What we do with your text.
And what we won't do with it.

Plain-language policy. The legal version of these commitments lives in Privacy and Terms.

What humanise.ai is

A free tool that rewrites AI-generated prose to read as if a human wrote it. It's powered by large language models running on third-party APIs. When you paste text into the humaniser, the following happens:

1. Your browser sends the text over TLS to a Cloudflare Worker.
2. The Worker passes the text to our model provider's API with a short rewriting instruction.
3. The humanised result comes back and is returned to your browser.
4. The Worker isolate that handled your request is destroyed.

The text touches three places: your browser, our Worker, and our model provider. It is never written to durable storage at any step.

What we do with your text

• We do not store submissions. Zero content retention, by design and by code. The Worker has no database attached and no write path to one.
• We do not train on submissions. Our model provider contract excludes API traffic from training data. Submissions cannot be used to train future model versions, ours or theirs.
• We do not sell submissions. No data brokerage. No "anonymised" data shared with partners. The business model doesn't require it.
• We log anonymised metadata only. Character-count buckets, HTTP status, latency, country (from Cloudflare), and a request ID. No IP, no content, no identifying headers. Retention: 30 days. See Security.

What we do with the result we give you

Once the humanised text is on your screen, it's yours. We don't track what you do with it. We don't watermark it. We don't call home. You can copy it, edit it, submit it, publish it, bin it — we have no visibility into any of that.

What we use AI to do

Exactly one thing: rewrite the text you paste so that it reads as human-written while preserving meaning. We don't use AI to:

• Monitor your usage patterns
• Profile or classify users
• Generate marketing content in your voice
• Make decisions that affect you (there is nothing to decide — it's a text-in, text-out tool)

Our position on AI detectors

AI detectors are probabilistic classifiers. They produce scores with meaningful false-positive rates, especially on:

• Non-native English writers
• Formal academic or legal prose
• Heavily edited text of any origin
• Short passages (under 300 words)

A detector score is not proof of anything. It is a signal that deserves context. Institutions that treat detector output as verdict rather than evidence are making a methodological error, and we've seen the consequences — real human writing misclassified as AI, real penalties applied on bad data.

Our tool rewrites for rhythm and cadence, which happens to be what detectors measure. We don't guarantee detector bypass, and we think anyone who does is overselling. Our measured median pass-rate on synthetic AI text across GPTZero, Turnitin, Originality.ai, Copyleaks, and Winston AI is above 85%. We publish the measurement and stand behind it.

See Guide: AI detection, explained for the full picture.

What we won't help with

The humaniser is a general-purpose tool, and we don't police inputs. But we'll state the lines clearly:

• Submitting generated work into contexts that explicitly prohibit AI-generated content is still a policy violation whether or not a detector catches it. The humaniser does not change the underlying fact of where the text came from. If your institution prohibits AI-drafted work, humanising it doesn't make it compliant — it makes it harder to detect, which is a different and usually worse situation.
• Plagiarism. Running plagiarised text through the humaniser doesn't make it original. It rewrites the prose surface; the ideas and sources underneath are still someone else's. Attribute properly.
• Fabricated citations. LLMs hallucinate references. The humaniser does not check them. Always verify every citation before submitting anything that has them.
• Fraud, impersonation, harassment, or any content that would be illegal under UK law. Our Terms of Service cover the list in legal form.

These aren't detection problems, they're ethics and policy problems, and they outlive any detector arms race.

Transparency commitments

• If our data practices change, we will update this page with a date-stamped changelog and post a notice on the homepage for 30 days.
• If we ever add features that process text beyond rewriting (e.g., storage for revision history), they will be opt-in, clearly labelled, and separately explained.
• If a partner or provider changes our data handling, we will disclose the change here before it takes effect.
• We will never introduce an "AI training opt-out" that's silently opt-in. There is no opt-out because there is nothing to opt out of — the default is no training.

Questions?

• Policy questions: policy@humaniseai.ai
• Data requests (access, erasure, correction): privacy@humaniseai.ai — we respond within 30 days per GDPR/UK-GDPR/CCPA requirements.
• Abuse reports: abuse@humaniseai.ai

Last reviewed: 2026-04-20 · Next scheduled review: 2026-10-20.