Trust & methodology
How to Verify a Claim on BestAIAgent.in
A practical, step-by-step guide to checking any claim on this site yourself, rather than taking a published field at face value.
Every other trust page on this site documents how we produce and gate content. This page is different: it's a walkthrough for you, the reader, on how to independently check whether a specific claim on a specific page actually holds up — in under a minute, using nothing but the page itself and its cited source.
Step 1: Find the receipt, not just the claim
On an entity page (an agent, model, framework or provider), look for the evidence block near the claim — it names a publisher, a retrieval date, and links to a primary source. If a field has no such block, it means the field is unknown by this site's own rule, not that the claim is unverified-but-true — treat it as genuinely unknown.
Step 2: Open the primary source directly
Click through to the linked source and read the passage yourself, rather than trusting our paraphrase of it. This matters most for anything volatile — price, plan limits, model version — since these can change after our retrieval date.
Step 3: Check the retrieval date against how fast the fact moves
A pricing fact retrieved 90 days ago is a different risk than an identity fact retrieved 90 days ago — see /trust/freshness-policy for the specific re-verification windows we apply per claim type. If a page looks stale for its claim type, that's worth flagging as a correction rather than assuming it still holds.
Step 4: Distinguish a fact from an editorial judgment
A number with a citation is a fact. A recommendation, a "best for X" framing, or a comparative verdict is an editorial judgment built on top of facts — it should be labeled as such on the page. If a judgment reads like a fact with no methodology attached, that's a real defect worth reporting.
Step 5: Report what doesn't hold up
If a source doesn't say what the page claims it says, or the retrieval date is older than the freshness window for that claim type, use /trust/corrections to report it with the specific field and the source you checked. This is the same reporting path a reviewer would use internally.
What verification does not mean
- A verified identity does not mean every field on that entity's page is verified — verification is field-level, not whole-page.
- A primary source can still be wrong; verification confirms we cited it accurately, not that the source itself is infallible.
- An evidence-ready comparison does not mean a universal winner — see /trust/comparison-methodology for what a comparison page does and doesn't assert.
Related methodology pages
Source receipts
- Internal — Evidence methodology (this site, retrieved 2026-08-29). The receipt structure this page teaches readers to check.
Last reviewed
- Page last reviewed: 2026-08-29
- Next scheduled review: 2027-02-28