Tactic · Disinformation
The fabricated screenshot
An image of text purporting to be a post, message or headline. It takes under a minute to make, cannot be verified from the image, and is shared at enormous scale.
Where is the link, and does the post it claims to show exist?
What it looks like
A screenshot is not a manipulated image. It is a genuine image of fabricated text, and that distinction is why it defeats every defence at once.
Image forensics look for splices, resampling and inconsistent lighting. A screenshot of text typed into a page has none of those, because the pixels really are authentic. Provenance signing does not help either: a screenshot is a new capture with no history, and for text on a page there is no original asset to match against. And platform moderation cannot touch it, because the post never existed on the platform — there is nothing to label, downrank or remove.4
Why it works
The largest survey of image-based misinformation, built from 135,838 fact checks with 32,320 image-and-claim pairs annotated by hand, gives screenshots their own annotation branch explicitly because of how common they are, and distinguishes genuine social-media screenshots from fabricated ones produced by online screenshot generators.1 Its authors note that screenshots of social media posts are especially common even though every platform makes resharing the real post trivially easy — which is itself the tell.
People are close to chance at detecting manipulated social media images. In a preregistered randomized experiment, a control group averaged about 56 percent correct across 32 images.2 Two earlier studies with 707 and 659 participants found overall accuracy of 66 and 62 percent, and correct localisation of the manipulation in only 45 and 56 percent of cases.3
Both of those measured manipulated photographs. For a fabricated screenshot the true rate is worse, because there is nothing in the image to detect. The failure is not that people look badly. It is that looking is the wrong operation.
accuracy at spotting manipulated social media images in a preregistered experiment — barely above a coin flip. Feedback raised it to about 69 percent; adding education on top of feedback added nothing.Hoq, Facciani & Weninger, HKS Misinformation Review, 2 Apr 2025, n=268
of participants who could correctly locate the manipulation in an image, even when they suspected one.Nightingale, Wade & Watson, Cognitive Research, 2017, n=707 and n=659
accuracy at machine-extracting the timestamp and the account handle from 4,504 archived screenshots, then matching them against web archives at F1 = 0.96 with zero false positives.Zaki, Nelson & Weigle, ACM Hypertext, 2025, n=4,504 screenshots
of recent image-based fact checks involve media, with context manipulations — including self-contextualising images carrying their own overlaid text — dominating throughout.Dufour et al., Google Research with Full Fact, May 2024, 135,838 fact checks
Documented cases
A broadcaster’s article that never existed, July 2022
An image circulated purporting to be a Sky News article linking vaccines to the climate crisis. Sky News confirmed to Full Fact that no such article had ever been published.5
Full Fact also noted the tells that were visible in the image itself — incoherent prose, faulty punctuation, and a variant name that was simply wrong. Those helped, but they are corroboration, not proof. The thing that settled it was asking the publisher.
Real and fake, side by side, May 2024
When Google rolled out AI-generated search summaries in May 2024, some viral screenshots were genuine: a Google spokesperson confirmed the results advising glue on pizza and eating a small rock daily, which came from the system ingesting a joke forum comment.
Others were fabricated. The most widely shared — a screenshot claiming the feature had given dangerous advice about depression — was invented, and the person who made it said so. A major newspaper issued a correction after citing it as real.6 Users openly circulated instructions for making more. This case is unusually useful because the two categories travelled together, and because a professional newsroom got caught.
What to do about it
- Ask for the link. A real post has a permalink. "I saw a screenshot" is a report about an image, not about a post.
- Search the exact wording in quotation marks, and again with the outlet’s domain. Real headlines are indexed. Fabricated ones return only the screenshot’s own circulation.
- Check the archive. Paste the account or the URL into web.archive.org and look for the date in the image. This works well enough that it has been automated at high accuracy.
- Look for the debris. A genuinely deleted post leaves quote-posts, replies, and screenshots taken at different moments by different people. A post that only ever existed as one image has none of that.
- Ask the source. Outlets publish denials and corrections; people say "I never posted this." Sky News simply told Full Fact.
- Say "unverified", not "fake", until you have actually checked. They are different claims, and only one of them is honest before you have done the work.
Sources
Every figure on this page comes from one of these. Each entry names the publisher, the date and the sample size, so you can check it rather than take our word for it.
- Dufour, N. et al. — "AMMeBa: A Large-Scale Survey and Dataset of Media-Based Misinformation In-The-Wild"Google Research with Full Fact and others · arXiv:2405.11697 · May 2024 · 135,838 fact checks, 32,320 annotated pairshttps://arxiv.org/abs/2405.11697
- Hoq, A., Facciani, M. J. & Weninger, T. — "Feedback and education improve human detection of image manipulation on social media"Harvard Kennedy School Misinformation Review · 2 April 2025 · preregistered, n=268https://misinforeview.hks.harvard.edu/article/feedback-and-education-improve-human-detection-of-image-manipulation-on-social-media/
- Nightingale, S. J., Wade, K. A. & Watson, D. G. — "Can people identify original and manipulated photos of real-world scenes?"Cognitive Research: Principles and Implications 2, article 30 · 2017 · n=707 and n=659https://doi.org/10.1186/s41235-017-0067-2
- Content Authenticity Initiative — frequently asked questions on content provenanceContent Authenticity Initiative · a living page with no publication date shown · checked 2026-09-02 · states that Content Credentials do not prevent anyone from screenshotting an imagehttps://opensource.contentauthenticity.org/docs/getting-started/faqs/
- Benedictus, L. — "‘Sky News’ screenshot linking vaccines with climate change is a fake"Full Fact · 22 July 2022https://fullfact.org/online/sky-news-fake-screenshot-climate/
- Wrona, A. — "Fake Google AI Overview screenshots go viral after new feature’s rollout"Snopes · 29 May 2024https://www.snopes.com/news/2024/05/29/google-ai-feeling-depressed/
- Zaki, T., Nelson, M. L. & Weigle, M. C. — "Web archives for verifying attribution in Twitter screenshots"Proceedings of the 36th ACM Conference on Hypertext and Social Media · 2025 · n=4,504 screenshotshttps://arxiv.org/abs/2510.22939
What this page is not sure about
Every page here carries this section. A site that tells you to check its sources should be the first to say where its own evidence is thin.
- The detection studies on this page tested manipulated photographs, not text screenshots. They establish a ceiling on how well people do by looking. For screenshots the real figure is almost certainly worse, and we have not found a study that measures it directly.
- We wanted to publish a figure for what share of image-based fact checks concern fabricated screenshots specifically. The largest dataset reports this in its charts rather than its text, so we have described the finding and not invented a percentage from a graph.
- This page describes documented fabricated screenshots and links to the debunks. It does not reproduce any of them, and it never will — a convincing fake outlives the article debunking it the moment somebody crops the caption off.