex
The 1,000 row limit in GSC means you’re not seeing everything. Filter to informational queries — the ones most likely to trigger AI Overviews — using a regex pattern that captures question words and research intent modifiers. Click Add filter → Query → Custom (regex) → Matches regex and paste this in:
^(how|what|why|when|where|who|which|can|could|should|would|does|do|is|are|was|were|will|has|have|had)[s+]|.*(best|top|vs|versus|compare|comparison|review|guide|tutorial|tips|alternatives|checklist|explained|definition|types of|list of|how to|what is|what are|why do|should i|can i|how much|how many|how long).*This surfaces question-format and research-intent queries where Google is most aggressively answering before the click. For larger sites this filter is essential — it separates AI Overview displacement from the noise.
Step 4: Sort by Impressions Descending
Sort by Impressions Last 3 months in descending order. A query with 50,000 impressions and AI Overview displacement is a much higher priority fix than one with 500 impressions. The sort order turns a filtered list into a prioritised work list.
Step 5: Manually Verify in Google
Before drilling into query-level data, open an incognito window and search for the flagging queries. The GSC pattern (position up, clicks down, impressions flat) can also be caused by featured snippet expansion, shopping blocks, video carousels, or People Also Ask expansion. If no AI Overview is present, the fix is different — rich result schema rather than AI citation restructuring.
If an AI Overview is present, note whether your brand is cited inside it. Present = brand signal working, click not guaranteed. Absent = citation gap, content restructuring is the priority fix.
Quick diagnostic — AI Overviews are costing you clicks when:
✅ Impressions stable year-on-year
✅ Average position improved
✅ CTR dropped
✅ Informational query pattern (how, what, alternatives, comparison)
✅ AI Overview visible in manual SERP check
All five = AI Overview displacement confirmed. Fewer than four = investigate other SERP features first.
Step 6: Drill Into a Page to See Query-Level Detail
Click the page in the Pages tab to filter to just that page, then switch to the Queries tab. Page-level data tells you that displacement is happening. Query-level data tells you which queries and reveals distinctions that change what you do next.
What I found in Coviant’s filezilla-alternatives page: The top specific comparison queries — ‘filezilla alternative’, ‘filezilla alternative for windows’ — were gaining clicks year-on-year. Losses were on broader generic terms: ‘filezilla’ itself (-10 clicks), ‘sftp alternatives’. Google’s AI Overview answers a generic alternatives list without a specialist source. Coviant’s page is specific enough to win the named-product comparison but loses the generic query.
The insight: Specificity is a partial defence against AI Overview displacement. Broad comparison queries lose clicks. Specific, named-product comparison queries hold them. Not visible from page-level data alone.
What to Do With the Queries You Find
The filtered list is not a problem list — it is an opportunity list. But be clear: you cannot force Google to remove an AI Overview or guarantee citation inclusion. What you can do falls into three categories.
Category 1: Get cited inside the AI Overview (on-site content)
Restructure the page so the answer appears in a standalone, extractable paragraph within the first 120 words of the relevant section. Explicit definitions. Attributed statistics with named sources. FAQ pairs where each answer is complete without context. The AI Citation Checklist gives you the six criteria with before-and-after examples. The AI Page Anatomy shows exactly how a page should be structured.
Cannibalisation warning: Before creating new specific pages (‘filezilla alternatives for healthcare IT’), ask whether an existing page already ranks for that intent. If yes, add an H2 section rather than a new URL. Restructure existing pages first, enrich with vertical H2 sections second, create new URLs only if the vertical intent is genuinely distinct and can’t be served by the existing page.
Category 2: Win the clicks that still exist (technical)
Rewrite meta descriptions for post-AI-Overview intent — ‘See the full comparison with scoring criteria’, ‘Download the complete checklist’ — signalling that clicking through gives something the AI Overview didn’t. Check GPTBot and OAI-SearchBot are not blocked in robots.txt. Check page speed — AI crawlers time out faster than human users. Ensure structured data is server-rendered. Full checklist at Technical SEO.
Category 3: Build entity authority so AI systems name you by name (off-site)
There is a difference between your content being extracted anonymously (‘according to sources…’) and being named by brand (‘according to Coviant Software…’). That difference is entity authority. AI systems only name brands they have sufficient cross-platform confidence in.
Wikidata — the open knowledge graph feeding LLM training data. A verified entry with correct properties gives AI systems a confirmed entity anchor. See Wikidata for SEO.
Entity schema — Organisation schema with knowsAbout and sameAs links to every verified profile. See Entity SEO.
NAP consistency — business name, address, and phone number must match exactly across all platforms. Inconsistencies reduce AI entity confidence.
Third-party citations — mentions in authoritative industry publications, sector directories, and press. AI systems weight external corroboration heavily. See Digital PR & Link Building.
The full diagnostic for why your brand may not be appearing in AI answers by name is at Why Isn’t My Website Appearing in AI Answers?
Measure Differently for the Queries You Can’t Recover
For pure informational queries where Google has moved to AI Overview answers, the click may be permanently reduced. Stop measuring success by click volume on those queries and track downstream brand search volume instead. A user who sees your brand cited in an AI Overview may not click that day but is more likely to search for you directly on their next higher-intent query. Rising brand searches while informational clicks fall is the signal that your AI Overview citations are working.
If you want to work through your own GSC data and build a prioritised action plan, the AI Visibility Audit is where that starts.
