Generate headline variations in named styles
Ten headlines across labelled styles — how-to, question, number, contrarian — none of them promising more than your one verifiable fact.
Last updated
Fill in the blanks
Assembled in your browser — nothing you type is stored or sent anywhere.
Your prompt
Model-agnostic — works in Claude, ChatGPT, Gemini or any AI chat.
Why this prompt works
- →Naming the styles up front (how-to, question, number...) forces variety along dimensions that actually differ, instead of ten rewordings of the same headline.
- →Anchoring every headline to one verifiable point is the guardrail against clickbait — a headline that isn't licensed by that fact gets skipped rather than invented.
- →Labelling each headline with its style makes the output scannable and lets you pick a style on purpose instead of guessing what a headline is going for.
Part by part
- 10 headline options
- A specific count, high enough to force past the obvious. The first three are always the same three, and the useful one is usually past the sixth.
- label the style in brackets
- Makes the set comparable. Unlabelled, you cannot see that eight of ten are the same style with different words.
- Every headline must be something the piece can actually back up
- The honesty constraint. Headline generation is where a prompt most easily produces a promise the article cannot keep.
- no promising a twist, secret or result
- Names the specific overclaims this task attracts, which is more enforceable than asking for accuracy in general.
- skip that style instead of forcing a dishonest one
- Explicit permission to return fewer options. Filling a quota with a dishonest headline is the failure this prompt is designed around.
Common mistakes
Write me some catchy headlines.Catchy is optimised for by overclaiming, so you get secrets, mistakes you are making, and results nobody promised. Each one is a small debt the article has to pay and cannot.
Write 10 headlines for this article.Without the style spread you get ten variations of one framing. The value of ten options is comparing genuinely different approaches, not choosing between synonyms.
Write clickbait headlines that get traffic.Clicks earned by a promise the piece does not keep produce immediate bounces, which is a worse signal than a lower click rate. The cost lands on the piece rather than on the headline.
Questions
Why state the key point separately?
Because it is what every headline is checked against. Without it the model infers the claim from the topic and generates headlines for an article that does not exist.
Why is the style label useful?
It shows you the shape of the set at a glance. A list where seven are question headlines has not explored the space, and the label makes that visible without reading all ten carefully.
Which style performs best?
It depends entirely on the audience and where the headline appears, which is why the prompt asks for a spread rather than a recommendation. Test rather than trusting a general claim about what works.
Should the headline match the page title exactly?
Not necessarily. A search result title has different constraints — length, keyword placement — from a headline on the page. Generating both from the same key point keeps them consistent without forcing them identical.