See the retrieval layer
Capture web-search queries explicitly surfaced by ChatGPT, Claude and Google Search in a local side-panel trace.
Open-source Chrome extension
See the queries behind AI search.
Open Queries shows the web-search queries surfaced by ChatGPT, Claude and Google AI Overviews in a local Chrome side panel—and unlocks the query trace and on-demand fan-out estimates after you accept the privacy setting.
Queries, not conversations · Observed stays observed · AGPL-3.0
Supported now
- ChatGPT
- Claude
- Google AI Overviews
The open retrieval layer
Useful to inspect. Rigorous enough to audit.
Open Queries keeps direct observations and probabilistic estimates separate from capture through aggregation.
Estimate with native evidence
Accept the privacy setting, then request plausible fan-outs ranked by each provider’s own token probabilities—or repeated native samples when logprobs are unavailable.
Audit the whole system
Extension, Worker, schemas and methodology are public under AGPL-3.0-or-later. No black-box ranking layer.
A narrow data contract
Transparent without surveillance.
- No chat messages, titles, account identity or conversation URLs
- Privacy starts unaccepted; accepting it reveals queries and unlocks fan-out estimates
- Sensitive-pattern filtering runs locally and at the Worker edge
- Eligible observed queries are written directly to D1
- Estimated fan-outs are never stored as observed queries
Open methodology
Read the evidence, not the marketing.
Query fan-out
What are fan-out queries?
A practical explanation of the hidden sub-queries answer engines use to retrieve evidence before responding.
5 min readAEO / GEOWhy AEO and GEO need query data
Traditional keyword tools describe human search demand, not the retrieval steps AI answer engines perform for users.
6 min readEvidence methodObserved vs. estimated AI search queries
How Open Queries labels direct UI evidence and probabilistic reconstructions without mixing the two.
4 min readTechnical methodologyEstimating fan-out queries with log probabilities
A mathematical account of provider-native inverse perplexity, token alignment, repeated native sampling and the limits of closed-model reconstruction.
14 min readMake AI search transparent
Build the query history AI search is missing.
Install, accept privacy and inspect. Every layer stays open.