How to Write Better Headlines with AI (Without Clickbait)

1. Give the AI the article, not the topic

The single biggest improvement I made was embarrassingly simple: I stopped asking for headlines about a topic and started pasting in the finished article.

“Write a headline for an article about editing” gives you generic filler. “Write a headline for this article” — with the full text attached — gives you headlines that know what the piece actually says, which specific promise it makes, and which detail is worth leading with. The model can only be specific if you give it something specific to work from.

My rule now: no headline work until the draft exists. Headlines written before the article are guesses. Headlines written after it are summaries, and summaries are what good headlines are.

2. Ask for ten, then throw away nine

I always ask for ten headline options. Not three, not five — ten. Here’s why: the first three are always the obvious ones, the safe ones, the ones every writer in the niche would write. Options four through seven are where the model starts reaching. And somewhere in the last three there’s usually one that surprises me — a phrasing I wouldn’t have reached on my own, built from a detail in my own article.

Then I throw away nine. Keeping ten headlines is indecision. The point of ten options isn’t to have ten good headlines; it’s to give yourself enough material that one genuinely good one shows up. Most of the time I end up editing the winner anyway — swapping one word, cutting the tail off, tightening it by a third. The AI gives me raw material. I do the choosing.

3. Ban the clickbait moves by name

Clickbait isn’t a vibe, it’s a set of specific tricks, and I ban them by name in the prompt. My banned list for headlines:

• “You won’t believe…” — if I wouldn’t believe it, why would my reader?
• “The secret to…” — there is no secret. There’s a method. Name the method.
• “…and the last one will shock you” — the last one never shocks anyone.
• “Doctors hate this” and every “X hates this” variant — lazy, and readers know it.
• “This one simple trick” — if it were one simple trick, the article would be one sentence.
• Numbered lists that lie about their contents (“7 ways” when the article has four real ones and three fillers).

I paste this list into the headline prompt itself. The model then has to find something interesting to say without reaching for the cheap moves, and that’s where the good headlines live — in the gap between “boring” and “dishonest.” A headline can be bold without being a lie. “How I cut my editing time in half” is bold. “Editors hate this one trick” is a lie wearing bold’s clothes.

4. Put the benefit before the cleverness

Every headline is fighting for the same thing: the reader’s next three seconds. And in those three seconds, the reader is asking one question — “what’s in this for me?” Cleverness answers a different question, one nobody asked.

So I make the AI lead with the benefit and earn the cleverness second. Compare:

• Clever first: “The Headline Whisperer Strikes Again”
• Benefit first: “Write Headlines People Actually Click (Without Lying to Them)”

The second one tells me what I’ll get and what it costs me — nothing, no deception. The first one tells me the writer thinks they’re funny. I rewrote my headline prompts to ask for “the specific benefit, then the twist if it fits,” and the results got dramatically more clickable without getting any more dishonest. A reader who clicks because they know what they’re getting stays. A reader who clicks because they were tricked leaves — and remembers.

5. Test the headline against the article

This is my honesty test, and it takes about ten seconds: read the headline, then read the article, and ask whether the headline’s promise is actually kept.

If the headline says “in 30 minutes” and the method takes two hours, rewrite the headline. If it says “the complete guide” and the article covers three of the five things, rewrite the headline. If it promises a secret and delivers common sense, rewrite the headline — or rewrite the article until the promise is true, which is sometimes the better fix.

I do this test on every headline before publishing, AI-written or not, because it’s the cheapest trust test in writing. One broken promise costs you more than ten good headlines earn you. Readers forgive a dull headline. They don’t forgive being misled, and Google notices the bounce too — a headline that overpromises and an article that underdelivers is a ranking problem as well as a trust problem.

6. Use numbers and specifics, not superlatives

AI headline drafts love superlatives: “amazing,” “incredible,” “ultimate,” “best ever.” Superlatives are the cheapest words in the language because they cost the writer nothing and tell the reader nothing. “The ultimate guide” could describe any guide on the internet. “9 fixes that actually work” describes one specific thing.

My fix is to ask the model for numbers and specifics from the article itself: how many steps, how long it takes, what it costs, what the result was. “How to humanize AI text in 9 manual fixes” beats “The amazing guide to humanizing AI text” because the first one makes a checkable promise and the second one makes a noise. If the article genuinely has a number in it — a time, a count, a result — put the number in the headline. It’s the fastest way to make a headline feel concrete, and concrete is what gets clicked.

7. Let the AI write the options — you pick the winner

Here’s the part I want to be clear about: the AI is good at generating options and terrible at choosing between them. It doesn’t know your readers, your standards, or which promise you’re actually willing to keep. That’s your job, and it’s the job that matters most in the whole process.

So my workflow is fixed: I write the article, I paste it in, I ask for ten headlines with the clickbait list banned, I throw away nine, I test the winner against the article, and I publish. The AI did the brainstorming. I did the editing. The headline that ships is one I chose, tested, and can defend — and if a reader asks me why the headline says what it says, I have an answer that isn’t “the tool wrote it.”

The honest truth

A good headline is a contract. It tells the reader what they’ll get, and the article has to deliver it. AI can help you write the contract — it can give you ten drafts of it in ten seconds — but it can’t sign it for you. That’s the part where your judgment shows up, and it’s the part no tool will ever do well, because judgment is the one thing you can’t outsource to a pattern-matcher.

The headlines I’m proudest of on this site aren’t the cleverest ones. They’re the ones that told the truth clearly enough that the right reader clicked and stayed. That’s the whole game: not more clicks, but the right clicks, from readers who got what they were promised and came back for the next one.

Frequently Asked Questions