Technology
Financial AI you can check beats financial AI you have to trust. AnalystBook is built as trust machinery — a record built by fixed rules, automatic accuracy checks, SEC audits and daily evaluation — not as a promise.
Every figure, score and change flag in AnalystBook is taken directly from the company's own SEC filings by fixed rules — the machine-readable financials (XBRL), proxy pay tables, the subsidiary list (Exhibit 21), insider forms, activist stakes. A dedicated extractor per document type builds an organized record: the same filing always yields the same figures, and no language model ever decides a number.
Ana and the Intelligence Brief work on top of that structured record through read-only tools. No web browsing, no answering from training memory. Because the data underneath is organized and period-matched, the AI can compare companies correctly, cite exactly, and know precisely what it doesn't know.
Every statistic in an answer is automatically checked against the record — one that doesn't match is dropped. Management 'quotes' are checked the same way: one that wasn't actually read from a filing gets caught and removed. Investigations run under strict time budgets, so an answer is always finished honestly from gathered evidence.
Every Intelligence Brief's figures are checked against raw SEC data and a computed year-over-year change report before publication. A brief with a mismatched figure is rejected and never served — the rejection is stored for audit. Each published brief carries its gate verdict and is monitored for staleness against the filing calendar.
Ana's behavior is pinned by a certified question bank across dozens of research desks — answers whose correctness has been verified — and evaluated daily. Every answer can be flagged in one click by any user, and flags land on an internal health dashboard the team reviews.
Every number links to the filing paragraph it came from; every AI claim carries a citation to the source document. You never have to trust the system — checking any claim is one click. That's the whole design.
FAQ
The data layer — financials, forensic scores, changes, segments, pay, ownership — is produced by fixed extraction rules reading official filings, not by a language model. AI is used only for the narrative layer (Ana's answers, the Intelligence Brief), and that layer is grounded in that record and checked against it before you see it.
The architecture is deliberately model-agnostic: the AI layer runs on AnalystBook's own tools and accuracy checks, so the underlying model can improve or change without changing the trust properties. The guarantees come from the record and the gates, not from any vendor's model.
Yes — no AI is infallible, and we say so rather than promising magic. What AnalystBook engineers is verifiability and refusal: figures are gate-checked against the record, claims carry citations to the filing, and questions the record can't support get an honest 'I don't have that' instead of a guess. For research only; not investment advice.
Traceability. Because everything derives from official filings, every figure can link to its exact source paragraph and every claim can be audited — by you or by our gates. Licensed feeds would break that chain (and quietly put someone else's methodology between you and the record).
14 days free, no credit card required. Every claim links to its filing.