Claude email humanizer

The engine receives one string and returns one string. The instruction is to rewrite the supplied text to accomplish the transformation the tool name implies while preserving the original meaning and factual claims, returning only the rewritten text with no preamble, commentary, quotation marks or fence. Diction, cadence and sentence length change. The information is meant to survive.

Envelope: 5 runs per rolling 24 hours, 1 to 12,000 characters, an optional 300-character instruction, live character and word counts, Copy on completion, nothing retained. There is no account, no history and no second output pane.

This is the difference that matters operationally and it is worth stating as a limitation rather than dressing up. refine.so, the closest email-specific competitor, implements field protection: "Subject lines, signatures, and CTAs go through unchanged", and it "treats names, links, and quoted phrases as fixed". That is a real feature and this route does not have it.

The consequence is a discipline rather than a defect, provided you know about it. Submit the body. A signature block in the input is text to be rewritten. A tracking URL in the input is text to be rewritten. Quoted thread history in the input is somebody else's words, rewritten and returned as though they had written them differently. The selection you make before pasting is the only field protection available.

One of these steps exists because of the gap described above.

Step 1

Isolate the body. Subject line, signature, disclaimer, links and quoted history stay outside the input, and go back around the result afterwards.

Step 2

Constrain the rewrite in the instruction field: a length, a register, a paragraph to leave alone. It is the only control surface on this route.

Step 3

Diff the output against the draft on entities first: names, dates, figures, anything a recipient could act on incorrectly. Prose second.

refine.so's description is the most precise in the result set: "formal openers, three-sentence body, a vague closing", with "corporate hedges" and "three-sentence symmetry" between them. What makes it a good description is that it is structural rather than lexical, so it survives a change of vocabulary.

The lexical layer is easier to fix and everyone names the same item: "I hope this email finds you well". Removing the opener and the softening sentences is free, immediate, and diagnostic. If the message loses its substance when the padding goes, the padding was carrying the message, and no rewrite will repair that.

Only marginally, and the most careful page in this category says so about its own product. refine.so writes that reply rate is driven by relevance, timing and the offer rather than by tone alone, and describes humanizing as one of the cheaper wins available. That is the correct magnitude of claim, and it is notably smaller than the one its competitors make.

The variables that dominate are outside any rewriter's reach: whether the recipient can act on the request, whether there is one request, whether it appears before the fold, and whether this week is a reason to answer. A rewriter operates on the fourth-order term. Worth doing, not worth mistaking for the work.

rewriteai.com advertises "99%" resistance to AI detection, against "10-30%" for ChatGPT output and "40-80%" for competing humanizers, plus "3× Higher reply rates vs raw AI emails" and better deliverability through fewer spam trigger words.

Each of those describes a system the vendor does not control. Detector thresholds move with configuration. Reply rates depend on the list, the offer and the sender. Inbox placement is decided by filters whose rules are not published and change without notice. We publish none of these numbers, not because ours would be worse, but because no one selling a rewriter is in a position to measure them honestly.

12,000 characters per run is high for this category: refine.so's free tool takes 200 words, rewriteai.com's free tier takes 300 words per input and 500 words a month. A long project update, a detailed reply or a multi-paragraph proposal fits in a single run here.

So the binding constraint is the allowance, not the cap. Five runs per rolling 24 hours rewards one considered pass over the message that matters, and punishes iterating on the same draft. Repeated rewriting converges toward a flat, uniform register, which is the thing you started out trying to remove.

No field protection for subject lines, signatures, links, names or quoted text. No generation from scratch, no personalisation from a contact record, no tone presets, no file upload, no API, no account, no stored history.

No detector claim, no deliverability claim, no reply rate figure, no speed figure and no language claim in either direction. No guarantee that factual content survives, only an instruction to preserve it and the recommendation that you verify. And a specific caution for regulated senders: approved wording stops being approved wording once it has been rewritten.