How the Niche Panel works
A transparent look at how LandingCrush generates AI Panel Previews, what makes them different from a single LLM, and how we track accuracy publicly.
This is not ChatGPT giving you feedback
When you ask ChatGPT to review your landing page, you get one perspective: a single, generalised AI voice with no specific buyer context and no deliberation, unless you set those up yourself.
The LandingCrush Niche Panel is different. It generates 50 distinct AI personas matched to your specific niche: each with a different role, budget, pain point, and scepticism level. They are simulated across three deliberation rounds, can change their votes, and end in a written verdict.
ChatGPT review
LandingCrush Panel
The 5-step process
You submit your variants
Upload 2 to 4 screenshots of your landing page variants. Add a title, select your niche (SaaS, E-commerce, Coaching, Agency, Health, Finance, or Education), and optionally provide the live URL. Variants are stored securely and never shared publicly without your permission.
50 niche-matched personas are generated
The swarm engine generates a pool of 50 AI personas matched to your niche. The model then simulates a sample of 12 of them, and their votes are scaled to the 50-persona panel. Each persona has a distinct archetype, skeptical CTO, budget-conscious SMB founder, growth-stage VP, first-time buyer, enterprise procurement lead, with a specific pain point, decision-making style, and scepticism level. No two personas are identical.
Three deliberation rounds run
In Round 1, each persona independently evaluates all variants and casts an initial vote. In Round 2, personas see a summary of the group's initial votes and revise their assessment: this is where disagreement surfaces. In Round 3, personas make their final vote, producing the final tally.
Scores are calculated and a verdict is written
The panel produces a Crushability Score (0 to 100) and a click-intent percentage for each variant, plus notes on Click Gravity, scroll traps, and bounce triggers. A Panel Verdict summarises the winner's key advantages in plain language, along with the top 3 strengths and weaknesses of each variant.
Your report is delivered
The full LandingCrush Report is available immediately at /test/:id. It is clearly labelled as an AI Panel Preview. As your page accumulates real community votes on LandingCrush, the AI prediction is verified against the real outcome and the result is published on the /ai-accuracy page.
How scores are calculated
| Score | What it measures |
|---|---|
| Crushability Score | 0 to 100. Calculated from the variant's share of the panel's final votes (up to 60 points) plus the panelists' average click intent (up to 40 points). |
| Click intent | The share of the scaled 50-persona panel that chose the variant. |
| Click Gravity notes | Written notes on how strongly the primary CTA pulls a click: language, placement, and contrast. |
Accuracy tracking
We do not ask you to trust the panel blindly. Every AI prediction is tracked against real community votes. When a test accumulates 50 or more real votes on LandingCrush, the community winner is confirmed and compared to the AI Panel's prediction. The result, correct or incorrect, is published publicly on the/ai-accuracy pagewith no filtering or cherry-picking.
Every published result stays on the accuracy page, so you can judge the panel's record for yourself.
Ethical framework
Always labelled as AI
Every report generated by the Niche Panel is clearly labelled as an AI Panel Preview. We never present AI-generated votes as real human opinions.
No data sold or shared
Your landing page screenshots and test results are not sold, and are not displayed publicly without your explicit consent. Screenshots are sent to our AI model provider to generate the panel report.
Accuracy published unconditionally
We publish accuracy data whether the panel is right or wrong. There is no minimum accuracy threshold required to continue offering the service: transparency is the product.
Personas are fictional
All 50 personas are entirely fictional AI constructs. They do not represent real people, real companies, or real demographic groups. Any resemblance to real individuals is coincidental.
Human votes always win
When real community votes conflict with the AI Panel prediction, the community verdict is treated as ground truth. The AI Panel is a preview tool, not a replacement for real user feedback.