This paper shows you can predict real purchase intent (90%...
Robert Youssef@rryssf
90 views
Jan 13, 2026
~2 min read
Advertisement
1
This paper shows you can predict real purchase intent (90% accuracy) by asking an LLM to impersonate a customer with a demographic profile, giving it a product & having it give impressions, which another AI rates.
No fine-tuning or training & beats classic ML methods.
This is BEYOND insane:
No fine-tuning or training & beats classic ML methods.
This is BEYOND insane:
2
Consumer research costs companies BILLIONS annually.
Traditional surveys suffer from biases, take weeks to run, and need hundreds of real participants.
But researchers just found a way to simulate thousands of synthetic consumers that think like real humans.
Traditional surveys suffer from biases, take weeks to run, and need hundreds of real participants.
But researchers just found a way to simulate thousands of synthetic consumers that think like real humans.
3
The breakthrough is called Semantic Similarity Rating (SSR).
Instead of asking LLMs for direct 1-5 ratings (which produces garbage), they let the AI write natural impressions first.
Then map those impressions to scores using embedding similarity.
Instead of asking LLMs for direct 1-5 ratings (which produces garbage), they let the AI write natural impressions first.
Then map those impressions to scores using embedding similarity.
4
Here's how it works:
Prompt: "You're a 35-year-old female, income $75k, interested in skincare"
Show product image
AI writes: "I love the natural ingredients but the price seems high..."
System maps text to rating using semantic similarity
Zero training data needed.
Prompt: "You're a 35-year-old female, income $75k, interested in skincare"
Show product image
AI writes: "I love the natural ingredients but the price seems high..."
System maps text to rating using semantic similarity
Zero training data needed.
5
They tested this on 57 real consumer surveys from a major corporation (9,300 actual human responses).
Results?
- 90% of human test-retest reliability
- KS similarity > 0.85 (near-perfect distribution match)
- Outperformed LightGBM trained on demographic data
Results?
- 90% of human test-retest reliability
- KS similarity > 0.85 (near-perfect distribution match)
- Outperformed LightGBM trained on demographic data
6
The wild part: demographics matter MASSIVELY.
When they removed demographic conditioning, accuracy collapsed.
Age and income level predictions mirror real human behavior patterns almost perfectly.
The AI actually understands how different people think about products.
When they removed demographic conditioning, accuracy collapsed.
Age and income level predictions mirror real human behavior patterns almost perfectly.
The AI actually understands how different people think about products.
7
This destroys traditional market research economics:
Old way:
• $50-200 per survey respondent
• Weeks to recruit panels
• Geographic/demographic limitations
• Biased responses from "professional survey takers"
New way:
• API costs: ~$0.01 per synthetic consumer
• Instant results
• Unlimited scale
Old way:
• $50-200 per survey respondent
• Weeks to recruit panels
• Geographic/demographic limitations
• Biased responses from "professional survey takers"
New way:
• API costs: ~$0.01 per synthetic consumer
• Instant results
• Unlimited scale
8
But here's what makes this genuinely useful:
The synthetic consumers provide RICH QUALITATIVE FEEDBACK explaining their ratings.
You don't just get a number you get actual reasoning like: "The eco-friendly packaging appeals to me, but competitors offer similar quality at lower prices."
The synthetic consumers provide RICH QUALITATIVE FEEDBACK explaining their ratings.
You don't just get a number you get actual reasoning like: "The eco-friendly packaging appeals to me, but competitors offer similar quality at lower prices."
9
The implications are massive:
• A/B test 1,000 product concepts overnight
• Simulate market reactions before manufacturing
• Test messaging across demographic segments instantly
• No more waiting months for consumer feedback
Concept-to-market cycles just got 10x faster.
• A/B test 1,000 product concepts overnight
• Simulate market reactions before manufacturing
• Test messaging across demographic segments instantly
• No more waiting months for consumer feedback
Concept-to-market cycles just got 10x faster.
10
Paper: "LLMs Reproduce Human Purchase Intent via Semantic Similarity Elicitation of Likert Ratings"
The synthetic consumer era just began.
Real market research panels might be obsolete within 2 years.
The synthetic consumer era just began.
Real market research panels might be obsolete within 2 years.
11
Your premium AI bundle to 10x your business
→ Prompts for marketing & business
→ Unlimited custom prompts
→ n8n automations
→ Pay once, own forever
Grab it today 👇
godofprompt.ai/complete-ai-bu…
→ Prompts for marketing & business
→ Unlimited custom prompts
→ n8n automations
→ Pay once, own forever
Grab it today 👇
godofprompt.ai/complete-ai-bu…
12
I'm moderating a panel at AI Skills'2026 to break down what's dead vs. what's next.
Jan 22. 3,000+ attendees. 4+ hours. Free.
Save your free seat: conf.cosprints.ai/?3
Jan 22. 3,000+ attendees. 4+ hours. Free.
Save your free seat: conf.cosprints.ai/?3