Google Search Console (GSC) holds a trove of the real questions people already type to reach you — and a regex filter turns that into a list of the question-shaped queries worth building AI-citable content around. This method comes from a widely-shared June 2026 LinkedIn tip by SEO practitioner Mark Williams-Cook: instead of asking an AI to guess your query data, filter GSC's own Performance report by intent. Here's how, and how to use the output for AI search.
Why mine Search Console for AI search at all?
First, an honest scope note. GSC reports classic Google Search queries — not the prompts people enter into AI Mode, ChatGPT, or Perplexity, which no platform exposes as a clean log. So this is a proxy, not a direct AI-query feed.
It's a good proxy, though. The longer, question-shaped queries you already rank for are exactly the intent AI engines decompose into sub-queries when they fan out. Mining them tells you which questions you have real demand and standing for — the strongest candidates to answer extractably and to add to your tracked prompt set. It pairs directly with how to choose which prompts to track.
How do you filter Search Console by regex?
The steps, per the source method:
- Open Google Search Console and go to the Performance report.
- Click "+ New" / Add filter at the top of the report.
- Choose Query.
- Switch the match drop-down to Custom (regex).
- Paste one of the intent patterns below and apply it.
The table then shows only the queries matching that intent. Export each one to build an intent-segmented map of the demand you already capture.
Which regex patterns surface AI-search intent?
Each pattern isolates a different intent. Treat them as starting points and adapt to your vocabulary:
| Intent | What it surfaces | Regex (Query → Custom regex) |
|---|---|---|
| Informational | Guides, tutorials, how-tos | `\b(how to |
| Comparison | "best", "vs", "alternative" | `\b(best |
| Product / service | "is X good?", "where to buy X" | `\b(price |
| Transactional | Buying, pricing, locations | `\b(buy |
| Navigational | Brand, reviews, support | `\b(review |
| SaaS / tools | Tool-seeking queries | `\b(?:tool |
| Likely questions | Queries of 4+ words | (\w+\s){4,} |
The SaaS/tools pattern is credited in the source to Pietro Mingotti; the 4+-word pattern catches the long, natural-language queries that look most like AI prompts. Apply each as a separate filter and read the proportions — which intents you already win, and which you barely show up for.
The questions people already ask Google to find you are the cleanest signal of the AI prompts you should be answering — extract them, then make each one a self-contained, answer-first chunk.
What do you do with the extracted queries?
Turn demand into citable content and a tracking list:
- Group by intent, then check whether you have an answer-first chunk for each high-volume question. Gaps are your content brief — this is the query fan-out content loop in practice.
- Promote the recurring questions to tracked prompts so you can measure AI visibility, not just classic rank — the bridge to why you might be invisible in AI search despite ranking.
- Refresh the pages that win these queries, since live AI retrieval favours recently-updated content.
What are the limits to watch?
Read this directionally, not as gospel. As practitioner Ryan Jones noted on the same thread, regex filters change GSC's metrics to sum across matching pages, which can inflate impression counts — so trust the patterns and proportions, not exact totals. GSC also samples and thresholds query data, and the same regex works in other tools that support it (such as Ahrefs) if you want a second view. And remember the headline caveat: this is classic-search demand standing in for AI intent, so corroborate it against your own AI-visibility snapshots.
That corroboration — taking these real questions, posing them across ChatGPT, AI Mode, Perplexity, and Claude, and tracking whether your brand is named and your pages cited over time — is exactly what Buffy Intel is built to do.
