A synthetic buyer is an AI agent modelled on a specific real buyer type that reads your page and reacts the way that buyer would: noticing what reassures them, stalling where they doubt, and leaving where the page stops answering their question. It is not a generic chatbot given a costume, and it is not a stand-in for a real research participant. It is a way to see your page through one buyer’s eyes before you have any traffic.

What it looks like in practice

Point a synthetic buyer at a landing page and it moves through the page in sequence, the way a person scans rather than the way a crawler indexes. A synthetic economic buyer scanning a pricing section stalls the moment it can’t find a number. A synthetic sceptic stalls at an unproven superlative with no evidence behind it. Same page, different reactions, because each is shaped around a different buyer’s priorities.

The output is not a score out of ten. It is a running account: here is where this buyer leaned in, here is the sentence that raised a doubt, here is the point they would have left.

How it differs from things it resembles

Synthetic buyerGeneric AI personaReal research participant
Shaped arounda defined buyer type and its prioritiesa broad “act like a customer” promptan actual person you recruited
Reads the page asa sequence, spending attentiona block of text to summarisethemselves, in real time
Best forstructural friction, fast, pre-trafficquick gut-checksground truth on emotion and intent
Cost and speedminutes, no recruitingminutesdays and real budget

The distinction from a generic persona matters. A generic prompt tends to be agreeable and summarises what a page says. A synthetic buyer is built to react as one type with one set of concerns, which is what makes it disagree with your page in useful, specific ways.

Why it matters

Before launch you have no analytics, so the usual signals for where a page loses people simply don’t exist yet. A synthetic buyer gives you a reaction to work from in under 10 minutes, on any URL including a staging or preview link. The named problems it surfaces map onto the Friction Index taxonomy, so “this feels off” becomes a specific thing you can fix. It fits inside the wider practice of pre-launch conversion testing.

Where it stops

A synthetic buyer is strong on structural friction: whether a value proposition is legible, whether a claim has support, whether a price is findable. It is directional only on emotional nuance and on exact pricing sensitivity, and it does not replace real user research once you have traffic. Treat a strong reaction as a prompt to investigate, not a verdict. You can see a full worked example in the sample report.

  • Buyer panel — several synthetic buyers run against the same page, so you see how each type moves through it.
  • Attention budget — the model of finite attention that depletes under friction and refills under clarity.
  • Friction Index — the named taxonomy of twelve friction types a synthetic buyer reports against.

Frequently asked questions

Is a synthetic buyer the same as a customer persona?

No. A persona is a static description you write and keep in a document. A synthetic buyer is an agent that actually reads your page and reacts, so it can surprise you by stalling somewhere your persona document would never have predicted.

Can a synthetic buyer tell me whether people will pay my price?

Only directionally. It reliably flags whether a price is findable, understandable, and comparable, which is structural. Actual willingness to pay depends on budgets and alternatives that real buyers carry, so treat any signal about price level as a starting point for research rather than an answer.

How is this different from asking ChatGPT to review my page?

A general model asked to review a page tends to summarise and reassure. A synthetic buyer is deliberately shaped around one buyer type’s priorities and reads the page as a sequence, spending a modelled attention budget, which is what makes its objections specific rather than polite.