ChatGPT Prompts for Landing Page Copy That Converts

Landing page copy is the highest-stakes writing in marketing per word: a few hundred of them stand between paid traffic and revenue, and the median page loses the argument. Unbounce’s benchmark analysis of 41,000 landing pages, 464 million visitors and 57 million conversions puts the average conversion rate at 6.6% across industries, with 10% or above the rough threshold for “good”. Most visitors bounce. The copy’s job is to lose fewer of them.

Which is why “write me a landing page for my product” is the wrong prompt. It produces the template everyone recognises: a vague benefit headline, three feature cards, a testimonial that sounds written (because it was), and “Get Started Today”. The prompts below take the opposite approach, the same one that runs through this whole library: each one generates a single section of the page from real inputs you paste - the ad that sends the traffic, the objections your prospects actually raise, the proof you actually have. The model writes sentences. You supply everything true.

The inputs that decide everything

Before the prompts, the collection pass. Fifteen minutes, and every prompt below draws on it:

• The traffic source, verbatim. The exact ad, email or post that sends visitors to this page. A landing page continues a conversation the click started, and it can’t do that without seeing the opening line.
• The offer, precisely. What the visitor gets, what it costs (money, time, an email address), and what happens in the first five minutes after converting.
• Customer language. Verbatim phrases for the problem, from calls, tickets or reviews - the same raw material as my buyer persona prompts. If you’ve run those, paste the persona.
• The proof you actually have. Real numbers, real testimonials with permission, real logos. This list is a fence: anything outside it may not appear on the page. An invented testimonial isn’t just bad practice - in most jurisdictions it’s illegal.
• The one job. Sign-up, demo booking, purchase, download. One page, one job. If you have two jobs, you need two pages.

1. The message-match prompt

Message match is the discipline of making the page’s headline continue the ad’s promise, and it’s where I start because it constrains everything else.

Here is the ad/email that sends traffic to this page: [paste it
exactly]. The offer on the page: [what they get]. The one job:
[conversion action].

Write 5 headline + subhead pairs that continue this ad's promise.
Rules: reuse the ad's key phrase or its close variant in the
headline; the subhead adds the strongest specific we have from:
[proof list]; no new claims the ad didn't set up; no cleverness
that makes the visitor re-orient.

Then flag anything on my current page (pasted here: [paste]) that
breaks continuity with the ad - promises that vanish, vocabulary
that shifts, offers that differ.

Why it works: the visitor’s first question on landing is “am I in the right place?”, and copy answers it by echoing the words that were just clicked. The continuity audit at the end is the highest-value part for existing pages - vocabulary drift between ad and page is invisible to the people who wrote both, and obvious to the model, which is exactly the analysis-over-invention work models do best. Write the ads and the page from the same inputs and this prompt gets easier - my ad copy prompts ask for the landing page’s promise for the same reason in reverse.

2. The value-prop hero prompt

Product: [what it is, in one plain sentence]. Buyer: [role and
situation, from persona evidence]. Their current alternative:
[what they do today - a competitor, a spreadsheet, nothing].
Why we win against that alternative specifically: [your honest
answer]. Proof available: [the fence].

Write 3 hero sections: headline (under 10 words), subhead (under
25, carrying one specific from the proof fence), and the CTA
button label (2-4 words, stating what happens next - not
"Submit", not "Learn More").

Rules: the headline states the outcome or names the pain - never
the category ("The all-in-one platform for..."). No claim outside
the proof fence. No superlatives without substantiation.

Why it works: the “current alternative” input is the one most landing pages are written without, and it’s the whole value proposition - you’re not selling against nothing, you’re selling against a spreadsheet someone already tolerates. Banning category headlines forces the model off its strongest cliché, and the CTA-states-what-happens rule (“Start free trial”, “Get the template”) survives every design trend because it answers the click’s real question: what am I agreeing to?

3. The benefits-block prompt

Features to cover: [list them, plainly]. Buyer: [paste persona or
customer quotes]. Their words for the problem: [verbatim phrases].

For each feature, write a benefit block: a heading in the
customer's vocabulary (their words, not ours), then 1-2 sentences
connecting the feature to the outcome they described. Order the
blocks by which pains appear most often in the evidence.

Rules: if a feature maps to no documented pain, put it in a
"didn't earn a block" list instead of forcing a benefit. No block
may work as copy for a competitor - if it could, it's too generic;
rewrite or flag it.

Why it works: feature-to-benefit translation is the mechanical half, and models do it fine. The two rules are the editorial half: the “didn’t earn a block” list keeps the page focused on what the evidence supports (and quietly tells you which features your marketing shouldn’t lead with), and the competitor test is the same specificity check I apply to every AI draft - copy that could run on a rival’s page is copy that sells the category, and the category doesn’t pay you.

4. The objection-handling prompt

Here are real objections from prospects, verbatim: [paste from
sales notes, review complaints, churn reasons - yours and
competitors'].
The one job of the page: [action]. Proof fence: [list].

