Accurate Headline Drafts Without Clickbait

Source and thanks

Original TokRepo prompt, CC BY 4.0. Reference: ChatGPT release notes (reviewed 2026-10-04), used as context for a general-purpose AI workspace; these templates do not depend on a newly announced feature.

source reviewed; runtime not tested

Complete reusable prompt

You are helping me draft three accurate headline options for one article I have already written. Do not browse, publish, post, email or change anything. You only produce drafts and checks that I will review and act on myself.

My inputs:

• Article title (working):
• Full article text or a faithful excerpt:
• Target reader (one sentence):
• Where the headline will be used (publication, newsletter, blog, social preview, internal doc):
• Platform character limit or style limit if any (e.g. 'under 60 characters', 'sentence case'):
• Tone I want (plain, curious, formal, conversational):
• Words or claims I must avoid:
• Any factual limits I already know (results not proven, small sample, draft figures):

If the full article text is missing, stop and ask me for it instead of guessing. If the target reader or the usage platform is missing, still proceed but label that headline set 'needs more context' and state what is missing at the end.

Task:

1. Read my article and identify at most five factual load-bearing points: the specific findings, decisions, numbers or events the article actually supports.
2. Write three headline options that differ in angle, not just wording. Use these three angles in order: a. News/what-changed: states the concrete development. b. Reader benefit or question answered: states what the reader gets, without overpromising. c. Contrast or surprise grounded in the text: only if the article genuinely contains it; otherwise say so and skip to a fourth honest angle.
3. Every headline must be traceable. For each one, add: the angle label; the article line or short phrase it rests on; the words in the headline that carry the strongest claim; and a truth-check verdict of Pass, Caution or Fail.
4. Apply these accuracy rules to each headline:
   • No absolute claims ('always', 'never', 'everyone', 'proven') unless the article supports them.
   • No numbers that do not appear in the article.
   • No curiosity gap that hides the subject entirely (say what it is about).
   • No promise of outcome the article does not deliver (money, health, certainty, guarantees).
   • No invented quote, source, study or product feature.
   • If the article itself is uncertain, the headline must stay uncertain too.
5. For any headline marked Caution or Fail, revise it once and show both versions with a one-line reason for the change.
6. Add a short 'Reader test': would the target reader, reading only this headline, expect the article they actually get? Answer yes, partly, or no per headline.

Output format (Markdown):

• Section 1: Article spine — bullet list of up to five supported factual points, each with a short quoted or near-quoted anchor from my text. If you cannot find them, say what is missing.
• Section 2: Headline table with columns: Headline | Angle | Evidence anchor | Strongest-claim words | Truth check | Reader test | Length.
• Section 3: Revised versions of any Caution/Fail headline, before and after, with reason.
• Section 4: What I still need to decide — a short list of questions for me, such as which angle matches the publication, and any claim I should verify with a source.
• Section 5: One-line reminder that these are drafts for my review and that I am responsible for any final published claim.

Worked example (fictional, for shape only — do not reuse its content): Input excerpt: 'Our six-month trial of shared calendars across 12 staff found most scheduling clashes disappeared, but reminders were still missed when staff worked remotely.' Target reader: small-team managers. Platform: internal newsletter. Possible headline: 'Shared calendars cut clashes in our 12-person trial — remote reminders still missed.' Evidence anchor: 'six-month trial... 12 staff... most scheduling clashes disappeared... reminders were still missed when staff worked remotely.' Strongest-claim words: 'cut clashes in our 12-person trial'. Truth check: Pass (scope limited to the trial, limitation stated). Reader test: yes — reader expects a limited trial result with a caveat.

Now produce the output using my inputs above. Keep it compact, use plain language, and do not add any publishing, scheduling or sending steps.

References and reuse

• ChatGPT release notes · Reviewed 2026-10-04

Original TokRepo prompt · CC BY 4.0. Reference documents retain their own rights.