What each part of the prompt does
- Claims as a list
- Turning the post into numbered claims separates what can be checked from what is opinion or prediction. Viral posts mix the two on purpose, and pulling them apart is most of the work.
- Honest ratings
- Four ratings, including "cannot judge", give ChatGPT a way to say it does not know. The ban on invented sources matters because a confident made-up citation is worse than an admitted gap.
- Questions it skips
- Asking what the post leaves unanswered gives your piece its own angle. Your audience gets something the original did not offer instead of a summary of it.
- Your checklist
- The closing list hands the final verification back to you. ChatGPT can narrow the work, but you still open the report, the dataset or the original announcement before publishing.
Variations to try
Same idea, one thing changed.
Thread or carousel output I want to write about a social media post that is getting attention in my field, but first I need to know what in it holds up. The post, pasted in full: [POST] Who posted it and where: [AUTHOR] My audience and the format I plan to publish: [MY FORMAT] Work through it: 1. List every factual claim in the post as a numbered line, separating facts from opinions and predictions. 2. For each claim, say what kind of source would confirm or refute it, and what you know about it. Mark each one as likely true, disputed, likely false, or cannot judge. Do not invent sources, studies or numbers; if you do not know, say "cannot judge". 3. Point out missing context, loaded wording or a sample that is too small to support the conclusion. 4. Name two questions the post does not answer that my audience would ask. 5. Draft my piece as a numbered thread of six to eight short posts, or slides if my format is a carousel. Quote or credit the original post, rely only on claims marked likely true, and frame disputed ones as open questions. End with a checklist of the claims I must verify myself before publishing.
Reply instead of a post I want to reply publicly to a social media post that is getting attention in my field, but first I need to know what in it holds up. The post, pasted in full: [POST] Who posted it and where: [AUTHOR] My audience and the format I plan to publish: [MY FORMAT] Work through it: 1. List every factual claim in the post as a numbered line, separating facts from opinions and predictions. 2. For each claim, say what kind of source would confirm or refute it, and what you know about it. Mark each one as likely true, disputed, likely false, or cannot judge. Do not invent sources, studies or numbers; if you do not know, say "cannot judge". 3. Point out missing context, loaded wording or a sample that is too small to support the conclusion. 4. Name two questions the post does not answer that my audience would ask. 5. Draft a reply under 80 words that adds the most useful correction or context without attacking the author. Quote or credit the original post, rely only on claims marked likely true, and frame disputed ones as open questions. End with a checklist of the claims I must verify myself before publishing.
If the result isn't right
- Paste the post text rather than a link; ChatGPT may not be able to open the platform, and screenshots can drop words or whole replies.
- When a claim is marked likely true, still find the primary source; the rating reflects what the model knows, which can be out of date.
- If the post includes a chart, describe it in [POST] with its axis labels and source line, since those are where misleading charts usually go wrong.