How to check whether a video's footage supports what its narration claims
Ask Vizard Agent to check two different things: whether the narration's factual claims stand up against official sources, and whether each shot actually documents the event being described. A video can pass the first test completely while being illustrated throughout with generic footage from somewhere else entirely.
What is the short version?
"Is this video accurate" is really two separate questions wearing one coat. The words can be entirely right while the pictures are stock footage of somewhere else, and a viewer will take those pictures as evidence for the words regardless of where they came from.
- Ask Vizard Agent to list the factual claims in the narration.
- Ask it to map each shot's actual visual content alongside them.
- Ask which shots are documentary and which are illustrative.
What do you need before you start?
The video and a reason for asking. Vizard Agent verifies against public and official sources, so an event with an official record — an earthquake, a policy, a match result — can be checked far more firmly than a private claim can.
- The video. The one being checked.
- Which claims matter. Or ask for all of them.
- The event. Named, with a date if you have one.
- How firm an answer you need. Confirmed, or plausible.
- Whether you can read the sources. Language matters here.
What do you type into Vizard Agent?
Ask Vizard Agent the two questions separately within the same brief. Rolled together they come back as a single answer, and that answer is nearly always about the words, because the words are by some distance the easier half to check.
Prompt
Variants worth knowing:
- "Two things please." Forces both checks.
- "Shot by shot." Stops a general verdict.
- "Generic illustration from elsewhere." Names the failure mode.
What does Vizard Agent actually do?
Here is the Vizard Agent sequence on a real news-style video that was submitted for checking. The claims and the pictures get investigated along two entirely separate tracks, and in this particular case the two tracks arrived at quite different conclusions.
- Reads the whole narration and identifies its factual claims.
- Maps the shot changes and what each one visually contains.
- Checks the claims against publicly reported sources.
- Goes to the original reports rather than relying on a search summary.
- Searches the official national agencies for the same details.
- Takes denser frames from each clip to judge what it really shows.
- Distinguishes documentary footage from generic illustrative visuals.
- Confirms the relationship between narration, on-screen text and picture.
- Looks for raw field footage from the event rather than broadcast commentary.
- Checks the timestamps on any original recording it finds.
Step four is the difference between checking and appearing to check. A search result summarising a claim is not a source; the report it summarises is, and they disagree more often than you would expect.
Step seven is the one that is usually missing. A clip of flooding is a clip of flooding, and the question is whether it is this flooding — and at a glance those two things look identical.
Step nine matters because broadcast material is not evidence of the event; it is evidence that a broadcaster covered it. Raw field footage from the scene is a different category, and worth separating out.
What does the result look like?
Two verdicts rather than one. The claims are confirmed, contradicted or unverifiable against named sources, and the shots are listed as documentary, illustrative or unidentifiable — so you can see exactly where a true video is being illustrated with something else.
When does this not work well?
Verification depends entirely on there being a record to verify against. Recent events, purely local incidents and anything official bodies have never documented will leave most of the claims unverifiable, and Vizard Agent reporting that is an honest answer rather than a failure.
- Very recent events. The record does not exist yet.
- Local incidents. Nobody official reported them.
- Sources in a language you cannot read. You cannot audit the audit.
- Generic subjects. "A factory" cannot be identified.
- Heavily edited footage. Recompression hides the identifying detail.
How do you fix a result that came back wrong?
Say which of the verdicts you doubt, and why you doubt it. Vizard Agent keeps every source it used along the way, so a disagreement between you becomes an argument about one specific document rather than about its judgement in general.
- "That source is not reliable." The claim is re-checked against another.
- "That shot really is from there." The identifying detail is re-examined.
- "You called it unverifiable." What would count as verification is agreed.
- "Check the on-screen text too." Titles and captions are audited as claims.
How does Vizard Agent compare to doing it yourself?
By hand you watch it, the claims sound plausible, the pictures look like the event, and you conclude it is fine. Checking properly means reading the official reports and identifying forty shots, which is a day's work that nobody does before sharing something.
| By hand | Vizard Agent | |
|---|---|---|
| The question | Is it accurate? | Are the claims true, and do the shots show it? |
| Sources | A search result | The report the search summarised |
| The footage | Looks right | Identified shot by shot |
| Illustration | Read as evidence | Listed separately |
| The verdict | One impression | Two findings, with sources |
Common questions
Can it verify anything? Only what has a public record. Vizard Agent says plainly when there is none.
What is illustrative footage? Generic material standing in for the event. Common, and not always dishonest.
Is illustration always wrong? No — but it should be labelled. Vizard Agent tells you where it is unlabelled.
Can it check my own video before I publish? Yes, and Vizard Agent finding it then is far cheaper than a correction afterwards.
Does it check the on-screen text? If you ask. Vizard Agent treats titles and captions as claims in their own right.
What about the numbers? Vizard Agent checks figures against the original reports rather than a summary.
Can it find the real footage? Sometimes. Vizard Agent searches the agencies and raw uploads, then reports what exists.
What if sources disagree? Vizard Agent reports the disagreement rather than quietly picking the convenient one.
How long does it take? Longer than watching it, and far less time than Vizard Agent would take you to do by hand.
Can it tell a presenter from field footage? Yes, and the distinction matters — one is coverage, the other is evidence.
Does this work in any language? Vizard Agent searches in the relevant language; you should be able to read the sources.
What about the date? Where the original files are reachable, Vizard Agent checks their recording timestamps directly.
Can it write up the findings? Yes. Vizard Agent lists the claims and the shots with their verdicts and sources attached.
Should I check every video? No. Give Vizard Agent the ones you are about to put your name to.