AFP Fact-Checking Stylebook

Introduction

The following Stylebook provides editorial guidance on fact-checking at Agence France-Presse.

Our mission is to fact-check falsehoods, hoaxes and misleading claims that circulate widely and have a significant impact on news or public discourse.

We prioritise fact-checking false claims that could cause harm.

We must show balance, so for example we should not focus on one candidate, party, or website, unless they are consistent spreaders of potentially harmful disinformation.

Accuracy is paramount, as is clearly and transparently explaining how we carried out our investigation.

Errors must be immediately and transparently corrected.

It is the responsibility of our fact-checking team to be familiar with the rules set out in this Stylebook and AFP’s code of ethics (Editorial Standards and Best Practices) and 20 Principles of Sourcing.

We are bound by the IFCN code of principles and the EFCSN code of standards.

Download the stylebook here.

ANONYMOUS SOURCES

An anonymous source cannot be quoted as part of a fact-check investigation. If a person wishes to be quoted only by his or her function, (e.g., a spokesperson for the police), and they are providing essential information that cannot be obtained elsewhere, the decision to use the source must be made with heads of the fact-checking service or their deputies. Using identified sources is essential to the credibility of our fact-checking and allows readers to verify the quality of these sources for themselves if they wish to.

See Sources

ARTIFICIAL INTELLIGENCE

AI-generated content brings new challenges for fact-checkers as existing digital verification tools are largely incapable of detecting if an image has been produced by AI. This is because AI-generated images are recreated pixel-by-pixel from a combination of existing images drawn from a vast database. Software designed to detect if an image has been created by AI currently gives unreliable results. When shared online, these AI images also do not contain any metadata, another source of clues as to an image’s authenticity. As a result, for now the fact-checking team must rely on existing methods to detect images generated by such AI tools as Midjourney, DALL-E, Craiyon and Stable Diffusion. These techniques include reverse image searches to find when the image was first posted online, possibly its origin, and similar images from which it may be based. The fact-checking team uses visual clues that expose weaknesses in the way AI-generated images are created, such as out of proportion features or background, physical clues like a person having six fingers, lack of or inconsistent reflections and shadows, and watermarks that some AI companies add to their generated content. An article on the challenges involved in identifying AI-generated images can be found here.

ATTRIBUTION

All facts quoted in a report must have clear and identified attribution.

BIAS

Be aware of the danger of confirmation bias, when you have decided a post is false and as a result may overlook or disregard evidence to the contrary.

CONTEXT

It is essential to put context high in our stories and that it is precise and accurate, which can involve linking to an AFP story that has been published online or another reliable source. A fact-check in isolation may mislead readers about the actual situation the post claims to represent, so context is vital.

False posts may use a misleading or out of context image to illustrate something that has happened or is happening. For example, a claim that an image shows the aftermath of a bombing by warplanes of country A of a town in country B, where in fact the image comes from a different conflict. In this case, we can say that the post was false, but make it clear that other images online do show the results of bombing by country A of civilian targets in country B, if such images are available. By debunking the post, we should not give the impression we are also dismissing claims of real attacks on civilians, so it is essential to add that context.

For example: Warplanes from country A have bombarded towns and villages in country B, but this image dates from an earlier conflict in country C and does not show the aftermath of recent attacks.

CORRECTIONS

If we make a factual error in a publication, we correct it and add a note at the bottom of the publication that explains what has been changed and the date of the modification. A record of all corrections published in recent years is maintained on the site.

If a mistake is substantial, we take down the article and leave an explanation of the error and the circumstances that allowed it to occur. We also publish the correction on the same social media channels as the original, to ensure it is widely seen.

When an error has been made in a format that is not editable, such as certain social media posts, we delete the incorrect post and publish a correction in the same channel.

We also note at the bottom of our articles any updates, clarifications or new information that has been added after publication.

DENIALS

Do not only use a denial by the person, organization or other stakeholders as proof that a claim is false. We must provide other independent proof in our investigations.

