Synthetic Consumer Lab

SSR Purchase-Intent Tester

Describe a product concept. An AI panel of synthetic consumers reacts in their own words, and Semantic Similarity Rating maps those reactions to a realistic 1–5 purchase-intent distribution — plus the rationales behind it. Based on Maier et al. (2025).

Product concept

Describe the product as you would in a concept test

Panel size (synthetic consumers)

Panel seed

Larger panels are more stable but take longer.

Polling synthetic consumers… this can take 20–60s.

Survey results

–

Mean purchase intent

–

Top-2 box (rate 4–5)

–

Consumers polled

What they said

How it works. Each synthetic consumer is a persona-conditioned LLM (gemini-3.5-flash) that replies in free text. Replies are embedded locally and compared to five Likert anchor statements per reference set; cosine similarities become a probability distribution over 1–5 (averaged across six anchor sets). Direct numeric prompting collapses to "3"; SSR recovers realistic spread. Validity depends on the model having seen real discussion of the product category.