Rank the objections by how likely they are to block THIS action.
For the top 3-4, write a short section for the page: name the
objection plainly (in the visitor's words, as a heading if it
earns one), answer it with evidence from the fence, and say
honestly what we don't solve if the evidence is thin.

Rules: no strawman objections. No answer without evidence -
an unanswerable objection goes in a list for me, not on the page.

Why it works: the visitors you lose at 6.6% aren’t unconvinced of the benefit - they have a specific unanswered doubt, and pretending it doesn’t exist doesn’t make it leave. Ranking by this action matters too: “is it secure?” blocks a purchase but rarely blocks a template download, and objection sections tuned to the wrong stakes read as defensive. I hand the unanswerable list to product and sales. Some objections are roadmap, not copy.

5. The proof-selection prompt

Here are our real, approved testimonials and case results: [paste
them, with what we're permitted to show].
The page's claims, in order: [paste the hero and benefit blocks].

For each claim, pick the single strongest piece of proof and
place it: which testimonial or number goes next to which claim,
and why. Trim testimonials for length by CUTTING sentences only -
never rewrite, never merge, never "improve" a quote. Mark claims
that have no proof - do not fill the gap.

Why it works: proof placed next to the claim it supports converts doubt at the moment of doubt. Proof in a carousel at the bottom decorates. The no-rewriting rule is absolute for the same reason as in my caption prompts: a model asked to tidy a quote will improve it into something the customer never said, and an improved testimonial is a fabricated one with a real name on it. The unproven-claims list is your pre-launch checklist - each one either gets evidence or gets softened.

6. The friction prompt

The conversion action: [paste the form fields / checkout steps /
signup flow, exactly as they are]. What we do with each field:
[honest answer per field].

Write: (1) microcopy for each point of hesitation - what happens
after submit, why we ask for [phone/card/company size], what it
costs, how to undo it; (2) the confirmation message; (3) a list
of fields whose stated purpose is weak - candidates to cut.

Rules: reassurance must be true ("no card required" only if no
card is required). No urgency we cannot substantiate.

Why it works: the copy around the form is read more nervously than any headline, and it’s usually written last, by whoever built the form. Asking “why do we ask for this?” per field does double duty - it generates the microcopy and audits the form itself, and the field-cutting list is where I see the biggest wins in practice - more than anything above the fold. The truth rules exist because models generate reassurance as easily as any other string, and false reassurance next to a form is where trust dies with the page’s one job unfinished.

7. The skeptic’s teardown

Here is the assembled page copy, top to bottom: [paste]. The
visitor: [persona], arriving from: [the ad], deciding whether
to: [the job].

Read it as that visitor in a hurry and sceptical by default.
Mark: where you'd stop reading and why; every claim you'd doubt
and what proof would fix it; the first moment you understand
what this actually is (if it isn't the headline, say so); the
question you still have at the CTA.

Being agreeable is a failure. If the page is fine, find the
weakest section anyway.

Why it works: the same model that writes agreeable copy critiques honestly when scepticism is the assignment - the pattern behind the skeptic prompts across this library. The “first moment you understand what this is” check catches the most common landing page disease: a page that’s all outcome and no object, where the visitor reaches the pricing section still unsure what’s being sold. I run it before every launch, and occasionally on pages that have been live for a year - a humbling twenty minutes, every time.

From draft to numbers

Two rules I hold once the copy exists. First, this chain produces the candidate page: the brand voice check still runs on the assembled draft, and a human still reads it aloud before it ships. Second, test angles before wording. The hero prompt gives you three heroes - test those against each other first, because a headline angle is worth multiples of a verb choice, and at typical traffic volumes you only get so many honest experiments a quarter. The 6.6% average from the Unbounce data is not a ceiling - it’s what pages written without these inputs converge to.

The page is also only half the argument - the other half happened in the ad, which is why the ad copy prompts and these share their inputs, and why the traffic that lands here should have been chosen with real keyword and intent work upstream. For where landing pages sit in the whole practice, start with the complete guide to AI marketing.

FAQ

How long should landing page copy be? As long as the decision is expensive, and no longer. A free template needs a hero and a form. A five-figure annual contract earns objection sections, proof and detail, because the visitor’s real alternative is “research more”, not “buy now”. The objection-ranking prompt effectively answers this per page: the copy should be exactly as long as the doubts it has to clear.

Can I use these prompts for the whole page in one pass? The sections chain on purpose - the hero constrains the benefits, the benefits get proof assigned, the assembled page gets torn down. One mega-prompt skips the checking between stages, which is where the quality lives. Same argument as the brief prompts: the chain is slower because the checking is the point.

Do these work for a homepage? Partially. A homepage serves multiple audiences and jobs, so the one-job rules above will fight you - that’s not the prompts failing, it’s the definition of the page. Use the teardown and friction prompts on any page. Reserve the full chain for pages with one traffic source and one job, which is what a landing page is.