DIS/MISINFORMATION/FAKE NEWS

These are terms that are used frequently in the media so here are some definitions:

Disinformation is created and designed to deliberately mislead people, often with a malicious intent. Misinformation is generally regarded as having been created unintentionally, although it can still have damaging effects. “Fake news” is a term that has been hijacked to mean everything from intentional disinformation to an epithet used by politicians and others to discredit unfavourable coverage. Therefore, we should avoid using the term “fake news” in our stories, except in quotes when necessary, and instead use dis/misinformation.

See Satire/Jokes

DUTY OF CARE/MINIMISING HARM

As a news agency we have a duty of care to the people we cover, and to minimize harm to them and by extension their families. In some circumstances we may choose to blur the faces or withhold the identity of people featured in a false post if they are wrongly being accused or being held up to ridicule. Likewise for names in the comments section. We should be particularly careful when the fact-check involves minors.

FUTURE EVENTS

Do not fact-check predictions or statements about an event that has not yet taken place. For example, if a politician, organization or government says that something is going to happen, the news side can investigate and seek confirmation or contradictory views. For example, prior to the war in Ukraine, we would not have fact-checked claims from the United States that Russia was poised to invade. In that case we would have had to quote Russian denials, which would not have been a fact-check.

We can make exceptions for science-based scenarios such as fake, doomsday posts predicting an asteroid is going to hit the earth, in which case we would quote astronomers.

GRAPHIC IMAGES

The guidelines on dealing with graphic images are given in the Stylebook and ethics document. They include this checklist on the use of a graphic image:

• Does it add important understanding to the story?
• Does the public have a right to know and see what happened?
• Does it meet your duty to inform as a journalist?
• Does it simply appeal to morbid curiosity without providing any essential context?
• Have you considered your duty of care, and ability to minimize harm, to the victim/s and their family?
• How will it be published to our audience? Will it be filtered by a client or published directly online?

Our criteria for the use of a graphic image contained in a fact-check are different from those of the photo and video services, whose responsibility is to show the reality of a disturbing scene. Our role is to fact-check online falsehoods, and we must be particularly careful because our fact-checks are published directly online.

HARASSMENT

Fact-checkers can be the target of online harassment and doxing, so it is important to take the necessary precautions and review our internal guidelines. Report any incidents to AFP management.

The Rory Peck Trust publishes guidelines on digital security that can be found here.

You can follow AFP's online course to help you to act more safely online and to anticipate cyber bullying.

HEADER PHOTOS

As much as possible, use the misleading or false image in the heading, clearly marked with the fact-check. Only use an AFP photo if the original content is too graphic or otherwise unsuitable to use in such a prominent position. If you do use an AFP photo, ensure that it is relevant to the story. Also, do not create confusion by using an AFP image of the real event as the header image in a story that debunks the use of a manipulated or fake image.

HEADLINES

Headlines must be as short as possible and state the fact clearly and stand independently of the header image. In other words, they should be able to stand alone.

For example, write: This 2014 image was taken in Paris and is not an anti-fracking protest in London.

Do not write: This is not an image of an anti-fracking protest in London.

Remember to include key words and proper names so that our headlines are search engine optimized.

HEALTH

Avoid formulations, including in headlines subheads, captions etc., that may be seen as an AFP endorsement of the efficacy or safety of a drug or vaccine.

We must quote respected medical sources and research to debunk the false claim.

Do not write: The vaccines were properly tested. or It’s false to say that they were not properly tested.

Instead use formulations such as: The WHO, FDA and other health organisations around the world said that trials proved the efficacy of the vaccine and deemed it safe enough to authorize use.

Similarly, we must avoid definitive formulations such as Vaccines do not cause XXXX. There could be a rare case somewhere of the adverse effect in question so we should attribute it to respected sources.

We can say: AFP has not found any study that says xxx despite searches on the databases of respected scientific publications. We should avoid definitive statements such as “there are no studies”, because bogus scientific and other research circulates online.

Give the date that you collected the evidence/quote on the topic you are investigating.

You can follow AFP's online course to help you to tackle health claims.

IMAGES

If you check an image and discover it is a composite, publish both images – A/ the one that is featured in the false post along with its undoctored original; B/ the other image from which elements have been taken.

