LookAI

Spot AI text, sentence by sentence.

One API call returns a verdict, a score and the exact passages that read as machine-written. 26 languages, humanized text included.

10,000 words free every month. No card needed.

curl -X POST https://lookai.io/v1/analyze \
  -H "Authorization: Bearer $LOOKAI_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Furthermore, the stock market crash..."}'
Response200 OK · 1.8 s
{
  "predicted_class": "ai",
  "score": 96.8,
  "confidence": "high",
  "ai_fraction": 1.0,
  "sentences": [
    { "index": 0, "score": 98.4,
      "label": "ai", "highlight": true },
    …
  ]
}

From sign-up to your first result in a few minutes

  1. 1

    Create a key

    Sign up, open the dashboard and create an API key. The first 10,000 words of every month are free.

  2. 2

    Send a text or a file

    POST plain text, or a PDF, DOCX, TXT or Markdown file. The result comes back in the same request, usually in 1 to 5 seconds.

  3. 3

    Show the verdict

    Display the class, the score and the highlighted passages. You pay only for the words analysed.

A verdict you can act on, not just a number

Every response carries one of three classes. When the evidence is weak the API says so instead of guessing, and points to the passages that moved the score, so a person can look before anyone is accused.

human
Written by a person. Nothing is highlighted, even if a sentence on its own looks suspicious.
ai
Generated by a language model, including text passed through a humanizer or a paraphraser.
inconclusive
Not enough evidence either way, often a human text with an AI-written passage, which is highlighted for review.

In our held-out tests, fewer than 1 in 100 human-written texts reach the AI band. Single sentences are never highlighted on their own: a passage needs at least two.

Sentence scoresessay-4812.docx
  • My grandfather used to say the crash started in the farms, years before Wall Street noticed.

    6
  • Prices for wheat and cotton fell through the twenties.

    11
  • Furthermore, the stock market crash of 1929 was the result of a complex interplay of factors.

    96
  • These elements combined to create an environment in which a sudden loss of confidence could trigger collapse.

    93
  • When I read the newspaper archive for October 24, what struck me was how calm the first reports sound.

    4
  • One columnist blamed "nervous women investors," which says more about 1929 than about the market.

    41
Where the document score falls
human
inconclusive
ai
03075100

Built for teams that review a lot of writing

Education

Check essays and assignments inside your LMS, with the passages to discuss highlighted.

Legal and compliance

Screen statements, submissions and reports before they enter a case file.

Publishing

Review pitches, freelance copy and reader submissions at the volume they arrive.

Hiring

See which parts of a cover letter or a written test the candidate wrote themselves.

26 languages, each calibrated on its own

The same writing lands differently in every language, so each one has its own calibration. Other languages still get a score, marked as low confidence.

Plans that grow with your volume

Pick the words you need each month. If you go over, the extra words cost $135 per million instead of stopping your app. Change or cancel any time.

Monthly plan
Selected plan
Growth
$121/ month
  • 1,000,000 words included every month
  • Extra words at $135 per million, never a hard stop
  • Use it inside your own app or product, for your own customers
  • Sentence-level highlights in 26 languages
  • Texts never stored, never used for training
words
Plan
$121.00
Extra words
$0.00
Estimated total
$121.00
Start free, choose a plan later

10,000 free words every month without a plan. Prices exclude VAT. More than 20M words? Talk to us.

Your texts stay yours

Built for schools, law firms and publishers that handle other people's writing.

How we handle data
Never stored
Texts and files are analysed in memory and discarded. They are not logged and never used for training.
Only what you need for billing
Each request keeps its date, word count, language and verdict, for your history and invoices.
Keys you can revoke
Keys are stored as a one-way hash, and revoking one stops it immediately.
Payments by Stripe
We never see or store card details. Everything is served over HTTPS.

Questions

Do you store the texts I send?

No. Texts are analysed in memory and discarded. We keep the date, word count, language and verdict of each request for your usage history and invoices, never the text itself, and nothing is used for training.

Which AI models does it detect?

Text from current models such as GPT, Claude, Gemini, DeepSeek, Mistral, Grok and Kimi, including text rewritten by humanizer tools and paraphrasers.

How accurate is it?

In our held-out tests fewer than 1 in 100 human texts reach the AI band. No detector is perfect, so treat a result as evidence for a person to weigh, especially for short texts or low-confidence results.

How are words counted?

Words are separated by spaces. In Chinese and Japanese, two characters count as one word. Failed requests are never billed.

How does billing work?

Plans are billed monthly by card through Stripe, from $40 for 300,000 words. If you use more than your plan includes, the extra words cost $135 per million and are added to your next invoice, so your app never stops. You can change plan or cancel any time, and you get 10,000 free words every month before you choose a plan.

Can I use it inside my own product?

Yes. You can build the API into your own app, platform or service and show the results to your own customers. The only thing you cannot do is resell raw API access or share your keys.

What are the limits?

From 30 to 15,000 words per request and 60 requests a minute per key. Ask us if you need more.

Try it on your own texts

10,000 words free every month, and an API key in under a minute.

Get an API key