Your best headline has probably already been written. A customer wrote it, in a review, a support ticket or a sales call, while describing the problem you solve in words you would never think to use. The people who write headlines know the product best, and that’s exactly the trouble. We describe what we built. Customers describe what was wrong before and what changed after, and their version usually lands better. Here’s the one-hour sprint I use to find their words and turn them into headlines: fifteen minutes to collect, ten to cluster, fifteen to read, ten to write and ten to set up a test.
Step 1: Collect the raw words
Give this fifteen minutes. Pull customer language from wherever it already lives:
- reviews of your product and of your competitors, especially the unhappy ones
- support tickets and live chat transcripts
- sales call notes or transcripts
- answers from onboarding and cancellation surveys
- emails, direct messages and community threads
Copy everything word for word. Keep the typos and the odd grammar, because that’s how people really talk. Put it all in one document, one comment per line, with the source at the start of each line.
No customers yet? Use the next best thing. Read your competitors’ reviews. Read the questions people ask in forums and communities. Look at what they type into search. Prospects who talked to you and didn’t buy are worth including too, because they tell you what held them back.
Two rules before anything goes near a model. First, remove names, email addresses, phone numbers and account details. Second, stick to one audience per sprint. Mixing new customers with long-standing ones, or buyers with people who cancelled, blurs the patterns you’re looking for.
Step 2: Let AI find the patterns
Ten minutes. Paste the document into the model with a prompt that asks for exact quotes and counts, not a summary. A summary turns their words back into yours, which defeats the point.
Below are customer comments about [product], taken from reviews, support tickets and sales calls. Each line starts with its source.
Group the comments into themes. For each theme:
- give it a short name
- count how many comments mention it
- quote the exact phrases customers use, word for word
- say whether the language describes the problem before, or the result after
Sort the themes by how many comments mention them.
Never paraphrase, and never invent a quote.
If a comment doesn't fit any theme, list it under "Other".
[paste comments]If you have more text than the model can take in one go, split it into batches and ask it to merge the themes at the end.
Step 3: Read the clusters yourself
Don’t skip this step. It’s the one that matters most. Give it fifteen minutes. The model finds patterns. You find meaning.
Read every theme and look for four kinds of language:
- The trigger. The moment someone realised they needed help.
- The pain. How the problem felt, in their words, not yours.
- The result. What changed afterwards, especially anything specific.
- The doubt. What nearly stopped them from buying.
Each kind has a job on the page. The trigger and the pain make strong opening lines. The result makes a strong headline. The doubt rarely belongs in the headline, but it belongs right under it, answered.
Counts tell you what’s common, not what’s important. Keep an eye on the small themes too. One vivid phrase can be worth more than a big, vague cluster.
Specific beats general. A phrase about the exact moment the problem bit, a missed deadline or a Sunday night lost to admin, will do more work than “saves time” in almost any headline.
Then check every quote. Models sometimes merge two comments, or smooth a phrase into something nobody actually said. Search your raw document for each phrase you like. If you can’t find it, you can’t use it.
Finish by picking three to five phrases, copied exactly.
Step 4: Write headlines in their words
Ten minutes. Write a few yourself first, so the model’s ideas don’t anchor yours. Then use a prompt that keeps it close to the phrases you picked.
Here are the phrases our customers use, copied exactly:
[phrases]
Here is what [product] does, in one sentence:
[description]
Write 15 headlines for [page or ad], aimed at [audience].
Rules:
- every headline must reuse one of the phrases, or stay very close to it
- no more than 10 words each
- plain English, no jargon, no exclamation marks
- 5 that name the problem, 5 that name the result, 5 that name the moment they realised they needed help
After each headline, say which phrase it came from.The last line keeps the model honest. If a headline can’t point back to something a customer said, cut it.
Then cut hard. Read each headline out loud and ask three questions. Is it true? Is it clear? Would a customer say it this way? Anything that fails one of them goes, however clever it sounds.
Step 5: Pick two and test them
Ten minutes to set up. Cut the list to your five favourites, then run a quick check before you spend anything:
You are a busy [audience] who has never heard of [product].
Read these headlines:
[headlines]
For each one, answer in one line:
- What do you think this product does?
- Would you keep reading? Why, or why not?
- Which words sound like marketing rather than something you would say?
Then pick the two you would click, and explain why in plain English.Treat the answer as a filter, not a verdict. A model pretending to be a customer isn’t a customer.
Then take the two strongest to the market, wherever you’ll get an answer fastest: two versions of an ad on a small budget, two subject lines sent to part of your email list or a landing page with traffic split between two headlines. Change one thing at a time. Decide which number you’re watching before you start: clicks for an ad or an email, sign-ups or enquiries for a page. Then let the numbers pick the winner.
If neither version wins clearly after a fair run, the difference was probably too small to matter to buyers. Next time, test two different themes, not two wordings of the same one.
Keep the words working
The sprint gives you more than a headline. Keep a running document of every phrase you found, with a note of which ones won and which didn’t. Share it with whoever writes your ads, onboarding emails and sales scripts, because the same words work there too. For ads, generate twenty angles from the phrases, cut to five and let the numbers pick two. Then run the sprint again every month. Customers change, and so do the words they use.
Book the hour
Block an hour this week for the page that matters most, and keep the free Prompts for growth teams pack open while you work. It has more prompts like these for research, content and ads, ready to copy.