Verified against ChatGPT · 2026-08-11
Generate hero section headline variants matched to where the visitor is coming from, not just A/B noise
Produces a set of hero section headline and subheadline pairs, each matched to a distinct visitor entry state (cold ad traffic, warm retargeting, direct/branded search) so variants test genuinely different hypotheses instead of just wording tweaks.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Write hero section headline and subheadline pairs for Homepage for a project-management tool aimed at small agencies., where each variant is matched to a genuinely different visitor entry state rather than being a wording tweak on the same idea — the goal is variants that test different hypotheses about what makes someone convert, not five ways to say the same sentence.
PRODUCT / PAGE
Homepage for a project-management tool aimed at small agencies.
ENTRY STATES TO COVER
Cold traffic from LinkedIn ads, warm traffic retargeted after a free-trial signup abandonment, direct traffic from branded search.
(e.g., cold traffic from a paid social ad who has never heard of this before, warm traffic retargeted after abandoning a cart or signup, direct traffic typing the brand name because they already decided)
CORE VALUE PROPOSITION
One shared view of client work and billing so agencies stop reconciling three tools at invoice time.
WHAT MUST NEVER APPEAR IN ANY VARIANT
Never compare directly to named competitors by name.
For each entry state listed, write one headline (under 10 words) and one subheadline (under 20 words) built on a different hypothesis about what that visitor needs to see first: cold traffic likely needs the core value prop explained from zero, since they have no context; warm/retargeted traffic already knows the value prop and needs a reason to come back now, like addressing what likely stopped them the first time or referencing urgency; direct/branded traffic already decided and needs confirmation plus a fast path to action, not a sales pitch from scratch. Do not write five variants that are the same headline with synonyms swapped — if two entry states would genuinely see the same headline because the difference doesn't change what matters to them, say so explicitly instead of manufacturing an artificial difference. Do not include anything from the "must never appear" list. Do not fabricate a statistic, urgency claim ("only 3 left"), or guarantee not implied by the value proposition given.
OUTPUT: one row per entry state (Entry State / Headline / Subheadline / One-sentence hypothesis this variant tests), then flag any two entry states you judged should share a variant and why.Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Most hero-headline generation prompts produce variants that differ only in phrasing because the model is optimizing for "different enough to be five distinct outputs" rather than "different enough to test a real hypothesis," which produces A/B tests that can't actually explain why one version won even when they do — this prompt forces each variant to be anchored to a distinct, real visitor entry state with a stated reason that state needs different information, which is the actual variable that should be driving headline differences in a real test plan. The instruction to explicitly flag when two entry states should share a variant rather than forcing artificial differentiation matters because a model asked for N variants will produce N distinct-sounding outputs even when the honest answer is that two segments don't actually need different treatment — surfacing that as an option prevents shipping a fake test design that dilutes traffic across variants that don't test anything meaningfully different. Grounding each hypothesis in what's mechanically true about that entry state (cold traffic has zero context and needs the value prop explained from scratch; warm retargeted traffic already has context and needs a reason to return now) gives GPT-5.1 a concrete decision rule instead of leaving it to guess generically at "personalization," which is how most default hero-copy output ends up interchangeable across segments. The ban on fabricated urgency claims like fake scarcity counters a specific known failure mode where models asked for conversion copy will reach for manufactured urgency language by default, which is both a trust risk and, for many ad platforms and jurisdictions, a compliance risk if the scarcity claim isn't real.
What you get back
Entry State: Retargeted trial-abandoners. Headline: "Still juggling three tools at invoice time?" Subheadline: "Pick up where you left off — one shared view of client work and billing, ready in the account you already started." Hypothesis: this segment already understood the value prop once; the test is whether naming the specific friction that likely stalled them gets them back into the funnel faster than a generic "welcome back."
Verified against
ChatGPT GPT-5.1 · 2026-08-11
Changelog
- 2026-08-11 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — Google Ads management, built and maintained for you — that's Scult's day job.
EXPLORE GOOGLE ADS MANAGEMENT
