How to check suspicious media

Before sharing a post, trusting a call, or sending money, check who it came from and whether the claim is supported. Use the steps below to investigate images, video, audio, and messages.

Aerial photograph of the Six Flags Over Georgia roller coaster surrounded by floodwater — Same photograph, different story. Reuters found that the flood image dates to 2009 and rated the Hurricane Helene claim miscaptioned. Read the fact check.

The same Six Flags flood photograph shown with its original September 2009 context — Same photograph, different story. Reuters found that the flood image dates to 2009 and rated the Hurricane Helene claim miscaptioned. Read the fact check.

Start with the source and the claim

1. Pause before acting

An urgent request can wait while you check who sent it. Do not share the post, send money, or give out personal information until you have checked the claim.

If someone says they are a relative or friend, call them on a number you already trust.

2. Find the original source

Open the full post and follow any credits to the original upload. Look for the complete recording and its first publication date.

Who first posted or recorded it?
Is there a full-length version?
Do independent sources support the claim?

3. Check the context

Compare the caption’s date and location with other reporting. Search a distinctive spoken sentence in quotation marks.

Example searches

The person’s name, place, and event
An exact phrase from the recording
The claim followed by “fact check”

4. Find where an image appeared before

For an image

Save the clearest copy or take a screenshot. Open the Google app, tap the Lens icon, select the image, and tap “Find image source.” Crop away app borders or captions if they hide the image. Look for the earliest credible use and its original description.

For a video

Use the InVID verification plugin to inspect a public video link and extract representative frames. Search several clear frames through Lens, especially ones with a landmark, sign, vehicle, or other distinctive detail.

InVID toolkit displaying a public Instagram video URL for verification — InVID helps turn a public video into searchable evidence. In the Google app, Lens can then look for pages that contain a selected image or frame.

Google Lens inside the Google app showing the Find image source control — InVID helps turn a public video into searchable evidence. In the Google app, Lens can then look for pages that contain a selected image or frame.

Content Credentials Verify showing a landscape issued by OpenAI, identified as AI-generated, with creation and format-conversion actions recorded — A real verified record. The file is issued by OpenAI, identified as AI-generated, and shows the actions recorded in its process.

5. Check the file’s Content Credentials

Content Credentials are tamper-evident records that can describe who or what signed a file, how it was made, and which edits were declared. They provide provenance; they are not a universal AI detector.

Save the media or copy its public URL.
Open Content Credentials Verify and upload the file or paste the URL.
Read the signer, creation method, edits, and validation status, then compare them with the claim.

“No credentials found” proves nothing by itself. Metadata may be absent, stripped by a platform, or never added. It does not establish whether the media is authentic.

Open Content Credentials Verify

6. Check for AI watermarks

Some AI tools add invisible watermarks or records identifying how a file was made. OpenAI Verify checks for supported OpenAI signals, while Gemini checks for Google’s SynthID watermark. A positive result can identify AI use; an absent signal does not prove the file is human-made.

OpenAI Verify

Upload the original file

Open the verifier and upload one image or audio file. The original download preserves more useful signals than a screenshot or converted copy.

OpenAI Verify reporting that a landscape was generated with OpenAI tools

OpenAI result

Read which signal was detected

This example reports that the file was generated with OpenAI tools. A positive result supports origin, but it does not establish the truth of a caption or claim.

Gemini reporting that a landscape contains signals indicating it was generated or edited with Google AI

Gemini and SynthID

Ask whether Google AI signals are present

Upload the file to Gemini and ask whether it was created or altered with Google AI. This example reports that Google AI signals were found.

Open OpenAI Verify Check with Gemini

7. Compare models in DeepFake-O-Meter.

DeepFake-O-Meter is a University at Buffalo research platform that can run image, video, or audio through several published detection models. Use it only on media you are permitted to share.

Open DeepFake-O-Meter and sign in to a free account.
Upload a file or provide a public YouTube URL, then select several applicable detectors.
Wait for each model and compare its result with the original source, the surrounding context, and any file history.

A detector score is evidence, not proof. Results can change with the model tested, compression, cropping, editing, video quality, or unfamiliar generation methods. A low signal does not clear a video; a high signal should be confirmed elsewhere.

Open DeepFake-O-Meter

DeepFake-O-Meter processing an uploaded landscape with four selected detection models — This real interface shows four selected models still processing. Wait for every selected model before comparing their results.

8. Verify the person on a separate channel.

If a call, voice message, video call, or urgent request claims to come from someone you know, stop the interaction. Contact that person using a phone number or account you already trust, not contact details provided in the suspicious message.

For high-risk situations, agree on a private family verification phrase. Do not rely on a familiar voice, face, caller ID, or speaking style alone.

Before you share or send money

Stop the action.Do not share, send money, reply, or accuse.
Save the claim.Note the link, account, names, place, event, and time.
Trace and compare.Find the original; search the context and a few keyframes.
Add technical evidence.Check credentials or a suitable detector when useful.
Confirm independently.Use an established source or a trusted channel for identity before acting.

Still unsure? Label it unverified and do not pass it on as fact.