← Strategies in School

Writing and AI

Humanize AI-generated Text

A revision prompt, with a clear look at what AI detectors can—and can’t—reliably tell you.

This prompt asks an AI system to revise prose with more specific reasoning, varied structure, and closer attention to the source. Those goals can improve a draft when the writer checks every claim and keeps their own voice.

COPYABLE PROMPTScroll inside to read the full text
Loading prompt…

How the prompt processes a draft

The prompt is a staged set of instructions. It first classifies the draft, then chooses editing rules, applies sentence-level changes, and asks the model to audit its own result. Each stage narrows the rewrite in a different way.

1. It chooses a rule set before rewriting

The model must label the text as an analytical report, literary analysis, or exposition/narration. It also chooses a light, medium, or heavy edit; if no intensity is supplied, the prompt says to use medium. Finally, it lists filler and repeated points for removal. This initial scan decides whether the later rewrite should preserve the paragraph order, reorganize the argument, or rebuild the structure.

2. It places limits on what may change

The fidelity rules tell the model to keep facts, figures, names, quotations, and the original conclusion. If the source lacks support, it must not invent citations or events; it may explain only a common-sense link, identify missing support, or cut unsupported material. Those constraints are meant to keep stylistic editing from quietly changing the substance.

3. It applies a global style pass

The universal rules target transitions, repeated claims, generic sentences, hedging, caveat placement, paragraph endings, and stock vocabulary. For example, the prompt asks the model to move a necessary qualification to the end of its paragraph and to end paragraphs on a concrete fact, action, or limit. That changes the order and cadence of the prose; it does not establish that any detector uses those traits.

4. It adds genre-specific requirements

For reports, it requests a clear early judgment and causal explanations. For literary analysis, it asks for close attention to words, syntax, and observable details. For exposition or narration, it asks for subject-specific details and varied paragraph lengths. Only the selected category should guide the rewrite, so a mistaken genre choice can make the result feel forced.

5. It edits sentences, then runs a checklist

The final rules favor a direct subject and verb, move the main point ahead of extra qualifications, and replace abstract noun phrases with actions. The self-check then asks the model to scan for replaceable claims, repeated structures, weak paragraph endings, and missed rules. This is another instruction to the same model, not an independent fact-check or a reliable guarantee that every constraint was followed.

Where the rules conflict

The prompt asks for unique insights while forbidding invented details. When the source is thin, those goals pull in opposite directions; its safer fallback is to state that support is missing or remove the unsupported claim. It also tells the model to delay caveats while requiring it to preserve real limitations, which can separate a qualification from the claim it limits.

Several fixed quotas and prohibitions compete with the request for natural, specific writing. A model can satisfy paragraph-length ratios or word limits mechanically while making the draft less clear. The opening claim that detectors flag particular argument patterns is asserted by the prompt without evidence, and no rewrite instruction can guarantee a detector result or make authorship impossible to identify.

How to review the result

Compare the rewrite with the source sentence by sentence. Check that every fact and conclusion still has support, restore qualifications that readers need beside the claims they limit, and remove any “insight” the source cannot justify. Use the instructions as an editing checklist, not as a way to conceal who wrote the work.