Content & Marketing · Updated July 17, 2026
ChatGPT prompt for an edit pass before publishing
This is the prompt we run over any draft before publishing. It does three things at once: it tightens the text, it flags every claim that needs a source, and it offers you three opening lines. What it does not do: invent the missing numbers or touch the facts.
The prompt to copy
Edit this draft. Pass 1: cut 20% of words without losing meaning. Pass 2: flag every claim that needs a source or number. Pass 3: rewrite the opening line 3 ways. Return the edited draft plus the flags. Draft: {paste}
Replace the field in curly braces with your draft as-is. The text comes back trimmed, with a list of claims to back up and three opening options to choose between.
When to use it
On any draft before you publish it: article, post, page, longer email. It is the last step before going live, right after you have written the first draft. The pass turns a vague reread into three sharp checks: length, claims, opening.
What you get back
- A text trimmed by about 20%, with no loss of meaning.
- The list of claims that need a source or number.
- Three versions of the opening line to choose between.
- No fact changed, no number added.
Real example: draft → trimmed text and flags
Input: a LinkedIn post draft by Léa at Brightside, an 8-person brand agency. An excerpt, as-is:
We are thrilled to announce that, after several months of intense and exciting work,
we have completely rebranded Fernhill Coffee, which allowed them to triple
their online sales in the very first month.
Output, the part that matters:
Edited text. We rebranded Fernhill Coffee. Result in the first month: their online sales tripled.
Needs a source. “triple their online sales in the very first month”: number to confirm. Source or screenshot needed before publishing [SOURCE?].
Notice what the AI did not do: it did not take the number at face value. It flagged it, because the prompt tells it to catch any claim that needs proof. Léa checks with the client, keeps the number if it holds, drops it if not. The checkpoint takes a minute and keeps an indefensible figure off the page.
This prompt ships inside the kit, alongside the other 49 in the library and the full Brightside worked examples.
Frequently asked questions
- Does this prompt work with Claude or Gemini?
- Yes. It relies on no ChatGPT-specific feature: paste it as-is into Claude or Gemini and the behavior is the same.
- Does cutting 20% of the words remove information?
- No, that is exactly the instruction: cut the extra word, not the meaning. The pass targets filler, empty phrases, and repetition. You review the trimmed text and keep control over anything that reads as cut too far.
- What do I do with the claims flagged as needing a source?
- The prompt surfaces them without fixing them: it is on you to supply the source or number, or drop the claim. That is the checkpoint that keeps you from publishing data you could not defend.