Content: which parts are information and which are pressure?
Which parts of this are information and which are pressure?
This chapter deals with the things an operation actually makes: forged documents, screenshots, out-of-context pictures, edited clips and synthetic media. It gives a short verification routine, explains what provenance information can and cannot tell you, and shows why neither a missing watermark nor a detector score settles anything.
All stages
Information and pressure
The stage question is which parts of an item are information and which are pressure. Information is anything you could check: a named source, a date, a document, a figure, footage of an event. Pressure is anything that works on you before you check: a shocking image, a deadline, an instruction to share before it is taken down, a caption telling you what to feel. Honest communication uses both. In manipulative content the pressure carries the load and the information is thin, borrowed or absent.
Sort the item into the two piles first. Doing so slows you down at the moment the content is designed to speed you up, and it tells you what there is to verify.
Forged documents and screenshots
The leaked letter or memo has been a staple of covert influence for as long as the practice has existed, and a screenshot makes forgery easy. Secondary Infektion's typical product was an image of a supposedly leaked letter or post attributed to an official, and Graphika identified some 250 images planted in its articles, almost all of them suspected forgeries. They were confirmed as forgeries by the people and organisations whose names had been used. One forged tweet was exposed by the observation that nobody had replied to it or quoted it at the time it was supposedly posted.
Treat a screenshot as a claim that a document exists. Look for the document on the named author's own channel, look for any reaction at the time, and ask the named source. Oddities of format or language are worth noting but are weak evidence either way. Authentic leaks are the harder case. The questions then are who is supplying the material and why now, and both can be reported alongside it.
Out-of-context media, cheapfakes and synthetic media
Paris and Donovan place manipulated media on a spectrum. At one end are cheapfakes, made with ordinary tools: a clip cut short, slowed or relabelled, or a real photograph given a new caption, date or place. At the other end are deepfakes produced with machine learning. They argue that the cheap end deserves at least as much attention because it is easier and already common. Wardle and Derakhshan likewise treat genuine content in a false context as a category of its own.
Synthetic media is real and improving, but the public record so far is undramatic. The operations disrupted by OpenAI in 2024 used generative tools mainly to produce more text and images with less effort. They drew little authentic engagement and were caught largely through their behaviour and through earlier research, not because anyone recognised machine-written prose.
Verification steps and their limits
Four checks cover most cases. Reverse image search: look for earlier appearances of a picture or a video frame, and note the earliest date and place. Find the original: the full recording, the whole document, the complete quotation. Read laterally: see what independent sources say about the item. Ask the named source through a channel you already trust.
Provenance information helps when it is present. Some cameras, editing tools and generators now attach signed records of where a file came from and how it was changed. Most media carries none, and platforms often strip it. The absence of a watermark, label or credential therefore proves nothing in either direction: unlabelled content is not thereby authentic and not thereby fake. Automated detectors make errors both ways, and a score is not a finding.
One further caution comes from Chesney and Citron, who named the liar's dividend. Once people know that convincing fakes exist, genuine recordings can be waved away as fabricated. Calling something fake is a claim that needs the same evidence as calling it real.
Techniques you will meet here, and the tell
A fabricated letter, memo or message is presented as a leak, which excuses the lack of an official source and adds the appeal of a secret.
The tell: There is only an image of the document, no one can produce the file or the original, and the named author denies it.
A fake post, headline or message is circulated as a screenshot so that it cannot be clicked through to a source.
The tell: The original cannot be found on the account or site it supposedly came from, and nobody reacted to it at the time shown.
A genuine photograph or video is attached to the wrong event, place or date.
The tell: A reverse image search finds the same picture published earlier, somewhere else, about something else.
Simple edits (a cut, a crop, a change of speed, a new caption) do most of the work that is popularly blamed on sophisticated fakes.
The tell: The clip starts or stops mid-sentence, and the full recording gives the words a different meaning.
Generated images, audio or video depict something that did not happen, or give an invented persona a face and a voice.
The tell: The only copy is the one being shared, no independent recording of the event exists, and the person depicted denies it through their own channel.
Words are put in the mouth of a trusted or hated figure, often on an image card that looks like a news graphic.
The tell: The quotation appears nowhere in the speaker's own record or in any report from the occasion named.
Genuine and inconvenient material is dismissed as fake because fakes are known to exist.
The tell: The denial offers no evidence of fabrication and relies entirely on the general possibility of it.
