Across 15,244 landing-page visitors over five days, copy built from applicants’ answers reached its highest form-submission rate before the final CTA: 18.74%, versus 18.00% for control. Four recurring themes became cards tested near the top or bottom of the page. The historical display marked the result not significant, with an 83.68% win probability; treat the bottom placement as a direction to retest with the same copy and design, following submissions through to later outcomes.
What we wanted to improve
This was not a neglected page. It already had tested headlines, clear CTAs, testimonials, and sections explaining the business and the webinar. That made the next question more interesting: could visitors recognize their own problems in the offer? We decided to test that recognition directly instead of adding another generic benefit statement.
Our hypothesis: make the problems recognizable
We proposed a section asking “Do these sound familiar?” and presenting four common struggles from applicants’ own answers. The hypothesis was that this recognition would make the webinar invitation feel relevant. We also wanted to learn when the visitor needed that message: immediately after the first CTA or after reading the full offer.
Build the copy from real answers
We started with historical open-ended application answers, including “What are you struggling with right now?” and “What’s holding you back?” Instead of inventing pain points in a brainstorming session, we used the OpenAI API to group recurring responses into four themes and built a visual card for each. The raw answers supplied the questions and language to work from; the model helped organize them into a section the page could present clearly.
The section connected those cards with: “If these challenges feel familiar, this offer is made for you.” The useful method is to find language people already use, then check that the final statements still describe their concerns accurately. Without an application archive, draw on interviews, sales calls, surveys, support conversations or reviews.
For your own analysis, remove names, contact details and identifying personal stories before processing answers. Work with aggregated themes, check the model’s summary against the responses, and use only material you have permission to process.
Two placements against the existing page
| Arm | Experience |
|---|---|
| Control | Existing page without the section |
| Group 1 | Section after the first CTA, near the top |
| Group 2 | Section before the final CTA, at the bottom |
The two treatments used exactly the same four themes and card design. One appeared just after the first CTA; the other appeared before the final CTA. Each could be compared with the existing page to assess the added section, while a direct comparison between the treatments addressed placement. Neither comparison singled out the wording of an individual card.
Results over five days
15,244 people took part over five days. The historical PostHog display gives the following primary form-submission results.
| Arm | Exposed users | Submitted | Did not submit | Rate | Relative change | Historical 95% relative-lift interval | Win probability |
|---|---|---|---|---|---|---|---|
| Control | 5,211 | 938 | 4,273 | 18.00% | Baseline | Baseline | 11.82% |
| Group 1: near top | 5,070 | 921 | 4,149 | 18.17% | +0.92% | −4.84% to +6.95% | 58.94% |
| Group 2: near bottom | 4,963 | 930 | 4,033 | 18.74% | +4.10% | −1.79% to +10.27% | 83.68% |
The bottom placement gained 0.74 percentage points on form submissions, a 4.1% relative increase, and 4.66% on the secondary webinar-joining metric. Application-form visits were tracked as a separate secondary metric.
PostHog marked the primary and all shown secondary metrics not significant. The bottom placement’s historical relative-lift interval includes a possible decline.
What the placement suggests
We chose the bottom placement because both form submissions and webinar joins moved in the same direction, even though the primary result did not reach the historical significance threshold. That was an explicit operating decision under uncertainty. Our working explanation was that visitors first needed to understand the offer; recognition became more useful immediately before the last decision to act. A dedicated follow-up would be needed to establish that mechanism.
A focused follow-up
Measure submissions and later outcomes
Keep the language specific enough for the intended reader to recognize a situation. Then follow the person through the funnel: our application-form tests show why submissions need to be followed through to sales outcomes.
Retest the bottom section against the original page over a planned duration. Keep the themes and design fixed, measure submissions and webinar joins per assigned visitor, and choose the decision rule before launch. Our headline tests help frame message comparisons; the event start-time test changes one specific detail. Use the PostHog experiment guide for setup.