MEDIABIAS.fyi

Check a claim before you share it

Tactic · Decontextualization

Recycled imagery

A genuine photograph from another place or another year, relabelled. Nothing about the image is fake, which is why looking harder at it never helps.

Where and when was this picture actually taken?

What it looks like

A real photograph carries no internal evidence of when or where it was taken. Every pixel is authentic. Only the caption is false.

That defeats the two things readers instinctively do — look harder at the image, and ask whether it looks fake — because both interrogate the wrong object. The claim being made is about provenance, and provenance is not inside the file.

Why it works

This is not the exotic case. It is the ordinary one. The largest survey of image-based misinformation to date, built by Google Research with the fact-checking organization Full Fact, drew on 135,838 English-language fact checks and hand-annotated 32,320 image-and-claim pairs collected between November 2021 and November 2023. Its finding is blunt: context manipulations dominate, and have for every period with enough data to say.1 The authors note these "often do not entail any manipulation to the image pixels themselves," which is exactly why they attract less attention than deepfakes while being far more common.

A separate dataset built from African and South Asian fact-checkers reached the same conclusion independently: across 1,676 fact-checked images from 2019 to 2023, out-of-context imagery "consistently remains the most popular type of visual misinformation."2

The counter is reverse image search, and here the research is humbling. A study of 880 people found that a media literacy intervention significantly raised participants’ intention to reverse-search — and produced no significant improvement in whether they actually told genuine images from misattributed ones.5 Knowing what to do is not the bottleneck.

59%

of COVID-era misinformation was reconfigured genuine material rather than fabrication. "False context" alone was 24 percent; wholly fabricated content was 38 percent.Brennen, Simon, Howard & Nielsen, Reuters Institute, April 2020, n=225 fact-checked items

87%

of social media engagement with that sample went to the reconfigured material, against 12 percent for the fabricated.Brennen et al., Reuters Institute, April 2020

1,676

fact-checked images across three African and South Asian fact-checkers, 2019–2023, in which out-of-context imagery was the most common category throughout.Tonglet, Moens & Gurevych, EMNLP 2024

no change

in actual accuracy at telling genuine from misattributed images after a literacy intervention, despite a significant rise in stated intention to check.Qian, Shen & Zhang, Journal of Computer-Mediated Communication, 2022, n=880

Documented cases

Nepal, April 2015

After the Nepal earthquake, a photograph of a small boy with his arm around a crying girl circulated as two orphaned Nepalese siblings. The photographer, Na Son Nguyen, identified it as his own picture of two Hmong children taken in Ha Giang province, Vietnam, in 2007 — a three-year-old comforting his two-year-old sister, who was crying because a stranger had approached.6

Nguyen said it had been shared by hundreds of thousands of people and had previously circulated as an image of Syrian war victims. Nothing in the photograph was altered at any point.

Türkiye and Syria, February 2023

Footage said to show a Turkish nuclear plant exploding during the 6 February 2023 earthquake was the Beirut port explosion of 4 August 2020. Full Fact matched it against Sky News broadcast footage from 2020 and noted that Türkiye had no operating nuclear plant at the time.7

The same Beirut footage had already been recirculated with false claims about Ukraine. The recycling is itself recurrent: a dramatic clip does not retire, it gets recaptioned.

Move 3 — Find the original

What to do about it

  • Reverse image search it. Right-click and search by image, or upload it to a search tool. You are checking where else it has appeared, not what it looks like.
  • Sort by oldest, not by relevance. You want the earliest appearance, not the most popular one.
  • Crop out overlaid text and watermarks before searching. They interfere with matching, and the people recaptioning images know that.
  • Run more than one engine. Coverage is uneven and results degrade badly on small or cropped images.
  • No result is not confirmation. It means the image is not indexed. Look for a photographer or agency credit instead, and check whether any wire service is carrying it.
  • Check whether someone has already done this. Full Fact, AFP Fact Check and Snopes log these, and a disaster image that has travelled has usually already been checked.

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.

  1. Dufour, N. et al. — "AMMeBa: A Large-Scale Survey and Dataset of Media-Based Misinformation In-The-Wild"Google Research with Full Fact, Factly, Duke Reporters’ Lab and Maldita.es · arXiv:2405.11697 · May 2024 · 135,838 fact checks, 32,320 annotated image–claim pairshttps://arxiv.org/abs/2405.11697
  2. Tonglet, J., Moens, M.-F. & Gurevych, I. — "Image, Tell me your story! Predicting the original meta-context of visual misinformation"EMNLP 2024, pp. 7845–7864 · n=1,676 fact-checked images, 2019–2023https://aclanthology.org/2024.emnlp-main.448/
  3. Brennen, J. S., Simon, F. M., Howard, P. N. & Nielsen, R. K. — "Types, Sources, and Claims of COVID-19 Misinformation"Reuters Institute for the Study of Journalism · April 2020 · n=225 items, January–March 2020https://reutersinstitute.politics.ox.ac.uk/sites/default/files/2020-04/Brennen%20-%20COVID%2019%20Misinformation%20FINAL%20(3).pdf
  4. Toler, A. — "Guide To Using Reverse Image Search For Investigations"Bellingcat · 26 December 2019https://www.bellingcat.com/resources/how-tos/2019/12/26/guide-to-using-reverse-image-search-for-investigations/
  5. Qian, S., Shen, C. & Zhang, J. — "Fighting cheapfakes: using a digital media literacy intervention to motivate reverse search of out-of-context visual misinformation"Journal of Computer-Mediated Communication 28(1) · 2022 · n=880https://academic.oup.com/jcmc/article/28/1/zmac024/6839987
  6. Snopes — "Nepal Earthquake Photo", rated FalseSnopes · 5 May 2015https://www.snopes.com/fact-check/nepal-earthquake-photo/
  7. Thompson, T. — "Video does not show Turkish nuclear plant exploding due to earthquake"Full Fact · 7 February 2023https://fullfact.org/online/turkey-earthquake-explosion-beirut/

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 59 percent figure is from a single topic (COVID-19) in the first three months of 2020. It is the clearest quantified breakdown we could find, and it is five years old. Use it as evidence about the ratio, not as a current measurement.
  • The large survey behind this page reports the dominance of context manipulation as a time series rather than as one headline number, so we have quoted the finding rather than inventing a percentage from a chart.
  • The reverse image search guidance we relied on is from 2019. The failure modes it documents still hold; its specific rankings of which engine is best almost certainly do not.