The Exa guide records an April 6, 2026 test. The October 18 audit below measures retrieval presence for a small query sample. These are separate observations.
April review: historical measurements
The guide reports APAC testing with exa-py 2.11.0, ten latency iterations and example queries including “AI agents,” “Claude 4.6 features” and “companies using transformers for healthcare.” Its 452 ms average latency and 79–90% estimated token reduction remain April observations. The token figures were estimated from character counts, not billed downstream-model savings.
A sanitized raw-response bundle is not attached to this page. The recorded figures have not been independently reproduced here. They do not establish current latency, relevance, prices or an available free allowance. Use Exa’s official documentation for current API and access terms.
October 18: small retrieval sample
The retained audit covered five sites with 35 searches and one controlled fresh-content fetch in total. AIHackers’ portion was:
| Query group | AIHackers appeared in the top 10 | Scope |
|---|---|---|
| Branded | 1 of 1 | The branded result ranked first. |
| Unbranded | 0 of 4 | Four selected questions; no general discovery estimate. |
| Domain diagnostic | 2 of 2 | Domain-filtered queries test retrieval under that constraint. |
The AIHackers pricing article ranked first in a filtered search and was absent in its unfiltered counterpart. Its extracted text retained the dated comparison and tax, fee and plan qualifications. The controlled fresh-content fetch concerned another site; it is not an AIHackers freshness test.
These hand-picked queries have no historical baseline. They cannot identify a ranking cause, Google traffic, audience gains or general visibility across LLM products. Search Console impressions and clicks would require separate evidence. No new searches or provider experiments were run for this correction.
Evidence availability
This note summarizes the retained Exa-visibility-findings.txt audit, dated October 18, 2026, and the existing April guide. It supplies a readable scope statement; it is not a raw query/response download. A later retrieval comparison should retain the same query set and separately label direct source fetches and search results.