A score from an automated detector is quoted as proof that something is synthetic or authentic.
The tell: The percentage is the whole argument, with no original located and no source consulted.
Worked example
In one week four items about the Alder Bay footbridge circulate in the local groups and on the look-alike site: a screenshot of a council memo, a photograph of a wrecked footbridge said to be the same design, a nine-second clip of Councillor Okafor, and a voice message said to be the council leader. Each is captioned with some version of share this before they take it down.
- 1Priya sorts each item into information and pressure. The captions and the urgency are pressure. The memo text, the photograph, the clip and the audio are the things to check.
- 2The memo exists only as a screenshot. Its letterhead uses a crest the council replaced two years ago. More usefully, the clerk confirms in writing that no memo with that reference exists. Priya records that the council denies the document and that its format does not match; she does not claim to know who made it.
- 3A reverse image search finds the bridge photograph in a news report from another country six years earlier, showing a different kind of bridge after a storm.
- 4She finds the full meeting recording on the council's own channel. The clip has Councillor Okafor saying the traders will just have to live with it. The full sentence says that if the loading bays are wrong the traders will just have to live with it, that this is not acceptable, and that the bays must be settled first.
- 5A reader tells her an online detector scored the voice message as 87 percent likely to be synthetic. Priya does not use the score. The council leader denies recording any such message, and no original can be found. She files it as unverified and denied, and notes that she cannot establish how it was made.
- 6She saves a dated copy of each item with the address where she found it.
Outcome. Priya has four findings stated at the strength the evidence allows: a document denied by its supposed author, a photograph from another place and time, a clip cut to reverse its meaning, and an audio message that cannot be verified. She has shown the originals where they exist. She has not accused any resident who shared them.
Content verification
Use one sheet for each document, image, clip or recording you are actually examining. Record what you checked and what you could not establish. Saved in this browser only.
Describe it briefly and note where and when you found it.
Choose the closest description.
Tick everything in the item that pushes you to react or share before checking.
List the parts that could be checked: names, dates, places, figures, the document or footage itself.
Tick each check you actually completed.
Give the earliest date and place you could find for the item, or say that you found none.
Choose the statement your evidence supports and no stronger one.
Write down what remains unknown, such as who made the item or how it was produced.
Decision check: what is this item, on the evidence?
0/7If any of these is not true: Do not share the item and do not describe it as fake or as real. Say that it is unverified, say what you checked, and keep a dated copy. If the item names a person or institution, ask them. If you are a journalist or moderator, hold publication or labelling until one of the five findings can be stated honestly.
Common mistakes
- ×Relying on a detector score, or on the lack of a watermark or label, in either direction.
- ×Calling unwelcome but genuine material a deepfake. That is the liar's dividend, and it is available to your own side as much as to anyone else.
- ×Stopping at the verdict. Saying an item is false persuades fewer people than showing the original photograph, the full clip or the denial.
- ×Over-attribution: treating a fake as proof of an organised operation. Hoaxers, satirists and people having a joke produce forged screenshots and recaptioned photographs every day.
- ×Forgetting that authentic material can mislead. A verified document or clip still needs its date, its context and an account of who supplied it.
What will you actually make, and does every element earn its place honestly?
Related
References
- Paris, B., & Donovan, J. (2019). Deepfakes and Cheap Fakes: The Manipulation of Audio and Visual Evidence. Data & SocietyThe spectrum from simple edits and recontextualisation to machine-generated media, and the importance of the cheap end.
- Chesney, R., & Citron, D. (2019). Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security. California Law ReviewThe liar's dividend: the existence of fakes lets genuine evidence be dismissed.
- Wardle, C., & Derakhshan, H. (2017). Information Disorder: Toward an Interdisciplinary Framework for Research and Policy Making. Council of EuropeFalse context, manipulated content and fabricated content as distinct categories.
- Silverman, C. (Ed.) (2014). Verification Handbook: A Definitive Guide to Verifying Digital Content for Emergency Coverage. European Journalism CentreStandard newsroom verification of images and video: find the original, the earliest appearance and the source.
- Graphika (Nimmo, B., Francois, C., Eib, C. S., Ronzaud, L., et al.) (2020). Exposing Secondary Infektion. GraphikaForged documents and screenshots as an operation's main product, and how they were identified as forgeries.
- OpenAI (2024). AI and Covert Influence Operations: Latest TrendsGenerative tools were used for volume and efficiency, and the disrupted operations drew little authentic engagement.