PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteiTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
There is no single visual, audio, or detector test that proves a file was made by AI. The most reliable approach combines the file’s source and surrounding context with provenance records, provider-specific watermark checks, and independent confirmation of the claim the media appears to show. Treat every result as evidence about a particular signal—not as a universal verdict on authenticity, truth, ownership, or intent.
This guide explains a repeatable workflow for images, video, and audio, what Content Credentials and invisible watermarks can establish, why a negative check proves little, and how to report uncertainty responsibly.
What you can—and cannot—establish
Recognition is an evidence problem, not a visual puzzle
Modern generative systems can produce media without the obvious defects people once associated with synthesis. A hand may look normal, a face may have consistent lighting, and a voice may sound entirely natural. Conversely, a genuine camera recording can contain blur, compression blocks, stitching errors, or unusual reflections. Those details can justify a closer check, but none is a dependable universal test.
Recommended Free Tools
Start by asking two separate questions:
- What is the file’s origin and editing history? Provenance tools may record a signer, software, creation event, or subsequent edits.
- Is the event or statement depicted actually true? That requires contextual and independent verification. A file can have a genuine camera origin and still misrepresent what happened; an AI-generated illustration can accurately depict a real place or event.
Keeping these questions separate prevents a common mistake: treating a provenance result as a truth certificate.
#1 Best Overall
Why a missing signal is inconclusive
A file may have been created with an unsupported model, exported through a path that does not preserve metadata, altered after creation, or made before a particular watermark system existed. Screenshots, reposts, crops, transcoding, and social-platform processing can remove or damage records. Therefore, “not detected” means only that the tested signal was not found in the submitted file.
A practical verification workflow
1. Preserve the best copy you can obtain
Save the original download or attachment instead of relying on a screenshot or screen recording. Keep the filename, extension, and any accompanying text. Record where and when you found it, the account that posted it, and the URL or message in which it appeared. If several versions circulate, retain each one and note which is earliest.
Do not edit the file before checking it. Resizing, recompressing, converting formats, or trimming a clip can remove metadata and weaken watermark detection. If you need a working copy for analysis, keep an untouched original alongside it.
2. Examine the source and surrounding context
Assess the account that published the media. Is it an established source with a track record, or an anonymous account created recently? Does the post identify a location, date, photographer, or witness? Are there independent reports, photographs, livestreams, weather records, maps, or public statements that fit the claimed event?
Look for internal inconsistencies: a supposed live event posted hours before it occurred, shadows that conflict with the stated time, a location whose signs or architecture do not match, or a caption that changes between reposts. These are prompts for verification, not proof of synthesis. Reverse-searching a still frame or searching distinctive phrases can help locate an earlier upload, but an earlier upload is not automatically the original.
3. Inspect Content Credentials or C2PA history
Content Credentials, built on the C2PA standard, can carry signed information about where a file came from and how it was edited. A record may identify a creator or organization, the application used, and a sequence of recorded actions. Credentials can accompany both AI-assisted and entirely non-AI media.
When a credential is available, check:
- the signer and whether that organization is recognizable and appropriate to the claim;
- the named tool or camera and the recorded creation time;
- the edit sequence, including whether generative operations are listed;
- whether the signature validates and whether the record belongs to the exact file you received.
An intact, validated record can provide useful history. It does not prove that the depicted event happened, that the signer is honest, or that the file contains no unrecorded copy or edit. An absent, broken, or unsupported credential is inconclusive rather than evidence of a camera original.
4. Run a provider-specific verification check
Use a verifier that matches the suspected origin and media type. These checks look for particular embedded signals; they do not classify every possible AI model.
Google Gemini verification
Google says Gemini can check supported images, video, and audio for its SynthID watermark and for compatible Content Credentials. The current help guidance describes one file per check, with a maximum file size of 100 MB. Videos must be shorter than 90 seconds, and audio must be shorter than one hour. Google also describes an approximate rolling daily quota of 10 image checks, 10 video checks totaling up to five minutes, and 10 audio checks totaling up to three hours. These operational limits can change.
A positive SynthID result supports the inference that a supported Google AI system contributed the detected signal. It does not establish that the output is accurate, lawful, owned by Google, or presented in its original context. A negative result means only that Gemini did not find the supported signal in that file.
