The rule behind every report: a number must come from a defined procedure, and a claim about what this product is must survive this page. Here is the whole machine, including its limits.
We fetch the URL (or use the text you paste) and reduce it to what a shopper sees: title, body copy, detected price, image descriptions. Personas are instructed never to invent page content. If something important is missing (reviews, returns policy, shipping costs, who runs the store), the absence itself is treated as a signal, the way a real shopper treats it.
Each of the 50 personas reviews the material alone; none sees another's answers, so there is no herd effect. Every persona answers the same structured questions: first impression, would buy (yes, maybe, no), purchase intent 1-10, trust 1-10, clarity 1-10, price reaction, biggest objection with a category, the one change that would convince them, and one blunt quote.
Scores are averaged, the buy split is counted, objections are grouped into six fixed categories (trust, price, clarity, need, shipping, design) and ranked by frequency. Quotes are selected for specificity, at most one per archetype. No model decides the summary numbers; they are math over 50 structured answers.
BUY SIGNAL requires average purchase intent of at least 7 AND average trust of at least 6. FIXABLE covers intent of at least 4.5. Everything below is WEAK. The thresholds are deliberately conservative: an inflated verdict would make the report useless, so the rules err on the harsh side.
A malformed answer gets one retry, then that persona is skipped and the report states it plainly, for example "48 of 50 personas responded". Numbers are computed over the personas that actually answered, never padded.
Fifty hand-written personas would drift and repeat themselves. Instead, ten archetypes carry a fixed shopper psychology (what annoys them, what convinces them) and five demographic variations each spread age (22 to 68), country, income and occupation, so no two voices in an archetype share a country or an age decade. The full roster with names and biographies is on the jury page: nothing about who reviews your page is hidden.
The fastest way to decide if this is useful: walk through the sample report page by page, then let five personas at your own page for free.