How Can You Tell if a Photo, Video or Post Is an AI Fake?

There is no single reliable test. A practical approach is to combine several checks: look for provenance data such as Content Credentials, trace where the image or clip first appeared, see whether fact-checkers have already covered it, and treat any AI-detector result as a hint rather than a verdict. Looking for visual glitches alone is no longer enough.

1. Trace the source with a reverse image search

Start by finding out where the picture came from. Google lets you search with an image by uploading a file, dragging it into the search box, pasting an image link, or, in Chrome, right-clicking and choosing “Search with Google Lens”. Google’s “About this image” feature shows when the image and similar images may have first been indexed by Google, where it may have first appeared, and where else it has been seen online, such as on news, social or fact-checking sites. This is a clue, not proof of when an image was created. An image that claims to show a breaking event but was indexed years earlier, or that only appears on obscure pages, deserves suspicion.

2. Check provenance and Content Credentials

The C2PA (Coalition for Content Provenance and Authenticity) provides “an open technical standard for publishers, creators and consumers to establish the origin and edits of digital content.” Its Content Credentials work like “a nutrition label for digital content, giving a peek at the content’s history.” According to the Content Authenticity Initiative, they record who created the content, when, which tools were used and what edits were applied, and they are “tamper-evident.” A credential records an image’s history; it does not prove that what the image shows is true.

The catch is that credentials only help when they are present. Meta, which reads C2PA and IPTC signals to label AI images, acknowledges that “it’s not yet possible to identify all AI-generated content” and that “there are ways that people can strip out invisible markers.” A missing credential therefore proves nothing, while a present one is a useful positive signal. Note that Meta’s statement dates from February 2024 and may not describe its current systems.

3. Look for watermarks such as SynthID

Google’s SynthID is “a tool to watermark and identify content generated through AI.” Google says you can upload content to Gemini, Search or Chrome and ask whether it was created or altered by Google AI. Its watermarks are described as imperceptible to humans and designed to remain detectable after some changes such as cropping, filters and lossy compression, though not every edit. This only tells you about content made with tools that apply the watermark. It is not a universal AI detector.

4. Be careful with AI-detector tools

Automated detectors are tempting but fallible. The Reuters Institute reviewed how they work and where they fail. Its key points:

• Results “require informed interpretation, and this can easily mislead users.”
• “Even when an AI-detection tool does not identify any signs of AI, this does not necessarily mean the content is not synthetic.”
• “The act of compressing, cropping or resizing a file can affect its quality and resolution and impact the detectors.”
• “Detection tools should not be considered a one-stop solution and must be used with caution.”

A detector can be a starting point alongside other open-source verification techniques, but never the only evidence. NIST has also published a report on synthetic content covering provenance tracking, watermark labeling and detection as separate approaches; it does not describe any single method as foolproof.

5. Check the context and search the fact-checkers

Ask who posted it, whether a credible outlet reports the same thing, and whether the claim fits what is otherwise known. Then search for existing debunks; a missing fact-check does not mean an image is real. Google’s Fact Check Tools include the Fact Check Explorer, and Google’s documentation says its API queries “the same set of Fact Check results available via the Fact Check Explorer tool.” Fact-checking organizations such as Full Fact publish their findings openly.

6. Health misinformation: deepfaked doctors

Real doctors’ faces and voices are being used to sell products. A December 2025 Full Fact investigation found deepfaked videos of real academics and doctors on social media promoting supplements, including products linked to Wellness Nest. Wellness Nest said affiliate accounts were “100% unaffiliated” and that it “has never used AI-generated content, never paid celebrities.” TikTok acknowledged a “moderating error” and banned one account, and YouTube added “altered or synthetic content” labels to flagged videos. These are the companies’ statements to Full Fact; the investigation does not independently establish who ran the accounts.

Practical checks: be wary of a well-known doctor suddenly selling a supplement; look for the same clip on the expert’s official channels or institution page; check whether the account is verified and long established; and do not buy from a link in a video. Discuss any real health decision with your own clinician or pharmacist.

7. AI “engagement bait” on Facebook

Some AI-generated images are posted to attract engagement. Researchers Renee DiResta and Josh Goldstein found that Facebook’s feed sometimes recommended unlabeled AI-generated images to users who neither followed the Pages posting them nor realized the images were AI-generated. Their study does not show how common this is across all Pages. If a post looks like an engagement trap, do not click through to unfamiliar links, and report the page.

8. Non-consensual deepfake porn: where to get help

If someone has made or shared sexual images of you, the fault is the abuser’s, and there are tools that help.

• StopNCII.org is for intimate images of adults (check its FAQ for eligibility, including deepfakes of you). It generates a hash (a digital fingerprint) of the images on your own device. The site states that your content “will not be shared, it will remain on your device.” Participating platforms can then search for matching hashes and act on violations of their policies, and you receive a case number to check progress or withdraw. It works with partner platforms, not the whole internet, and does not guarantee removal of every edited copy.
• NCMEC’s Take It Down is for nude, partially nude or sexually explicit images or videos that were made when the person was under 18, even if they are an adult now; it is not a general service for adult victims. Users can remain anonymous. It assigns a hash value, and online platforms use it to detect the content on their public or unencrypted services and take action to remove it.

Also save the URLs and account names where the content appears as evidence, and report the content directly to the platform. Do not download or share sexual images of a minor, even to report them; use the services above instead.

A quick routine

• Pause before sharing anything that triggers a strong emotion.
• Reverse-search the image or video frames.
• Look for Content Credentials or a platform AI label, remembering their absence means little.
• Search for fact-checks and other credible coverage.
• Use detector tools only as one input.