The Google Search Console Regex Trick Everyone's Getting Backwards

Welcome to the mid week issue of the Click Raven newsletter.

SEO pros have been sharing the regex trick below for Google Search Console.

Trick: Filter your Query dimension with ^([^ ]+ ){3,}[^ ]+$ to isolate 4+ word queries, and increase the number for longer ones to find what people are asking AI Mode and AI Overviews and where your brand is showing.

Here is the thing, the first version of this advice is a bit incomplete.

At the 4 to 8 word range, with impressions in the hundreds or thousands, a lot of what shows up is not real human behavior.

Nobody types the identical 8 word phrase 2,000 times, so why would a query show 2,000 impressions? And there is no way 2,000 people would type an 8 word phrase exactly the same way.

That pattern is far more consistent with autocomplete selections, AI Overview query fan out, or SEO and AI visibility tools scripting identical searches at scale.

Search Console gives you no dimension to separate any of that from genuine intent.

What I normally do instead is push the regex to 15+ words and drop the impression threshold to single digits.

Like this: ^([^ ]+ ){15,}[^ ]+$

That is where it gets interesting.

See the real examples, unedited, each with 1 impression:

• "why have i gained 12kg even if im excersizibg 30 hours per week and am on a healthy diet"

• "wife is always tired even though she didnt do anything energy consuming, what could it be?"

• "how do i know if my new longevity supplements and diet are actually reversing my biological age?"

These read completely differently to the mid range noise. There are typos, run on sentences, and first person context in there, just the way you and me search.

Nobody builds a monitoring bot to ask about their own wife's fatigue, and nobody scripts a query with a spelling mistake in it.

Real natural language varies too much for the exact same long sentence to repeat at volume, which is why low impressions at extreme length are the stronger human signal here, not the weaker one.

That is the opposite of how most people use this filter. Everyone hunts for the biggest impression numbers. The real signal sits in the ones with 1 or 2.

Summary:

1. Push the filter to 12-15+ words. Identical repetition by bots or fan out becomes structurally unlikely at that length.

2. Stop filtering for volume. A long, imperfect query with 1 impression is more likely a real person than a clean 8 word phrase with 2,000.

3. Read for grammar. Spelling errors and casual phrasing are a feature here, not noise. Scripted tools do not misspell words.

4. Treat these as genuine voice of customer data, the raw material for FAQ content, GEO and AEO targeting, and product messaging.

Google still will not tell you what people type into AI Mode or AI Overviews.

Their new Generative AI performance report in Search Console shows impressions, pages, countries, devices and dates, but no queries and no clicks.

Until that changes, this is the closest thing marketers have to actual verbatim intent, if you know where in the data to look.

Feel free to reply here if you have any questions.

Until next time,
Ian @ Click Raven.