AI Media FAQ

You have a file, a deadline, and one awkward question: “What can I honestly say about this?” This FAQ is for that moment. It does not promise a universal AI detector, because that would be the wrong promise. It explains what Check AI Media can read, what it refuses to guess, and where you should go when a question belongs to an official tool instead of an independent site.

Can you prove a file is not AI?

No. A missing signal does not prove human authorship. It only means the check did not find the kind of readable signal it was built to look for. That difference is easy to overlook, especially when a result page feels official or technical. A file may have been made with AI and exported without credentials. It may have carried metadata once and lost it during a repost. It may be a screenshot of something that originally had provenance information. None of those cases can be turned into “not AI.”

The strongest honest statements are specific. You can say a file carries Content Credentials if a check finds them. You can say a file includes certain metadata fields if they are present. You can say no readable signal was found in a particular check. You should not say the file is human-made just because the result is quiet. If that boundary is new to you, the method page explains how we separate readable evidence from assumptions.

Do you upload my file?

The current browser-based tools run on your device. That means the file does not need to be sent to our server for the EXIF or Content Credentials checks now available on the site. This is important because people often check unpublished work, sensitive client files, or images that include private people and places. A check should not create a new privacy problem.

There is a practical reason to read the page you are using, though. Different checks can require different handling. If a future feature needs server processing, the page should say so before you choose a file, and the upload should be clearly labelled rather than hidden behind a vague button. The plain-language version of our approach is on the privacy page.

Do you remove watermarks or credentials?

No. We do not remove, replace, hide, or weaken watermarks, labels, Content Credentials, or other provenance information. We also do not recommend “remover” sites. A tool that helps conceal provenance can make it harder for viewers, buyers, editors, and platforms to understand where media came from. That is the opposite of the purpose of this site.

If you are trying to prepare an image for publication and a visible label is creating layout problems, the right next step is not to erase context. Ask for a proper export, keep a copy of the original, and document what changed. If you are checking whether a file carries credentials, use the C2PA checker and preserve the source file as received.

Can you check SynthID?

Not here. SynthID is Google’s technology, and we do not claim to read it ourselves. As of this writing, the correct path for a SynthID-specific answer is Google’s official SynthID Detector. Google has not published a public detection API that lets an independent website make the same claim with the same authority. Any site that says it can definitely detect or remove SynthID should be treated with caution unless it can show a credible official basis.

We still write about SynthID because people search for it when they are really asking broader questions about AI provenance. That creates confusion. The SynthID detector guide explains the difference between Google’s official check, general metadata clues, and ordinary file signals that are not SynthID at all.

What does “signals found” mean?

It means the file contains a clue this site can read or point to. That clue might be a metadata field, a credential record, or another checkable indication. It is not a moral judgment about the creator, and it is not a complete history of the file. “Signals found” should make you slow down and read the details, not stop thinking.

There are also weaker signals that require context. A file name can mention AI. A visible label can appear in a corner. An export field may name software. Those details can be useful, but they are not all equal. A signed credential deserves different treatment from a casual note in metadata. When you share a result with someone else, include the kind of signal found rather than flattening everything into “AI detected.”

Is this site Google?

No. Check AI Media is not Google and is not affiliated with Google. We mention Google when the topic requires it, especially for SynthID, because sending people to the correct official source is more honest than pretending we can perform that check ourselves. SynthID is a Google trademark, and the name is used here only to identify the technology people are asking about.

For a compact statement of non-affiliation and result limits, read the disclaimer. For most everyday checks, the practical rule is simple: keep the original file, use the right checker for the signal you care about, and describe the result in words that match the evidence.