The exact Google Search Console regex to isolate AI-shaped queries is ^(what|how|why|when|which|is|are|does|can|should)\b.* — apply it as a custom Query filter with the Matches regex option. This isolates question-shaped queries — the ones AI Overviews and AI Mode preferentially surface — from your general query stream. WebFlur runs this filter on every B2B GSC we audit; it is the fastest AI-search signal you can extract from free tools.
GSC does not have a native "AI query" filter, and it never will — Google does not label queries that way. But question-shaped queries have an 80%+ overlap with the queries AI Overviews and AI Mode actually surface, and question-shape is trivially detectable with regex. Below is the exact setup, 4 additional patterns for narrower segments, and the impressions-vs-clicks matrix that tells you what the numbers mean.
The exact regex to copy-paste
Copy this string exactly:
^(what|how|why|when|which|is|are|does|can|should)\b.*
Anatomy: ^ anchors to the start of the query. The alternation group captures the 10 most common English question-opening words (covering ~85% of question-shaped B2B queries per an internal WebFlur sample). \b is a word boundary so "whatabout" would not match but "what about" would. .* matches the rest of the query.
Two conscious trade-offs: (a) it's English-only — for multilingual B2B sites, add localized alternations (qué / cómo for Spanish, quel / comment for French, etc.); (b) it under-matches queries phrased as noun-shaped searches ("AI SEO pricing") that AIO still often surfaces for. Both are addressed by the additional patterns below.
Step-by-step: setting up the filter in GSC
Total setup time: about 5 minutes. Done once, gives you a permanent bookmarked view.
- Open GSC Performance → Search results. Set date range to Last 90 days for enough signal.
- Click +New → Query → Custom (regex).
- Paste the regex
^(what|how|why|when|which|is|are|does|can|should)\b.*into the filter box. - Ensure Matches regex is selected (not the default Contains). Apply.
- Bookmark the resulting URL — the filter is URL-encoded, so the bookmark preserves the view. Open it weekly.
Optional but recommended: name the bookmark descriptively (e.g. "GSC — AI-shaped queries, 90d") so you can share it with the team. If GSC feels slow with the regex applied, drop the date range to Last 28 days first — the filter is CPU-intensive on very large query sets.
4 additional regex patterns for narrower segments
The primary regex above is the broad AI-shape filter. The four below segment further — use them when you need a specific slice of AIO exposure.
^(what|how|why|when|which|is|are|does|can|should)\b.*\s+\S+\s+\S+\s+\S+.*
Filters the primary set down to queries with at least 5 words — the "long-tail" AIO surface. These are where B2B intent is highest and where competitor coverage is thinnest.
^(best|top|which)\s+.*(vs|versus|compared to|for)\s+.*
Isolates comparison-style queries that AIO renders as answer-with-table format. These are where WF-AIO-3 (mandatory HTML comparison table) has the highest impact.
^(what is|what are|define|meaning of)\b.*
Isolates definitional queries — the ones AIO extracts using the WF-AIO-1 answer-first opener pattern. If your GSC shows this segment growing but your clicks are flat, the answer-paragraph rewrite is the highest-ROI fix.
^(how much|how many|cost of|price of|pricing for)\b.*
Isolates cost- and quantity-shaped queries — high commercial intent and increasingly served by AIO answer boxes. Track this segment separately from the general AI-shape filter.
The impressions-vs-clicks matrix — what the numbers mean
Once the filter is saved and you have a week of data, the 4-cell matrix below tells you what to do next.
| Impressions | Clicks | Signal | What to do |
|---|---|---|---|
| Up | Up | Healthy AIO citation with your page as source | Keep shipping the current pattern. Add more clusters. |
| Up | Down | AIO expanding your visibility without giving the click (zero-click) | Reframe goal to brand mention inside AIO. Ship answer-first opener on displaced pages. |
| Down | Down | AIO displacing your rank without citing you (worst case) | Run manual citation check. Retrofit three-schema stack + answer-first opener urgently. |
| Flat | Flat | AIO does not surface for your category (rare in B2B in 2026) | Focus on GEO for LLM engines (ChatGPT / Perplexity / Claude) instead of AEO. |
Most B2B sites we audit sit in cell 2 (impressions up, clicks down) — that is the signature of the September 2026 auto-expand rollout. Full diagnostic in why is my organic traffic down — the AI Overviews diagnostic.
What to watch for weekly (the leading indicators)
Weekly review takes about 10 minutes. Three things to look at in order:
- Week-over-week impression change in the AI-shape segment. Rising = AIO exposure is growing for your category. Falling = something changed (either AIO shrunk in your category, or your ranking dropped upstream).
- Week-over-week CTR change in the AI-shape segment. Falling CTR is the earliest signal of AIO displacement — often 2–3 weeks ahead of the raw click count moving. Catch it here, ship the recovery playbook before the click drop becomes material.
- Individual query drill-down on the top-10 falling CTR queries. For any query where CTR dropped 20%+ week-over-week, run it manually in Google incognito. If AIO shows and cites a competitor, that query needs the recovery playbook this week, not next.
The GSC filter is the leading indicator; the full 5-source measurement stack (per-engine citation check, AIO tracker, revenue attribution) is the confirming layer. Full stack: how to track AI SEO — metrics, rankings & diagnostics.
- GSC — Performance report documentation: Reference for the query filter syntax used in this piece.
- MDN — Regular Expressions: General regex reference if you want to author your own custom patterns for niche segments.
