Citations as a contract.
We cite what we use. We expect the same in return.
Primary first, secondary second
When the research agent visits a company, model, or person, it prefers in this order: official press release, SEC or court filing, official documentation, major outlet (TechCrunch, Bloomberg, Reuters, WSJ, FT, The Information, Stratechery), then trade press, then community writeups. Each retrieved URL is stored alongside the run and shown on the profile under Sources.
No fabrication, no laundering
The research agent's prompt explicitly forbids inventing figures and inflating uncertainty into precision. If a valuation is rumored but not in a citable source, the field is left null rather than guessed. If a number is a range, the displayed value is tagged estimated. We do not "borrow" verbatim from competitors — every brief is written from the agent's own retrieved-source synthesis.
Attribution we ask for
The AIDB master index, leaderboards, and newsroom are released under CC BY 4.0. Use, remix, and quote freely with attribution:
AIDB — The IMDb of AI (aidb.digital). Retrieved <date>.
Academic / paper citation:
AIDB Reference Atlas. The IMDb of AI. https://aidb.digital, 2026.
If something's wrong
Every profile carries a Refresh button. Pressing it re-runs the Sonnet 4.6 agent against live web search and updates the record on success — usually in under a minute. If a figure is materially wrong (e.g. an old valuation displayed as current), email the admin or hit Refresh yourself. The previous run is preserved in research_runs so the change is auditable.
Which model wrote what
The model used for each run is stored on the research_runs row and visible in the company profile sidebar:
- Profile enrichment + boardroom debates — Claude Sonnet 4.6 with web_search
- Discovery + tools catalog + bulk classification — Claude Haiku 4.5 with web_search
- Footer assistant — Claude Haiku 4.5 streaming, grounded in the AIDB index
- Live benchmarks — pulled from artificialanalysis.ai
What AIDB is not
AIDB is editorial reference. It is not investment, legal, or business advice. We do not endorse, rate (except in the explicit tools catalog), or recommend the companies we cover. The freshness badges describe data age, not company quality.
See also: /methodology · /about · /faq