OpenAI verification
OpenAI describes a checker for supported image and audio files that looks for supported OpenAI signals, such as SynthID or a trusted C2PA manifest. A detected signal indicates likely generation by an OpenAI model associated with that signal. No signal means that no supported OpenAI signal was found; it does not rule out generation by another provider, an unsupported model, or a file that lost its metadata during export or editing.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesCoverage can vary by product, model, export path, and file type. Before relying on a result, confirm that the submitted format and media type are supported by the current checker.
5. Verify the claim independently
Provenance does not answer whether the scene, quotation, or allegation is true. Search for the earliest available source and corroboration from independent, reliable outlets. Compare geography, weather, shadows, landmarks, accents, and timelines. For a purported news event, look for reporting from organizations that were not simply copying the same post. For a voice message, seek a direct statement from the person or institution supposedly speaking.
The higher the consequence—financial instructions, public-safety warnings, political claims, or allegations against an individual—the more independent confirmation you should require before sharing or acting.
6. Describe exactly what you found
Report the signal and its scope: “The file contains a validated credential naming a generative editing step,” or “The Google check found a SynthID signal.” If no signal was found, say “This check did not find the supported signal,” not “This is real.” Explain which file was tested and whether it had been reposted or converted. Avoid categorical claims based only on a detector score, an odd-looking hand, a voice impression, or missing metadata.
What each signal can tell you
| Signal | Useful for | Important limits |
|---|---|---|
| Content Credentials / C2PA | Signed information about origin, signer, tools, and recorded edit history when present and valid. | Records may be absent, unsupported, invalid, or stripped. They do not prove the depicted claim is true or that a file is categorically AI-generated. |
| Invisible watermark, such as SynthID | Detecting a supported provider’s embedded signal in a compatible file. | Provider and format coverage is limited. Cropping, recompression, or other alterations can defeat detection. A negative result is inconclusive. |
| Platform label or creator disclosure | A visible notice that a platform or poster identified synthetic or materially altered content. | Depends on honest disclosure and the platform’s detection coverage. Labels may not appear on every copy. |
| Visual or audio inspection | Finding details that justify further checking and comparing the file with its context. | Human perception is fallible. Strange details are clues, not proof, and realistic synthesis may have no obvious artifact. |
| General-purpose detector | Possibly a triage signal when its method, training data, media type, and test conditions are documented. | No universal current accuracy figure is established for all images, video, and audio, especially after common edits or recompression. Do not treat a score as a final verdict. |
Media-specific clues to investigate
Images
Zooming in can reveal mismatched text, impossible reflections, repeated textures, inconsistent jewelry, or lighting that does not agree across objects. Use these observations to form questions: Is the sign legible in another photograph? Does the building exist at the stated address? Do shadows match the reported time and sun direction? A genuine photograph can also contain lens distortion, motion blur, or a damaged sign, so visual oddities alone cannot classify it.
Compare the highest-resolution version available with earlier uploads. A screenshot may hide or erase both evidence of editing and evidence of provenance. If a credential viewer identifies a generative operation, preserve a copy of the credential report with the original file.
Video
Inspect continuity across frames: faces, hands, logos, text, reflections, and background objects may change subtly. Check whether audio matches mouth movement and whether camera motion fits the claimed device or location. Then verify the timeline and location independently. A short, heavily compressed repost is particularly difficult to authenticate because it may have lost metadata and detail.
For a clip presented as live footage, find the earliest upload and compare multiple angles or contemporaneous reports. A provider check can identify a supported watermark, but it cannot establish that the scene was recorded at the time or place claimed.
Audio
Listen for abrupt changes in room tone, breath placement, pronunciation, emotional timing, or background noise. Compare the speaker’s wording and vocal characteristics with authenticated recordings, while remembering that microphones, codecs, illness, distance, and stress can change a real person’s voice. A voice impression is a reason to seek confirmation, not proof of cloning.
Check the original container rather than a clip copied from a messaging app. If the file is longer than a verifier’s limit, obtain a supported original or document that the check could not be run; do not infer authenticity from the inability to submit it.
How to interpret positive, negative, and conflicting results
A positive provider watermark
Read it narrowly: the tested file contains a signal associated with that provider’s supported tools. Preserve the result, the file hash or filename if available, and the date of the check. Do not extend the finding to claims about who prompted the model, why the file was made, whether every frame or sample was generated, or whether the depicted content is true.