IMPACT AND REACH

Explain the impact of the false news. This can include details of the extent of shares/views across all social media platforms and can also include the number of followers.

We should not use "widely-circulated" or "viral" without any figures to support it, or anything other than specific figures about post shares/views. We may rate content on Facebook, but do not focus your report just on Facebook and/or Twitter. Our stories should explain how the false news is polluting online as broadly as possible, including referencing closed messaging apps where relevant. There is no absolute number of shares to decide whether to do a story or not. 

One hundred shares may be insignificant in many countries but may be significant in a country, or in a local language, where there are strict controls on social media or where the social network ecosystem is different. It also depends on the languages: a post in English will logically have more shares than a post in Slovak or Catalan. Also, where we have multiple appearances of the same post but each one has few shares, we can then estimate that the accumulation makes it viral enough to treat it.

INVESTIGATIONS

Be clear about how an image was found and how the fact-check was conducted. Readers should be able to understand how it was done and even follow the same steps themselves. This is also important for members of the fact-checking team working in different languages, who will need to be able to follow the process if they are translating the story.

LENGTH

We should write clearly and concisely, but at the same time our fact-checks must be complete, so do not miss out important elements of your investigations to save two or three paragraphs of length. Also, context must be fully explained. People who do not wish to read the whole fact-check will usually just look at the headline and the header image (hence the importance of having a header image that, whenever possible, clearly shows the result of the fact-check and is not just a generic image).

LIGHT/OFFBEAT STORIES

Although our priority is to fact-check stories that have significant news value and may cause harm, we should also cover more light-hearted and offbeat stories. These are popular and have an educational value because people will see how easy it is to manipulate images and understand how we verified them. It may encourage them to hesitate before sharing outlandish posts in the future.

LONG VIDEOS/DOCUMENTARIES

If a long video or documentary is promoting a false premise, we can fact-check key elements as an illustration. Often the falsehoods are wrapped in a cocoon of verified facts, so the important thing is to show that there is sufficient disinformation to discredit the whole premise of the film.

NUMBER OF SHARES

Give a rounded number of shares, not the exact number (say more than 500 times, not 510 times). Also specify how long the content has been shared online for: it can sometimes be several years old but reappear due to current events. Use with care other metrics, such as view count of a video or number of comments/reactions, to illustrate the virality of a post. Users may not have watched a video entirely or may comment to say it is false.

OTHER FACT CHECKERS

We do not use other fact-checking organisations as primary sources. However, if another media outlet’s fact-check has contributed to our investigation, we must cite that fact-check. If you are led to evidence from another fact-check, you must still verify it yourself. Then clearly attribute it in your fact-check. You can also quote another organisation’s fact-check, for example if it came to the same conclusion using a different method of checking.

OTHER MEDIA

We should link to AFP content whenever possible. If AFP has not covered the story referenced in the fact-check, we can link to local media if they are regarded as a reliable source. We should avoid linking to content behind paywalls, which will be a source of frustration for readers who are not subscribed.

OPINIONS

We do not fact-check opinions.

PROPAGANDA

Propaganda is a term that is often used in the context of the spread of false information. It is a deliberate attempt by political and other actors to influence how people think and behave by using multiple methods including biased news reporting, television and radio broadcasts, social media posts, film, posters, music, memes, satire, advertisements, and other forms of communication. It may be based on fabrications or exaggerations, be biased and misleading, it is often designed to play on people’s fears and prejudices, but it is not necessarily false.

QUOTES/FALSE STATEMENTS

We must fact-check two kinds of false statements:

A/ Factually incorrect statements made by public figures.

B/ Statements attributed to public figures that are either fabricated or are misrepresentations of what they said.

In both cases, the statement we are checking should be attracting sufficient online attention and involve a prominent personality.

In the case of A, we must:

Establish that the person spoke in the context that was claimed, and then check the content of what they said.

Only fact-check statements when you can access the original version of what was said, e.g., a recording or an official transcript.  Do not base your fact-check on what was quoted in a report on other media or on social media.

In the case of B,

These fabricated or misrepresented quotes should be different enough from the person's public views and detailed enough to merit a fact-check.

