What we help produce — and what we decline.

What we help with

The overwhelming majority of writing the engine sees is the work professionals do every day — emails, posts, articles, summaries, applications, drafts, replies, briefs, decks, documentation, marketing copy, and the long tail of correspondence. We support this work without restriction. The rewrite engine exists to make this writing clearer, warmer, more specific, and more recognisably human, and that purpose is the default mode of the product.

What we decline

Some categories of content fall outside what we are willing to help produce, regardless of the user's intent. These are the categories where harm to readers, to third parties, or to the integrity of writing as a profession is substantial enough that the line is non-negotiable.

• Deceptive impersonation. Content presented as written by a real, named person without their involvement — fake testimonials, fake reviews, ghost-written communications passed off as the named author's first-person work where the named author has not reviewed the work.
• Disinformation campaigns. Content engineered to mislead readers about verifiable facts, particularly at scale — manufactured news, fabricated statistics, content designed to look like independent reporting when it is not.
• Targeted harassment. Content directed at a specific identifiable person and intended to harm them — coordinated abuse, doxxing, threats.
• Content that exploits minors. Any content sexualising or exploiting people under 18, in any form.
• Operational instructions for serious harm. Specific, actionable instructions for producing weapons capable of mass casualties, or for executing attacks against critical infrastructure.
• Academic dishonesty in graded work. Content explicitly intended to be submitted as a student's own work in violation of a stated academic integrity policy. (Note: academic writing assistance — feedback, editing of the student's own draft, learning support — is supported. The line is misrepresentation in a graded context where the institution prohibits it.)

How decisions get made

The policy above is implemented through a combination of automated checks and human review. Automated checks catch the clearest cases and are deliberately tuned to avoid over-blocking — the cost of refusing a legitimate request is real, and we treat it that way. Edge cases escalate to human review. We err on the side of supporting the work; we draw the line at the categories above and explain when we do.

If you think we got a decision wrong

Tell us. Send us the prompt, the response we returned, and a brief note on why you think the decision was wrong. Real people read these. Over-refusal is a failure mode we take seriously and act on.