All packs

Kit · 29 pages · 11 prompts + house style

Catch what your AI made up

An AI draft that's wrong doesn't read wrong. Check one without reading it again.

After using this: You'll be able to tell the difference between a draft that reads well and one that's actually true, and you'll have a check that still works on the fortieth article.

Who this is for

You're the last person between an AI-written draft and a customer. The writing sounds fine, which is exactly the problem.

Not for you if

You want a better draft to begin with. That's Write help articles humans and AI can use. This one picks up after the draft exists.

$69

One-time purchase · HTML and PDF

Licence

Instant delivery after payment.

Team licences and invoices

How it works

Every instinct you've built for reviewing writing is tuned to catch clumsiness. When a person doesn't know something, it shows: they hedge awkwardly, they go vague, they leave a gap. When a model doesn't know something, it writes the most plausible sentence instead, in the same confident voice as everything around it. So a bad AI draft doesn't read badly, and reading it again more carefully is the one method that can't work. This kit is the alternative: you don't review these, you verify them. Turn the draft into claims, check each against a source, and know the seven shapes a fabrication takes so you can find them without reading.

What's inside

  1. 01How to use this kit
  2. 02Why a bad draft doesn't read badly
  3. 03The seven things it gets wrong
  4. 04Check it against something, don't read it through
  5. 05The review that still works on the fortieth article
  6. 06When to reject instead of fixing
  7. 07The prompts
  8. 08Telling whether it worked
  9. 09Give your AI your review rules
  10. 10What to do next

Preview a prompt from the pack

The full prompt, exactly as it appears in the pack. Nothing removed.

Find every hedge in this documentation draft and tell me what it is hiding. The draft: [paste in full] Find every instance of: usually, typically, generally, normally, in most cases, should, may, might, often, a few, several, relatively, fairly, approximately, around, some time, shortly. For each one: 1. Quote the sentence. 2. Say what specific fact would replace it if we knew. 3. Say whether the thing genuinely varies, or whether the hedge is standing in for not knowing. Explain how you can tell. 4. If it genuinely varies, write the version that says what it depends on and gives the range, because that is useful where a hedge is not. Then rank them by how much damage the vagueness could cause. A hedge about how long something takes is mild. A hedge about whether something can be undone is not. Do not remove genuine uncertainty. The goal is to replace vagueness that hides ignorance, and to keep vagueness that reflects reality while making it specific.

Questions

Catch what your AI made up, $69

Instant delivery after payment.

Pairs with

Prompt pack$59

Write help articles humans and AI can use

Write it once, so a person can skim it and a bot can answer from one section of it.

30 pages · 11 prompts + house style

See what's inside
Kit$89

Give everything one name

If you call it three things, your AI thinks it's three things. Fix the names safely.

43 pages · 13 prompts + house style

See what's inside
Kit$129

Keep your support bot honest

Your bot is live and nobody checks what it tells customers. This is that review routine.

35 pages · 12 prompts + house style

See what's inside

We keep no cookies and use no third-party tracking. The only optional thing is a count of which pages get visited, which helps us see what people find useful. Turning it off changes nothing else on the site. What we store