If it was misrepresented, we must find the original quotation and show how the statement was edited or otherwise presented in a misleading way.

If it was fabricated, we should try to obtain a denial from the individual or a spokesperson and do our own investigation. What was the person doing at the time and place they were supposed to have spoken? Is there any trace of such a statement in legitimate media or on official sites? We should avoid overly affirmative sentences, such as “he/she never said that” and use phrases such as “There is no trace of such a statement on official sites, nor in the media, and his/her spokesperson has firmly denied it”.

REPEATING FALSE CLAIMS/POSTS

Our role is to debunk falsehoods, not to spread them, so we must take care how our fact-checks are presented. Whenever possible we must open with verified facts and follow it with the falsehood. When it is possible, the most efficient way of doing this is known as the truth sandwich - the truth, followed by the falsehood, followed by the truth.

For example, write: Paediatricians have warned parents not to try to make infant formula at home, contradicting social media posts that encourage parents facing shortages to make it for themselves. Paediatricians say homemade formula could lack key vitamins and nutrients and be harmful to babies.

Do not write: Social media posts claim parents facing baby formula shortages should make it themselves. But paediatricians told AFP they do not advise their patients to use homemade formula, warning it may lack vitamins and nutrients key to helping infants grow and thrive.

SATIRE/JOKES

Generally, only rank these if they have many reactions and shares. If the claim is clearly ridiculous, be careful not to give the impression that AFP is taking it seriously. Read the comments and reactions to see if it is being shared purely for fun or if enough people are taking it seriously to merit a fact-check. We also must take into consideration that what may appear blatantly ridiculous in one culture may not be so obvious in another. Satire, jokes and memes are an effective way of spreading disinformation and racial and other slurs, in which case a fact-check may be merited.

SCREENGRABS

Screengrabs should show the essential content of the post and should not be cropped in a way that omits important information such as the number of shares and engagements.

SOURCES

Sources should be clearly identified by name, age if relevant, title, occupation, affiliation, credentials, and any other factors that strengthen the credibility of the story. If we are quoting them as experts, we should be specific about their credentials and can link to relevant websites. We should have at least two independent sources in each fact-check, in addition to any denials from the interested parties.

We should contact our sources directly but be aware that contacting known peddlers of disinformation or conspiracy theorists can result in harassment and doxing. Whether or not to contact them should be carefully considered. For guidance on sourcing, consult AFP’s 20 Principles of Sourcing.

TERMS

We should use the following terms in fact-checks:

• False -- We state an item is false when multiple and reliable sources disprove it.
• True -- We state an item is true when multiple and reliable sources have confirmed the information to be authentic.
• Misleading. -- We state an item is misleading when it uses genuine information (text, photo or video), taken out of context or mixed with false context.
• Altered Photo -- When a photo has been manipulated to deceive.
• Altered Video -- When a video has been manipulated to deceive.
• Missing Context -- When a claim is authentic but might be deceiving without further information.
• Satire -- When a claim is false and has the potential to fool, but it has no intention to deceive (e.g., humour, parody).

TRANSLATIONS

Each translation must be adapted to the local and/or regional context, and enriched, if necessary, by other sources and other background elements, or other elements that reinforce the demonstration of the falsehood of the content.

WRITING

Our writing style should be straightforward and factual and does not need to be embellished with colourful language. Avoid puns and irony, which may not translate into different cultures and languages.

Every fact should not only be true, but must be shown to be true, so you must transparently report your investigation by:

• Providing evidence for all the facts that you state. Links, embedded posts/videos, street maps, etc.  
• Explaining the steps that you took to conduct your investigation. For example, explain that you used Google reverse image search, INVID, etc.  
• Write as concisely as possible. Every extra claim needs to be backed up and proved transparently.  
• When providing a link to a reliable source such as a story or piece of evidence that proves a claim is false, include the active link and add the archive link in brackets.   Conserving evidence is essential to the transparency of our work. For false posts, we don’t want to increase traffic to suspect websites so for those just put the archive link. Archive links using sites such as perma.cc and Wayback Machine.