A valid credential with no AI operation listed
This may document a camera capture or conventional editing workflow, but it is still provenance information rather than a truth certificate. Confirm that the signer and chain are trustworthy and that you are examining the original signed asset.
Free tools Windows power users keep installed
One-click scans. No signup required.
No watermark or no credential
Record the exact wording of the result and the file conditions. The absence may reflect an unsupported tool, an older file, a conversion, metadata removal, or an alteration that defeated detection. Continue with source and context checks.
Best Value
Conflicting signals
Preserve all versions and test the earliest, highest-quality file separately. A repost may have lost a credential while a creator’s original retains one. If a credential says one tool was used but a provider check finds a different signal, treat the discrepancy as an unresolved provenance question and seek the original uploader or independent corroboration.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Platform labels and disclosures
Meta has described using C2PA and IPTC information together with creator disclosures to label some AI-generated or altered media. Its published guidance also says invisible markers can be stripped and that it is not yet possible to identify all AI content. Treat a platform label as a useful notice about what the platform or creator identified, not as a complete inventory of every synthetic file. Platform policies and detection coverage can change, so check the current policy when a label matters to a decision.
What the current evidence says about detector accuracy
Official guidance from the major providers explains their own watermark and credential systems, but it does not establish a comparable independent benchmark with false-positive and false-negative rates for general-purpose image, video, and audio detectors after ordinary edits, resizing, recompression, or reposting. There is therefore no defensible universal accuracy percentage or overall ranking to quote.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
A detector may still help triage a large collection if you know what it was trained to recognize and how its output was validated. Keep its result separate from provenance evidence, and escalate consequential cases to human review and independent reporting rather than relying on a single score.
Google’s deployment figure, correctly understood
Google says its SynthID technology has watermarked more than 100 billion images and videos and 60,000 years of audio generated by company systems. This is a Google-reported deployment volume, not a measurement of detector accuracy, coverage across the wider internet, or independent verification of each item.
EU transparency rules from 2 August 2026
The European Commission says Article 50 transparency obligations under the EU AI Act apply from 2 August 2026. The related Code of Practice is a voluntary compliance tool, while the underlying Article 50 transparency requirements are legal obligations. The framework includes duties for providers marking certain AI-generated outputs and for deployers labeling specified deepfakes and AI-generated public-interest text.
These are EU rules with defined roles and contexts, not a worldwide requirement that every synthetic image, video, or audio file carry a universal label. Whether a particular obligation applies depends on the provider, deployer, service, content, and location. This article is not legal advice; consult the current Commission guidance or qualified counsel for a specific case.
A decision guide for common situations
| Situation | Best next action | What not to conclude |
|---|---|---|
| You have the original file and suspect a provider’s tool. | Inspect credentials, then run that provider’s supported verifier and document the result. | A negative check does not prove camera origin. |
| You have only a screenshot or repost. | Find the earliest, highest-quality copy and preserve the current one for comparison. | Missing metadata is not evidence of fabrication. |
| A dramatic clip claims to show a current event. | Verify date, place, and event through independent reporting and other footage. | A watermark result does not prove the event occurred. |
| A platform displays an AI label. | Read the platform’s explanation and retain the labeled copy. | The label does not explain every edit or establish motive and truth. |
| A detector gives a high score but no provenance evidence exists. | Use the score only for triage and seek the source, credentials, and corroboration. | The score is not a universal probability that the file is AI-generated. |
Checklist before you share or act
- Save the original file and note where, when, and from whom you received it.
- Locate the earliest available upload and preserve reposts separately.
- Assess the account, caption, timeline, location, and independent corroboration.
- Inspect Content Credentials or C2PA history and validate the signer and chain.
- Run a relevant provider check only if the file type, size, and duration are supported.
- Record whether a signal was found, not found, or could not be tested.
- Verify the underlying claim through independent sources, especially for high-consequence content.
- Describe uncertainty precisely instead of calling the file simply “real” or “fake.”
The safest conclusion is often narrow: a particular provider signal was found, a signed edit history is present, or no supported signal was detected. That level of precision protects readers from both synthetic media and overconfident debunking.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
