One Ad, a Week of Headlines Worth Testing
Most ad copy ships as one headline someone felt good about. It's not a bad instinct, but a single headline with no real alternative means that if it underperforms, there's no way to know whether the problem was the angle, the offer, or something else entirely, because nothing was ever compared against it.
Tip
One good headline tells you nothing without something genuinely different to compare it to. A set of near-identical variations doesn't count.
Testing isn't asking Claude for ten headlines. It's asking for three different reasons someone might click.
Why "give me ten options" doesn't actually help
Asking for ten headlines usually produces ten variations on the same underlying idea, different words, same angle. Comparing ten versions of one idea tells you about phrasing preferences, not about which angle actually works. A real test needs headlines built on genuinely different premises, not the same premise restated.
Building a real test set
- 1
Fix the audience and the single outcome
Who's seeing this ad, and what one action it needs to drive. Every variant should be optimizing for the same target, or the comparison won't mean anything.
- 2
Ask for angles, not phrasings
Name the specific angles you want: one benefit-led, one problem-led, one proof-led. Naming them prevents Claude from defaulting to ten versions of whichever angle came to mind first.
- 3
Pair each headline with a CTA that matches its logic
A proof-led headline earns a different call to action than a benefit-led one. "See the results" fits proof; "Start free" fits benefit. Mismatched pairs undercut the angle before the reader even clicks.
- 4
Tag each variant with what it's actually testing
Label them by angle, not just A, B, C, so when results come back you know which premise won, not just which arbitrary version did.
Audience: freelance designers on our free plan who've hit the project limit. Outcome: click through to upgrade. Give me three headlines, each under 8 words, on three different angles: one leading with the benefit of unlimited projects, one leading with the frustration of hitting the limit mid-project, one leading with the number of freelancers who've already upgraded. Pair each with a matching call-to-action button under 4 words.
”Checking a set before it ships
Each headline is built on a different premise, not a different phrasing of one premise
Every CTA's logic matches its own headline's angle
Each variant is labeled by what it's testing, not just lettered
Common mistake
Running five variants that all lead with speed or ease, then concluding the audience "doesn't respond to ads." What actually happened is one angle got tested five times. A flat result across near-identical variants says nothing about the other angles you never tried.
Inside Claude Tutorial
Comparing genuinely different options beats generating more of the same.
Asking for options that differ in substance, not just wording, is a habit that pays off well beyond ad copy. The app builds this pattern through short lessons you apply to real examples.
Of these three headline and CTA pairs, which would you actually expect to perform best with freelance designers specifically, and why? Be specific about which part of their situation each angle is speaking to.
”Continue reading
- Writing Better Marketing Copy with Claude: the genuinely-different-angles technique this article builds on, applied to a single high-stakes piece instead of a test set.
- Building a Brand Voice Profile Claude Uses on Every Piece of Copy: keeping every variant on-voice even when the angles themselves are meant to differ.
- Turning a Campaign Goal Into a One-Page Brief Your Creative Team Can Use: setting the audience and outcome this test set depends on before you start generating variants.