How to read a fact-check report without relying only on its verdict

Locate the claim the report actually checked

Compare the report’s wording with the message you received. Does it preserve the date, location, number and qualification? A report can be accurate about one claim while failing to answer another. This often matters when a forward contains an old video plus a new caption.

For a long message, expect several distinct claims rather than one all-purpose answer. A report might establish the origin of a photograph but leave a numerical claim unanswered. Do not use the verified part to certify unrelated sentences, the sender or an entire organisation.

Find out what material was examined

Look for an account of the input: the original file, a screenshot, a transcript, a linked page or a preview. Check whether the tool could access the full material. If a report says it examined only a title and cover image, its conclusion should not pretend to establish everything spoken in the video.

Pay attention to quotations. Compare the words with the original passage or recording, including the surrounding sentence. A transcript can omit a negation or mistake a name. If the report misunderstood the input, fix that problem before debating the verdict.

Open the sources that carry the conclusion

Start with the source most directly tied to the disputed fact. Locate the relevant text, date or image and read the surrounding context. Ask whether it establishes the claim, contradicts it, merely repeats it or supplies background. A citation can be real while its role in the argument is weak.

Consider a hypothetical report that labels a public notice false because a website has no matching announcement. That absence may justify doubt, but it does not by itself prove that the notice was forged. A dated correction from the issuer or an original notice with materially different wording would provide different evidence. Read for that distinction.

Trace repeated accounts to their origin. Five links quoting the same press release are not five independent investigations. Conversely, a single original record may be the right evidence for a narrow question about what that record says. Quality and fit matter more than the length of the source list.

Check the steps between evidence and verdict

A strong explanation makes the reasoning inspectable. If the source shows a video existed years earlier, that supports rejecting a claim that it was first filmed yesterday. It may not identify the people, the location or the original uploader. The conclusion should stay within what the evidence establishes.

Watch for broad labels applied to mixed claims. A number may be correct but missing the denominator; a quote may be real but attached to the wrong occasion. Ask which component changes the meaning and whether the report explains it. This is more informative than arguing over a word such as misleading without discussing the evidence.

Read uncertainty as part of the result

A report that cannot settle a question should explain the missing evidence. NOT PROVEN and FALSE are different conclusions. The first says the available evidence did not settle the claim; the second requires evidence against it. A short or unsuccessful search does not turn one into the other.

Also distinguish confidence in a narrow observation from confidence in the whole story. You may be sure a screenshot contains a sentence without knowing who created it. A confidence label or number does not replace that explanation or establish a system’s measured accuracy.

Check freshness and the correction route

Look at both the report’s date and the dates of its evidence. A later announcement, revised document or newly available recording may change the answer. A date alone is not proof that the sources were rechecked, so read any update note.

If you find an error, send the exact claim, the disputed passage and the source that changes it through the publisher’s stated correction channel. Explain the difference rather than only saying the verdict is biased. Avoid adding private details unrelated to the correction. A useful correction should identify what changed and why readers should interpret the result differently.

Share the reasoning with the result

If you share a report, include the narrow conclusion and its main limitation. For example: the image appeared earlier, but this check did not establish its original location. Do not remove uncertainty to make the message shorter. That can turn a careful report into another misleading forward.

ForwardCheck reports are AI-assisted and can make mistakes. ForwardCheck is a beta tool, not an independent newsroom, and a generated report should not be assumed to have had a person review it before publication. Use its source links and methodology to assess the reasoning, then decide whether the evidence is sufficient for the claim you are sharing.

Common questions

Can a true verdict apply to only part of a message?

Yes. Read the precise claim and its scope. A supported statement does not validate the rest of a forward.

What is the most useful way to challenge a report?

Identify the exact passage, provide relevant evidence and explain how it changes the conclusion or exposes a missing qualification.

Should I share a report that leaves the claim unresolved?

You can share it as an unresolved check when that is useful, but preserve that limitation. Do not present uncertainty as a confirmed verdict.