Checking whether something is AI
Detectors, labels, research and trackers, by type of media.
Each entry says what it does, who runs it, and what it costs. Nothing here is ranked. Compiled on September 27, 2026, with every link checked that day; tools in this area appear, change and shut down often.
Labels the tools themselves attach
Some companies mark what their tools make, with an invisible watermark or with a record attached to the file. These checkers read those marks. Each one can only find the marks that the tools behind it add.
- Content Credentials Verify
Shows any Content Credentials attached to a file: an industry-standard record (C2PA) of which camera, phone or software made it and what edits followed. Some cameras, phones, news organizations and AI tools attach them.
Content Authenticity Initiative (Adobe-led) · FreeImage · Video · Audio - Gemini app: “Was this made with Google AI?”
Upload a file and ask. It checks for SynthID, the watermark Google puts in images, video and audio made with its own AI tools. Google says Content Credentials checking is rolling out in the same place.
Google · Free with a Google account · announcement, May 19, 2026Image · Video · Audio - OpenAI: Verify OpenAI-generated content
Upload an image or audio file to check it for Content Credentials and SynthID from OpenAI tools such as ChatGPT.
OpenAI · FreeImage · Audio - ElevenLabs Audio Detector
Checks audio for the SynthID watermark ElevenLabs has added to its generated speech since June 2026, then falls back to a classifier. Covers ElevenLabs audio only.
ElevenLabs · Free with an account · older classifier, no accountAudio
Detectors that analyze the file
These look at the file itself for signs that it was generated or edited. Most return a likelihood score rather than a yes or no.
- DeepFake-o-Meter
Runs an upload through several published research detectors at once and shows each one’s score side by side. The code is open source.
University at Buffalo Media Forensics Lab · Free with an accountImage · Video · Audio - Hive Detect
Upload a file or paste a link. Also available as a Chrome extension and as a bot on X.
Hive (company) · Free web tools, no accountImage · Video · Audio - Sightengine
Drag-and-drop check for AI-generated images; video checks through its API.
Sightengine (company) · Free monthly credits, more with a free accountImage · Video - AI or Not
Checks images, video and audio.
AI or Not (company) · Pricing on their siteImage · Video · Audio - InVID-WeVerify verification plugin
A browser extension built for journalists: splits video into keyframes for reverse searching, reads metadata, magnifies details, and, for registered users, includes a synthetic-image detector.
EU-funded research projects (InVID, WeVerify, vera.ai) · Free · ChromeImage · Video - Hiya Deepfake Voice Detector
Chrome extension that listens to audio or video playing in a tab and scores whether the voice is synthetic.
Hiya (company) · Free with an accountAudio - Resemble Detect, Press Access
Audio, video and image detection for journalists and editors at press outlets: 90 scans a month, shared by up to three people.
Resemble AI (company) · Free for qualifying pressImage · Video · Audio - Pangram
Paste text or upload documents. Works in many languages.
Pangram Labs (company) · 20 free checks a day with an accountText - GPTZero
Paste text or upload a file. Also a Chrome extension and a Google Docs add-on.
GPTZero (company) · Free up to 10,000 characters, more with a free accountText
Where something came from
Not AI detectors. These trace an image or video back to earlier copies, or show what is inside the file.
- TinEye
Reverse image search: finds earlier and other copies of an image online.
Idée (company) · FreeImage · Video - Google Lens
Search with an image to see where else it, or something like it, appears.
Google · FreeImage · Video - Forensically
Photo forensics in the browser: error level analysis, clone detection, metadata.
Jonas Wagner (independent developer) · Free · Runs in your browserImage - FotoForensics
Error level analysis and a metadata viewer for photos.
Hacker Factor · FreeImage - Bellingcat Online Investigation Toolkit
Bellingcat’s catalogue of open-source investigation tools, with notes on what each does.
Bellingcat (nonprofit) · FreeAll media
How detectors have tested
Independent studies, newest first. Several report AUROC, a score on which 50 is a coin flip and 100 is perfect.
