# Draft three versions and pick the best: checklist

Write three different versions of the same piece, rate each one against criteria you set, and keep the winner with the reason shown, so the choice is not just a feeling.

## Before you start

- [ ] Have ready: What you need written
  Say what the piece is, what it must include, and anything the reader must do afterward. Add facts the AI should use.
- [ ] Have ready: Who will read it
  Name the reader and how well they know the topic.
- [ ] Have ready: What makes a version good
  List two to five things the winning version must do, one per line. For example "short enough to read in a minute" or "friendly but not casual".

## Step 1: Write three versions

- [ ] Write three versions that take clearly different approaches.
- [ ] You have: Three complete versions, each labeled with its approach.
- [ ] Check: The answer includes the words "Version C".
  A plain rule you can check by eye. If not: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it.

## Step 2: Rate each version against your criteria

- [ ] Rate every version on every criterion, with a reason that points to the text.
- [ ] You have: A table with a rating and a reason for every version against every criterion.
- [ ] Check: Every row of the table has one of "strong", "partly" or "weak" in the Rating column.
  A plain rule you can check by eye. If not: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it.

## Step 3: Keep the winner and say why

- [ ] Pick the version that best meets the criteria, and show the reason in the open.
- [ ] You have: The winner named with the reason, its final text, and any line worth borrowing.
- [ ] Check: The answer includes the words "Winner: Version".
  A plain rule you can check by eye. If not: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it.

## Final checks

- [ ] Check: Read the winner and the reason in the last answer next to the ratings table. Does the reason name criteria the person set, and do the ratings really support choosing that version? Answer FAIL if the reason uses criteria that were not listed or contradicts the table.
  Checked by a second AI prompt. If not: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it.
- [ ] Check: Every row of the table has something written in "Reason" (a blank or a dash does not count).
  A plain rule you can check by eye. If not: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it.
- [ ] Check: Compare the final text with the brief below. Does every date, name, number and promise in the final text come from the brief, with nothing invented?
  Checked by a second AI prompt. If not: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it.
- [ ] Check: Read the final text as the reader would. Is it the one you would send? If not, which version would you pick instead, and which criterion did the ratings miss?
  Checked by you. If not: Carry on, but write down the problem so whoever uses the result knows about it.

## You end up with

- [ ] The final text of the winning version, ready to use.
- [ ] Two other versions and a table of ratings, in case you want to borrow from them.
- [ ] A plain reason for the choice.

---

From Logic Lab (logiclabhq.com). Free to use and adapt, with credit.