Pulled my top 200 GSC queries into a prompt tracker — here's what survived the cut

Data engineer brain takes over: I exported 90 days of Search Console data, ran every converting query through a conversational rewrite script, and ended up with 200 candidate AI prompts to track. Then had to cut it to 25.

The filter I used, in order:

1. Drop any query where the Google click-through was under 0.5 %. If people scroll past it on Google, they're probably not asking AI about it either.
2. Drop anything where 3 runs on ChatGPT returned the same 5 incumbents with no variation. Those slots are locked, I can't win in the next 6 months.
3. Keep anything phrased as a 'for [specific use case]' long-tail — those are winnable.
4. Keep all 5 defensive brand probes ('is [brand] still active', 'is [brand] legit', etc.) regardless of volume.

Ended up with 21 buyer-intent + 4 defensive probes. Running weekly cadence on the top 10, monthly on the rest.

What surprised me: 6 of my surviving prompts were queries I had NO GSC data for — pure AI-native queries that nobody types into Google. Perplexity Sonar 3 is surfacing a completely different buyer intent layer than what Google ever showed me.

Anyone else finding that AI-native queries are a different population from organic search queries? Curious if this is category-specific.