AI Detector
Whakama ā-pūnaha OK
380+ tauira
Kāore he tohu wai
Kāore he kōwhiringa e hiahiatia ana
Kāhua:
+ GPT-5, Claude, Gemini
One place to check whether something was made by AI — paste text below for an instant read, or jump to the image and video detectors. We're honest about the limits up front: no AI detector is reliable enough to accuse anyone. Use these as one signal among several, never as proof.
What AI detection can and can't do
What it's good for
- Auditing your own draft before you publish — if your writing scores 90% AI, you leaned on it too hard.
- One signal among several when reviewing a submission, alongside draft history and a conversation.
- A quick sanity check on a viral image or clip before you share it, paired with reverse-image search.
What it can't do
- Prove anyone cheated. Stanford found non-native English writers get false-flagged as AI up to 61% of the time.
- Reliably catch modern models — false-negative rates climb above 30% on recent text, image, and video generators.
- Serve as court-admissible forensic proof. Where they exist, C2PA content credentials are more reliable than any visual heuristic.
Which detector do I need?
| What you have | Use |
|---|---|
| An essay, article, email, or any block of writing | AI Text Detector |
| A photo, screenshot, or artwork | AI Image Detector |
| A video clip — generated footage or a possible deepfake | AI Video Detector |
Ko ngā kōwhiringa hōhonu
Whakamutunga
Kua iti te tohu.
Ka whiwhi atu ētahi atu tohu
❤️ E hiahia ana ki te Free.ai? Whakapāpāho ki ōna hoa!
Kaituhi ki kia whiwhi ai i tētahi pātahitanga tohutoro, ā, ka whiwhi i ngā tohu 25,000 ia hoa.
E hiahiatia ana ētahi atu?
Ka whakaingoatia te wāteatanga mo ngā tohu 30K / rā + tāpiri 10K
Ka whakaingoatia te pūkete
E whakapāpā ana i tō tātau tono...
Free AI detector. Check if text was written by AI, or scan an image or video for AI generation. Honest about accuracy — a heuristic, not proof.
He pēhea te whakamahi AI Detector
1
Kei roto i tō tou tāuru
Type i te kupu, tuku i tētahi faila, whakaahua rānei i te mea e hiahiatia ana e koe. Kāore he tatau e hiahiatia ana.
2
Ka tirohia te whakatūnga
Ka tukatuka tātau AI i tō tātau tono i roto i ngā wā kotahi mā te whakamahi i ngā tauira pūtake tūwhera pai rawa.
3
Whakahua & tiritiri
Whakataki, tārua, tiritiri rānei i tōna hua. Whakatika noa iho mō te whakamahinga whaiaro, hokohoko rānei.
Ka whakamahia tēnei utauta mā te API
Ka whakamātautau tēnei utauta mai i tōtou waehere. Ko te wāhi mutunga o te REST e ōrite ana ki te OpenAI, te mana tohu-tokona, kāore he SDK tāpiri e hiahiatia ana. Ko ngā utu tohu e ōrite ana ki te whakawhitinga whatunga.
curl -X POST https://api.free.ai/v1/chat/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"model": "qwen7b", "messages": [{"role": "user", "content": "Summarize this: ..."}]}'E pā ana ki ngā utauta AI wātea
AI Detector — FAQ
It is one place to check whether something was made by AI across three modalities. Paste text right on this page for an instant read, or open the dedicated AI Image Detector and AI Video Detector. Each returns a probability score, a verdict, and the signals it spotted — and each is upfront that it is a heuristic, not proof.
Yes — three tools under one roof. Text detection looks at linguistic fingerprints (burstiness, AI-tell phrasing). Image detection looks at visual artifacts (warped hands, garbled text, impossible reflections). Video detection samples frames and checks each, plus a deepfake face-swap mode. Pick the one matching what you have, or use the quick text check here.
None is reliable enough to accuse anyone, but text is generally the most studied and image the most mature; video is the hardest because top models produce clips that pass frame-level checks. Across all three, modern state-of-the-art generators routinely beat detectors — expect 60–80% accuracy on recent outputs, lower on the newest models.
No. Courts require provenance-based verification — C2PA content credentials, cryptographic signing, original-source chain-of-custody. Every visual or linguistic heuristic, ours and the paid competitors alike, has too high a false-positive rate to be evidence. Use these as a first-pass signal, never as the basis for an accusation or a grade.
Writing (essay, article, email) → the AI Text Detector. A photo, screenshot, or piece of art → the AI Image Detector. A video clip, including possible deepfakes → the AI Video Detector. The three cards at the top of this page route you to each, and the quick box here handles text instantly.
The box on this page gives you a fast verdict — score, label, and a one-line summary — so you can triage in seconds. The full AI Text Detector adds per-sentence confidence and a labeled list of every AI-tell and human-tell signal, plus text-type baselines (academic, blog, email, creative, code). Click "Full per-sentence analysis" to go deeper.
Free, no signup for your first checks. Everything runs in a daily token pool on our own GPUs — no credit card, no per-scan fee. Signing up unlocks the full daily pool, generation history, and access across all of Free.ai's 400+ tools.
Those are mostly single-modality and paid — GPTZero ($10–24/mo) and Originality ($14.95/mo) for text, Hive (enterprise B2B) for images/video. Free.ai bundles text, image, and video detection for free with no signup, and we are honest about accuracy instead of advertising the 99% claims independent research keeps contradicting.
Sometimes — that is the core limitation. Stanford found non-native English writers get false-flagged as AI up to 61% of the time; heavily edited or stylized photos can read as "too perfect". Always read the signals and the confidence level. A "Likely AI" verdict with low confidence and thin signals is probably a false positive.
Do not use any AI detector — ours or a paid one — to accuse a student of cheating, reject a job applicant, or flag an email as "suspected AI". The error rates are far too high to support those decisions. Legitimate uses: auditing your own drafts, a sanity check before sharing a viral clip, or one signal among several in a human review.
Text is sent to our self-hosted GPUs, analyzed, and discarded. For video, frames are sampled in your browser and only those still frames are analyzed, then discarded — the video file never leaves your device. We log only timestamp and token cost for billing, never your source content. Details at /privacy/.
Yes. Text detection: POST to /v1/chat/ with the detector system prompt. Image and video frames: POST multipart to /v1/image/describe/ with mode=ai-detect or deepfake-detect. Bearer auth, 10K tokens/month free. If you build on detector output, read the accuracy caveats above — do not make accusations from a score alone. Snippets at /api/.
Ka whakaingoatia te wāteatanga mō ngā tohu 30,000
Ka waihanga tētahi pūkete wāteaKāore he kāri ā-pūtea e hiahiatia ana
He pēhea te whakawātea i tēnei utauta?