- DF26: We Cannot Tell Fake From Real Anymore
2,691 videos of one person speaking to camera, 2,420 of them made by seven recent video generators. Nine current research detectors scored 44.0 to 69.7 AUROC on it, against 75.6 to 94.3 on an older benchmark (CelebDF++). People shown the videos picked out the fakes 52.6% of the time.
Ukrainian Catholic University, Czech Technical University in Prague and others · Sep 7, 2026Video - AI detectors rarely flag human writing, but sometimes miss AI text imitating real authors
Pangram, GPTZero and Originality.ai missed under 1% of AI text from basic prompts, and 10 to 18% when the AI was told to imitate a particular author (about 26% for scientific writing). Of 495 human passages, Pangram and GPTZero flagged none as AI; Originality.ai flagged 19 (3.8%).
Epoch AI (research nonprofit) · Jul 15, 2026Text - AI-detection tools have made huge leaps forward — how good are they?
News feature on AI-text detectors in universities and scientific publishing.
Nature · Aug 2026 · may be paywalledText - Fit for Purpose? Deepfake Detection in the Real World
Academic, government and commercial detectors tested on real political deepfakes from social media. All struggled to generalize, and simple manipulations threw them off, especially for video.
Purdue University researchers · Oct 2025Image · Video - Deepfake-Eval-2024
Deepfakes that circulated online in 2024, gathered from 88 websites in 52 languages. Open-source detectors’ scores fell by about half compared with older benchmarks: 50% for video, 48% for audio, 45% for images.
arXiv · Mar 2025Image · Video · Audio - Speech DF Arena
Public leaderboard of speech-deepfake detectors, scored on 14 datasets from studio recordings to phone calls, with an open toolkit for reproducing the results.
Hosted on Hugging FaceAudio - TRIED: Truly Innovative and Effective AI Detection benchmark
A framework for judging detection tools on real-world use: the formats and compression people actually share, languages, explainability, and who can get access.
WITNESS (nonprofit) · Apr 2025All media
Who tracks AI fraud
Groups that document incidents, publish methods, or take reports.
- WITNESS: Deepfakes Rapid Response Force
Connects journalists and fact-checkers with media-forensics experts on hard cases, and publishes case analyses.
WITNESS (nonprofit) · For journalists and fact-checkersAll media - Reporter’s Guide to Detecting AI-Generated Content
Methods for spotting AI-generated images, video and audio, by Henk van Ess. Updated September 2025.
Global Investigative Journalism Network (nonprofit)Image · Video · Audio - AI Incident Database
Searchable record of reported AI harms, including deepfake scams and impersonation, with periodic roundups.
Responsible AI Collaborative (nonprofit)All media - Political Deepfakes Incidents Database
Coded collection of political deepfakes and “cheapfakes,” mostly from the U.S.: who was targeted, how it spread, whether it was verified.
Purdue University GRAIL labImage · Video · Audio - Deepfake Incident Database
Dashboard of reported deepfake incidents targeting people and organizations, with a weekly roundup, the Deepfake Watchlist.
Resemble AI (company that sells detection)All media - FBI public service announcements
The FBI’s fraud warnings, including a July 20, 2026 alert about deepfake videos of senior FBI officials used to target scam victims a second time. Reports go to ic3.gov.
FBI Internet Crime Complaint Center (US government) · Jul 20, 2026 alertAll media - Family emergency scams
Consumer guidance on impostor calls, including cloned voices.
Federal Trade Commission (US government) · report fraudAudio
No longer running
Still recommended in some lists online.
- TrueMedia.org
A free nonprofit deepfake checker that combined several detection models. It closed January 14, 2025 and released its code.
Nonprofit · Closed · code on GitHubImage · Video · Audio
Descriptions come from each tool’s own site or the study itself; accuracy claims made by the companies are left out. Entries listed as “All media” cover every type.
A companion to Very Weird, Very Quick.