The listings that most need a split test are the ones that do not qualify to run one. That is the gap this fills, and nothing more. Reynard is not affiliated with Amazon in any way.
Manage Your Experiments needs Brand Registry and roughly 1,000 detail page views on that ASIN in the last 30 days. One experiment per ASIN at a time, results recalculated every week or two, US marketplaces only. So a clean read on a main image takes four to eight weeks and traffic the listing does not have.
It is a chicken-and-egg problem. The listing converts badly, so it gets few sessions, so it never reaches the threshold that would let you find out why it converts badly. Amazon staff on the seller forums are consistent about this: there is no way around the eligibility bar. The advice is to change the content manually and watch Business Reports.
That is the one situation where Reynard is not competing with a better instrument, because the better instrument is not available to you. A run reads the listing cold, per segment, in a few minutes, and tells you which version deserves the traffic you are about to buy. It does not tell you which version wins. MYE does that, later, when you qualify.
The decision sellers say they got wrong is almost always the main. Not the whole image stack, not the A+ Content — the main, and specifically how it reads at thumbnail size in mobile search results, next to nine competitors, at a glance. It is judged full size on a desktop monitor by someone who already knows what the product is. It is bought at 120 pixels by someone who does not.
The shape of the story is the same every time: months of PPC spend, an ACOS that never comes down, a rebuild of the campaigns, and only then the realisation that the traffic was fine and the listing was the problem. Sessions were arriving. Unit session percentage was the number that needed fixing.
A run puts the main image, the title and the price in front of a weighted cross-section of the population who have never seen the product, and reports what each segment thinks it is, who they think it is for, and what they think it should cost.
One run per decision. Each one is a thing you would otherwise settle by argument, or by launching and hoping.
What the product reads as before a single word is read, and whether the thing you are selling is even identifiable at thumbnail size.
Whether the title says what the product does, and which bullet a segment stops reading at.
Which segment stops accepting the price first, and whether the objection is the number or what the number implies about the product.
Whether the longer copy adds a reason to buy or repeats the bullets, and whether a parent with several children reads as choice or as confusion.
This is not an argument against PickFu. PickFu recruits real people and says so, and a poll of real respondents beats a simulation on the question of what real people prefer. The workflow most sellers describe is already a funnel: narrow six main image candidates down to two with a poll, then split test the survivors on live traffic. Reynard's slot is one step earlier than both, and it costs less.
Once you have the traffic to run a real experiment, run it. A simulation is what you use when you cannot. And when you can, the live sessions are behaviour, not opinion, and behaviour wins.
One more thing to be clear about: this reads a weighted cross-section of a population, not an Amazon shopper who searched for the product, has purchase intent and a basket open. Liking something is not buying it, and the panel is nowhere near a checkout.
There are good reasons to distrust this category. These are the three that land, and what we actually say to them.
The sharpest version of this: ask a model what matters to ordinary gym-goers and you get what fitness influencers think ordinary gym-goers want. That is the real limit of the method, and we are not going to argue it away. Census weighting fixes the composition of the panel — how much weight the loud group gets against the quiet one. It does not fix the fact that the loud group wrote more of the training data. Treat a run as a read on signal, not on demand.
The output is not one verdict. It is segments that disagree with each other and a score that comes back low often enough to be worth reading. The useful part is which segment refuses first and what it names, not the average across all eight. If every segment agrees enthusiastically, that is a reason to be suspicious of the run, not to feel good about the listing.
Correct, and we are not going to pretend otherwise. It is a wind tunnel, not a flight. You use it to find the flaw before you spend the money, not to prove anything to anyone. There is no statistical significance here and there never will be.
Read this part before you pay for anything.
A poll at the incumbent starts around a dollar per response with a fifteen-dollar floor, and a proper multi-option main image test lands well above that. Reynard is a free first check with no account, then $12, $25 or $39 for a full run, one-off, no subscription. In euro markets that is €9, €19 or €29.
That is the whole pricing argument. If a poll of real people fits the decision and the budget, run the poll.
What the main signals per segment, before the copy does anything.
Read the comparisonWhich segment stops accepting the price first, and why.
Read the comparisonNo accuracy figure has been published, and this page explains exactly why not.
Read the comparisonThe first check costs nothing and needs no account. Write the thing you are testing, pick the market, read the report.
Read the comparisonOne check costs nothing and needs no account. You see the format of the output before you decide whether it is worth paying for a full run.
Start the free check