Check exact statements, including the visual ones.
A temperature map can show surface heat while the narration describes air temperature. Both measurements may be accurate, but putting them together can mislead the viewer. This guide follows six such claims through the decision to keep, change, remove or investigate them.
Mark numbers, dates, quotations, identities, comparisons and causal statements. Also mark visual assertions: a recreated chart presented as observed data, a stock shot standing in for a location, or a caption assigning a statement to the wrong speaker. The source list is the starting point; the job is to determine what each source actually supports.
Repair an attractive but unsupported explanation.
Original teaching example. The proposed episode concerns a city heat map. No local temperature data or guest interview has been collected. Its scientific starting points are NASA’s land-surface temperature explanation and EPA’s account of trees and vegetation.
| Draft claim or treatment | Check and decision |
|---|---|
| “The red patch proves the air here is 12 degrees hotter.” | Reject. No selected dataset, comparison, unit context or local air measurement supports this number. |
| “The satellite map measures the same thing as the weather report.” | Correct the measurement distinction using the NASA source. |
| “Trees can cool through shade and evapotranspiration.” | Retain with appropriate scope; EPA supports the mechanisms, not a guaranteed local amount. |
| “Every tree cools every street by the same amount.” | Reject the universal claim. The supplied evidence does not establish it. |
| “Our guest measured this exact street.” | Remove until a real guest’s work and the location are verified. No guest is booked in this example. |
| Paint a red region on an invented map and present it as data. | Label it as an illustration or replace it with cleared actual data and sufficient metadata. |
A corrected passage
A land-surface temperature map tells us about the surface seen by the sensor. It is not the same measurement as the air temperature in a weather report. Trees can influence temperatures through shade and evapotranspiration, but these sources do not establish a particular temperature difference on this street. We would need the relevant data and measurement context to make that claim.
Trace the claim to the strongest usable evidence.
Prefer the original dataset, study, record, recording or firsthand account when it can answer the question. A secondary article may be useful context, but check whether it has narrowed, rounded or expanded what the original says. Several pages repeating one claim do not become independent corroboration.
Record the exact source location, the version or date checked, and the scope: place, population, period, unit and denominator. A percentage without its base, an old cumulative count presented as a recent result, or a comparison between different measurement windows can sound precise while being wrong.
For quotations, compare the exact words with the recording or authorised original and retain the surrounding meaning. A paraphrase belongs outside quotation marks. An automated transcript helps locate a passage; it is not authority for words that the recording does not contain. The Associated Press’s published news values provide an accuracy and correction reference; the worksheet here is our practical production method.
Choose retain, narrow, qualify, remove or investigate.
Not every failed check means discarding the whole argument. Sometimes the claim needs a narrower subject, an explicit date or the correct unit. Sometimes the compelling number has no defensible source and must go. Preserve the useful question while removing certainty the evidence cannot carry.
A qualification should sit where it changes the interpretation. Burying “results may vary” in a description will not fix a spoken promise of a universal result. Equally, do not overload a simple supported sentence with caveats that do not affect its meaning. The aim is accurate understanding.
Give unresolved claims an owner and a deadline. A missing source does not become verified when the script is approved for tone. If the evidence cannot be obtained, decide whether the argument can stand without that claim before the shoot or finishing work makes it expensive to change.
Carry the check through the edit and packaging.
After picture lock, check the actual spoken words, on-screen text, chart labels, image credits, captions, thumbnail and description against the approved register. A safe script can become unsafe to trust after a punchier title or a shortened clip. Log changed claims rather than assuming the earlier review still applies.
If a mistake is found after release, identify the affected asset and the correction needed. Preserve an internal record of what changed and when; use an appropriate visible correction where the audience needs it to understand the work. Do not silently treat an amended page as proof that the original publication was correct.
The editable fact-check register includes the six decisions above, corrected wording and fields for the actual source, reviewer, date, affected assets and correction status. Use it alongside the research brief and approval map.
Questions about the method.
Can a fact checker verify a script from its source links alone?
The links must be read against the exact claims. Check source scope, dates, units, quotations, comparisons and what the source does not establish. The finished edit and graphics also need review.
Should I remove every uncertain statement?
No. Some questions and interpretations are useful when clearly identified. Narrow or qualify what the evidence partially supports, and remove or investigate claims that cannot be defended.
Tell us what you want to make.
Tell us about the channel or idea, the team you have and where you need help. We’ll discuss the scope on a 30-minute call.
Book a 30-minute call rishwajeet@machinehouse.media