Remove AI tells from a draft

This draft reads as though a machine wrote it. Fix it, then tell me exactly what gave it away.

Draft:
[PASTE THE DRAFT]

Who reads this: [AUDIENCE]
Where it appears: [EMAIL, LANDING PAGE, DOCUMENTATION, POST]
What it has to achieve: [THE ONE THING]
How I normally write, if you can tell from the draft: [OR "IT IS NOT MY WRITING"]

Produce three sections.

1. THE REWRITE — same meaning, same length or shorter. Vary the sentence lengths, including some short ones. Cut anything that could be said of a different subject without changing it. Do not replace a formal word with a casual synonym and call it a rewrite.

2. WHAT GAVE IT AWAY — quote each phrase or structure from my draft and say what makes it read as machine-written. Cover at least: openings that restate the title before saying anything; sentences built as "it is not just X, it is Y"; triples where the third item adds nothing; hedges attached to claims that were never in doubt; transitions used as decoration; a closing paragraph that summarises what was just said; and any sentence that is true of every company or every product in the category. Do not list a tell that is not in my draft.

3. WHAT IS ACTUALLY MISSING — the specifics that would have made this unmistakably mine: a number, a name, a date, a concrete example, something I know that a general model does not. List what to add and where. This section is the real fix.

Do not add em dashes to compensate. Do not make it folksy. If a sentence in my draft is genuinely good, leave it alone and say so.

The thing that marks writing as machine-produced is not any single word. It is that the text is perfectly fluent and says almost nothing that could only be said about this subject — the paragraph would survive unchanged if you swapped the product for a competitor's. Readers register this quickly, and what they conclude is not "a model wrote this" but "nobody involved knew anything specific", which is worse.

So a rewrite alone does not fix it. Softening the vocabulary produces the same empty paragraph in a friendlier register. That is why section three exists and why it is the one to read first: what the draft needs is the number, the date, the name, the thing that happened — the specifics no general model could have supplied, because it does not know them and you do.

Section two is there so the effect compounds. Having your own tells quoted back at you a few times changes what you write next, which the rewrite by itself never does.

1. Name the one thing the text has to achieve

   Vagueness in the brief is the most common source of vagueness in the draft. If you cannot name the single job, the rewrite will preserve the hedging because it has nothing to be specific about.

2. Do section three yourself

   It tells you where a specific belongs. It cannot supply the specific — if you let it invent a plausible number, you have traded generic writing for false writing, which is a worse trade than it looks.

3. Take the rewrite as a diagnosis, not a replacement

   Publishing the rewritten version gets you fluent text with the same missing substance. The value is in seeing which sentences survived contact and which turned out to be saying nothing.

4. Do not use this to defeat a detector

   AI-detection tools are unreliable in both directions and regularly flag human writing. Aim at text that carries information only you have; that is also the only thing that reliably reads as human.

• When the draft has nothing to sayNo rewrite adds substance. If section three comes back long, the honest next step is to go and find out the specifics, not to run the prompt again.
• For academic or regulated writing with a required registerFormal hedging is often the correct form, not a tell. Say what conventions apply or you will be advised to strip out something the reader expects.
• For non-native speakersSome of the same markers appear in careful second-language writing that is entirely human. Fix the missing specifics; do not let the tells list become a reason to distrust your own English.

• WritingApolloWrite a kind rejection emailTurns a blunt or over-apologetic rejection into a short, clear message that respects the reader's time without sounding cold.
• WritingApolloTurn a vague request into a good promptRewrites what you asked into a prompt that can actually be answered, and names the things the model was silently guessing.
• LearningApolloLearn a topic and get quizzed on itExplains something at three levels of detail, then questions you on it and reports which specific part you did not actually have.
• ResearchRagnarökCompare options and see where they differCompares the options on the things that would actually change your decision, and says separately which differences do not apply to you.