Berman says only 2 of the top 100 ad picks had 50% off
A simulated 100-shopper panel picked an undiscounted Black Friday ad more often than the half-price one.
Matthew Berman posted the discount breakdown from his Jev agent's run over 3,092 real Meta ads. Of the top 100 picks, only 2 carried a 50% off offer. The run itself is the one we covered earlier, and this is the number that came out of it.
The smaller panel result is the one that makes the point. Running 100 simulated shoppers against a set of ads, 32 picked a Black Friday ad with no discount at all. The 50% off ad got 12. That is the gap Berman is pointing at, a deep discount losing to offers that did not cut price, inside the same panel.
The prompt he is telling people to copy
Berman's suggested test is cheap enough that there is no reason to argue about it in the abstract. Paste your offer and five competitor ads into Claude, then ask it to act as 100 different people who buy in your category, have each pick the one ad they would click, tally the picks and explain why the winner won. That is the whole method. The 3,092 ad run is the same idea at scale with a Jev agent doing the ranking.
What the number is and is not
These are language model personas picking ads, not customers spending money. Nothing in the post ties the picks to click-through, conversion or revenue, and there is no holdout against real campaign data. So the honest read is that this is a screen you run before you commit to a price test, a way to notice that your discount might not be the thing doing the work, not evidence that it is not.
For people building on it, the two numbers are measuring different things. The 2 out of 100 is a property of how Berman's agent ranked a large ad corpus. The 32 versus 12 is a single panel on a small set of ads. They point the same direction, which is interesting, but they are not the same experiment and stacking them into one claim about discounting would be overreading a single day's run.
32 of 100 AI shoppers picked a black friday ad with no discount.
my agent w/ jev ran this on 3,092 real meta ads. of the top 100 picks, only 2 had 50% off.
Know what Jev is good at before you bet on it
One email a day. Under three minutes. What builders found out about Jev yesterday.