How to use an AI Panel preview without confusing it for the swipe
The AI Panel is a paid simulation with fictional personas and a written verdict. The swipe is a human crush or pass. Keep the columns separate, and let community votes break a tie.
Picture Miguel, a founder in Miami, pasting an AI Panel verdict into a document he has titled "Winner" on a Friday afternoon. The verdict likes variant B. The prose is confident. It names a skeptical buyer and a clarity score. Miguel is about to tell the team the test is over, even though the homepages have not been approved for a public vote and nobody has crushed or passed anything. The panel did what it is built to do. It produced an exploratory preview. The document title is the mistake. A preview is not a winner. A winner, on LandingCrush, is a claim you earn from community votes, and only after you are honest about the count.
The AI Panel Test is a separate purchase from Creator Pro. The published prices are $29 for one variant and $49 for a head-to-head of two to four variants. You upload screenshots, pick a niche, and can add a live URL. The methodology describes 50 niche-matched AI personas and three deliberation rounds. The report is labeled an AI Panel Preview. Scores you will see discussed in the product include Crushability, Click Gravity, Trust, and Clarity, plus a written verdict and strengths and weaknesses. None of that is a room of customers. The personas are fictional. If you need the longer public description, use the methodology page and the accuracy page, and treat both as descriptions of the tool, not as a promise that the preview will match your revenue.
Put the preview in a column that cannot be averaged
Open the note where the team stores homepage evidence. Make three columns and refuse to merge the cells.
Column one is the AI Panel Preview. Paste the verdict, the variant it prefers, and the scores it actually returned. Write "fictional personas" at the top so a later reader does not promote them to interviewees. Column two is the community vote: crush rate, vote count, weekly votes, velocity, and the comments that mention the change you tested. Column three is your analytics, still empty until the live page changes. If someone adds a row called "blended score," delete it. A 78 Crushability score and a 61 percent crush rate are not numbers on one scale. Averaging them creates a third metric the product does not define.
The post-approval AI design critique is not column one either. That critique runs after a homepage is approved. It scores hero clarity, call-to-action effectiveness, visual hierarchy, and mobile responsiveness from 0 to 10, with suggestions. Free submissions use the standard queue. Creator Pro uses the priority queue. It is another model note. Give it a corner of column one if you want it nearby, and label it "post-approval critique," not "panel" and not "votes."
You might be thinking the panel is close enough when you are in a hurry. Hurry is when the label matters most. A hurried team will ship the verdict and call it research. The about page says this feedback is not professional research and does not guarantee an outcome. Your note should say the same thing in shorter words. "Preview. Not the swipe. Not revenue."
Use the preview to choose what to submit, not what to conclude
The panel earns its price when the page is not ready to be public, or when you have more variants than you can ask strangers to judge this week. A $49 test of two to four screenshots can tell you which direction the simulated panel argues for, in language you can check against the image. Read the weaknesses before you read the winner line. A weakness that says the buyer is unnamed is useful even if you ignore the preferred variant. You can fix that line and then decide what to submit.
Do not submit only the panel's favorite and delete the others without a record. Save the verdict next to the files. If you submit the favorite, say you used a preview to choose a candidate. Then let the swipe test the candidate. The swipe is /swipe after approval. Right is a crush. Left is a pass. The people are registered participants, not the 50 personas. If you need two variants in the feed, Creator Pro is the plan with unlimited submissions, at $19 a month or $190 a year. The panel fee does not include that slot. They are different charges because they are different tools.
A preview is also a bad place to hide a disagreement. If the verdict prefers B and your five-second hero check fails on B, believe the check you can point at. Score the hero before you pay for language that flatters a page you cannot explain. The panel will often sound smoother than the screenshot. Smooth is not the same as specific.
Let the community break the conflict, after the count is real
LandingCrush states that community votes are the ground truth when they conflict with the panel. That rule is only meaningful when the community has actually voted. The product waits for 50 or more real community votes before it compares an AI prediction with the community result and publishes that kind of check on the accuracy page. Until you have a count in that neighborhood, do not declare that the panel "missed" or "nailed it." Say the vote is still directional. The reading of thin rates is in how to read a crush rate.
When the count is there and the columns disagree, write the disagreement without smoothing it. "The preview preferred A on clarity. Voters preferred B, 64 percent crush on 80 votes against 41 percent on 74 votes, and the comments mention the demo label." Then decide. The rule says the votes win the preference claim. You may still decline to ship B if the comments praise something you are not willing to promise. That is a strategy choice. Call it a strategy choice. Do not edit the vote column so the preview looks right.
Comments outrank scores inside the vote column. A Crushability number cannot tell you that three people misread the buyer. A roast comment can. Cluster those comments the way the roast article describes. The panel's top weakness is a hypothesis. The cluster is evidence from the feed.
Do not brief the team with persona theater
The methodology is public about how the panel is different from asking a single general chatbot for a polite review. Use that distinction if someone wants to paste the page into a chat and call it the same test. Do not use it to imply you interviewed a skeptical CTO. You did not. The archetype is generated. Quoting "the CTO persona said" in a customer meeting is how a preview becomes a fabrication. Quote the verdict as a verdict. "The panel preview argued that variant B named the buyer and variant A did not." That sentence is checkable against the screenshots. A fake quotation is not.
Niche selection is part of the input. The methodology lists niches such as SaaS, e-commerce, coaching, agency, health, finance, and education. Pick the niche that matches the page. A finance page run as a lifestyle niche is a sloppy input, and a sloppy input makes a preview that is even easier to overread. You still do not promote the output to a study.
Variants in a panel test should follow the same one-difference rule when you want a reason. Four unrelated redesigns will produce a favorite. You will not know which change the simulated round latched onto. Freeze the offer. Change one thing. If you already broke that rule, label the preview "direction pick" and not "headline test."
Close the preview when the swipe has a record
Archive the panel document when column two has a count you trust and a decision sentence. Leave the preview in the archive so you can see, later, whether it agreed. That is what the accuracy page is for at the product level. Your archive is the small version. Do not keep the preview at the top of the homepage channel in Slack after the votes exist. People will quote the smoother paragraph.
Then ship or do not ship, and measure on your site. Column three is the only column that can talk about trials, demos, or revenue. The panel cannot see it. The swipe cannot see it. If column two says voters crushed the page and column three says the demo rate fell, believe column three about the business and believe column two about the preference. They are allowed to diverge. Divergence means the voters in the feed are not the same as the visitors on your domain, which is the normal case, not a scandal.
The next time a verdict sounds finished, will you still be able to point to the column that has human votes in it?
Common questions
Q: Is the AI Panel the same as community voting?
A: No. The panel is a paid simulation with fictional personas, labeled a preview. Community voting is a crush or a pass by people in the swipe feed.
Q: What does it cost?
A: The published AI Panel prices are $29 for one variant and $49 for two to four variants. Creator Pro, which covers unlimited homepage submissions, is a separate $19 a month or $190 a year.
Q: What if the panel and the voters disagree?
A: The product treats community votes as the ground truth when they conflict. Wait until you have a real count, on the order of the 50 votes the product uses before it audits a prediction, before you call the conflict.