Ask ChatGPT to "write a landing page for my product" and you'll get a neat wall of claims that could sell almost anything. The model isn't broken. It just has nothing to work with: no voice of your customer, no real proof, and no single job for the page. So it fills the gaps with the average SaaS pitch.
This post is for solo founders rewriting a one-page offer, where the visitor is supposed to buy, book, or try. It walks through what to collect before you prompt, then three prompts you can copy: one that drafts hero variants from real customer language, one that turns that language into benefit blocks and an honest FAQ, and one that roasts the draft the way a skeptical buyer would. No invented conversion stats anywhere.
Most people don't read web pages word for word. Nielsen Norman Group's research has said this for decades: visitors scan, pick out phrases that match what they came for, and leave when the page doesn't answer them quickly. A page full of "seamless," "powerful," and "next-generation" fails that scan, because nothing on it is specific enough to grab onto.
When founders paste a vague brief into ChatGPT, three failure patterns keep showing up: Feature soup. The hero lists what the product has instead of what the buyer gets. Invented proof. "Join 10,000 happy customers" or "increase conversions by 40%" appears even though you never said anything like it. No single job. The page tries to educate, sell, collect a lead, and book a call at once, so the visitor never knows which action to take.
The fix is to change the inputs, not to hunt for a cleverer prompt. Give the model the words your buyers already use, ban claims you can't back up, and give the page one primary action.
Fifteen minutes of prep beats an hour of regenerating the same generic draft. Have these ready: A one-sentence offer. Who it's for, what they get, and the primary action (buy, book, or start). Real customer language. Quotes from sales calls, support tickets, Reddit threads, competitor reviews, or DMs. Paste the exact phrases, not your polished paraphrase of them. Three objections you've actually heard. Price, "I can do this myself," timing, trust. Skip the hypothetical ones. Proof you can name. A named result, an outcome you could screenshot, a short testimonial with a real first name. If you don't have any, leave this blank. Blank is better than fake. Traffic context. Where visitors come from (an ad, a tweet, a newsletter), so the headline can match the promise that brought them.
One note on benchmarks: Unbounce's Conversion Benchmark Report, built from tens of thousands of landing pages, puts the median conversion rate across industries somewhere in the mid-single digits. That's useful as a sanity check for your own analytics later. It is not something the model should be writing into your copy.
Paste your transcripts or quotes first. Then ask for hero options that reuse the buyer's words instead of your feature list.
What good output looks like: headlines you could imagine a buyer saying out loud, and CTAs that name the outcome ("Get the pack") rather than "Submit." If any variant invents social proof, throw it out.
The five labeled angles are there for a reason. Each one comes with its own one-line risk, so you aren't just picking the variant that sounds nicest. You're picking the trade-off you're willing to make.
Once you've chosen a hero angle, build out the middle of the page from the same quotes. Features stay in the page, but as supporting detail under the benefits, not as the headline act.
Why this works: the model stays anchored to language you gave it. The FAQ handles objections in the visitor's own words, instead of turning into a legal-sounding wall of fine print. And the explicit "none" option for proof gives the model permission to leave testimonials out rather than make them up.
Before you publish, have the model attack the draft the way a tired buyer with too many tabs open would.
Run this pass once. Then go through what's left and cut anything you couldn't defend if a buyer asked "says who?" on a sales call.
The scan test in step 3 ties back to how people actually read pages. If the three phrases someone would notice in ten seconds aren't the ones that explain your offer, the page has a problem no amount of extra copy below the fold will fix.
If someone clicked a tweet about rewriting cold emails, the first line of your hero should echo that promise. It shouldn't pivot to a generic "AI toolkit for founders." Message match is simple: the headline should read like the next sentence of the ad, post, or search result that brought the visitor there.
Then pick one primary action and repeat it. Secondary links, like other guides or an FAQ anchor, are fine lower on the page. Competing buttons in the hero ("Buy," "Book a call," and "Download a sample" all at once) usually mean the visitor does none of them. Solo-founder pages win by being obvious, not by offering every possible path.
After you publish, watch the first screen. If people bounce without scrolling, the problem is the hero or the traffic match, not the FAQ. Change one thing per week so you know what actually moved.