Most landing-page copy fails for the same reason: it’s written from the inside out. The founder describes the product the way they think about it, in their own words, and then wonders why it doesn’t land. The people reading it don’t use those words. They describe their problem in their own language, on their own turf, and the gap between the two is where conversions leak out.
There’s a simple fix, and it’s now a same-day job. Go to where your customers already talk about their pain, take the exact words they use, and put those words back in front of them.
The loop
- Find the watering hole. Somewhere your customers already talk about the problem unprompted: a subreddit, review pages for competing products, support tickets, a Facebook group. You want raw complaints, not survey answers.
- Pull the raw talk. Have an LLM write you a scraper, or export the reviews. Volume matters here; you want hundreds of posts and their comments, not a curated sample.
- Feed the pile to an LLM with a large context window and ask what emotionally drives these people to buy. What they’re afraid of, what they’ve already tried, what finally made them pay.
- Have the model point you to the highest-signal posts, then read those yourself. This step is not optional. The summary tells you the themes; the raw posts give you the actual phrases.
- Rewrite the page so it describes the reader the way they describe themselves, in their vocabulary. Make it mobile-first, because that’s where they’ll see it. Then A/B test it against the old copy and let the numbers settle the argument.
Why it works
The old version of this, voice-of-customer research, was real work: surveys, interviews, transcribing calls, weeks of it. The bottleneck was never the idea, it was the labor of turning a mountain of raw customer talk into a clean map of what actually drives a purchase. That labor is what collapsed. An LLM reads the whole forum in a sitting and hands you the drivers back in structured form.
But notice the step that keeps it honest. You don’t publish the model’s summary. You have it point you to the specific comments, and you read those yourself. That’s the difference between copy that sounds human and copy that sounds like a summary of humans. The LLM is the index, not the author.
The honest caveats
- Scraping a platform at scale runs into rate limits, and every platform has its own terms about automated collection. Know the rules of whatever you’re pulling from before you point a script at it.
- The step most people quietly drop, reading the raw posts yourself, is the one that makes the whole thing work. Skip it and you’re back to writing from the inside out, just with extra steps.
- Picking the right watering hole and asking the model the right questions still take judgment. A dead forum or a vague prompt gives you generic answers dressed up as research.
Bottom line: your customers have already written your best copy. It’s sitting in a forum thread right now, in their words, describing their own pain better than you ever will. The job isn’t to invent the language. It’s to go find it and give it back.