The key blocks
- Break a story into testable claims; repeated copies of one statement are not independent sources.
- Record event, publication and retrieval times separately, preserving the original timezone alongside Taipei time.
- AI can organize evidence, but each headline claim still needs checking against the source. Keep unknowns visible.
Break the headline into claims
Imagine a group message saying that a protocol has been hacked and all its funds are gone. Before asking AI for a sharper headline, split the sentence into three claims: a security incident occurred, funds were lost, and the loss includes everything. This is a fictional checking exercise, not a report of an ongoing event. Separate claims make the limits of the evidence easier to see.
For each claim, record the original speaker, an accessible source and what it establishes. A team announcement that some functions are paused does not establish that every asset is safe or lost. Ten accounts reposting one screenshot may still trace back to a single unconfirmed source. Check how each source obtained its information before counting it as independent confirmation.
Verify the address independently
Find the announcement through a known official entry point. Check the full domain, publisher and location of the article instead of following an unfamiliar message’s supposed official link. CISA’s phishing guidance encourages verification through the organization’s website; the FBI’s IC3 warns that HTTPS and a browser lock alone do not establish trust. For crypto readers, this protects both source evaluation and wallet access.
Check that a citation actually reaches the relevant original, rather than a homepage, error page or unrelated article sharing a keyword. Reading a news announcement should not require a recovery phrase or wallet authorization. For a screenshot, record the account, displayed time and how you obtained it, then look for the original post. If it cannot be found, say that the screenshot remains independently unverified.
Keep three clocks, not one
Use separate fields for the event time, the source’s first publication time and your retrieval time. An updated timestamp may reflect an edited paragraph rather than a new event. Words such as “today” and “earlier” can mislead readers in another timezone. For deadlines, outages or launches, give the full date, time and timezone when the source provides them.
For example, 17:00 UTC on 1 November 2026 is 01:00 on 2 November in Taipei, UTC+8. This is a conversion example, not an event timestamp. Preserve the original value before producing a local display. If a source provides only a date, mark the time as unspecified instead of inventing midnight. Similarly, a rolling 24-hour market change is not the same measurement as a local calendar-day change.
Compare AI’s wording with the evidence
Ask for an evidence table containing the claim, its location in the original, its time and the missing evidence. Then open each source yourself. Pay particular attention to verbs such as considering, proposing, applying, approving and taking effect, as well as asset quantities and units. Cross-language number conversions need calculation, not a mechanical substitution of the word following the digits.
In its AI standards update of 23 July 2026, AP kept verification and editorial responsibility with its journalists. Assistance does not replace checking. That principle informs this checklist; it is not an AP endorsement of BlockBlah. Translation must also preserve uncertainty: “the team says it is investigating” cannot become “a hack is confirmed” merely to make a sentence shorter.
Decide what is ready to share
Write your conclusion in one sentence, then check whether every consequential word has support. If the evidence establishes only that a team made an announcement, attribute it to the team. Hold back when the central claim cannot be verified. If public-interest reporting requires discussing an unverified claim, place that qualification where readers first encounter it rather than burying it at the end.
Keep the checking time, sources, limitations and subsequent corrections so another reader can repeat the process. IFCN’s principles emphasize transparent sourcing, methods and corrections; citing them does not claim that BlockBlah is a verified signatory. This original, AI-assisted media-literacy guide does not fact-check a particular incident or provide investment advice. Useful speed starts with knowing which parts are confirmed and which remain unknown.
Sources & further reading
For information and education, not investment advice. Crypto assets are volatile. Verify independently.
