1. The Experiment

Michael Taylor asked: "Can AI predict what will go viral online?" He created 1,903 AI personas based on real Hacker News users' comments and had them vote on 1,147 headlines.

"They got it right 60% of the time — 20% better than flipping a coin."

Marketing leaders say they'd use AI market research at even 70% alignment with humans. 60% is approaching useful territory.

2. Why AI Got It Wrong

Analyzing failures revealed: virality depends on luck and social momentum. AI personas evaluated headlines in isolation, but real users are influenced by upvote counts, page position, and daily trends.

"One early upvote can change everything — identical content takes completely different paths in parallel universes."

Princeton research: 70-80% of success was determined by initial "luck."

3. Practical Insights

• Use AI for iteration, not one-shot prediction: Test 10 versions to filter obvious failures; experiment with promising candidates.
• Run multiple simulations: If it succeeds 6+ times out of 8, there's real potential.
• Focus on relative ranking over absolute values: AI is stronger at distinguishing "clearly good" from "clearly bad."

4. Try It Yourself

Copy a Hacker News user's comments and use the provided prompt in ChatGPT or Claude to create a persona. Then test your ideas against this virtual focus group.

5. Conclusion

AI democratizes market research insights once reserved for large companies. But perfect prediction is near-impossible due to the chaos of social dynamics. AI is not a crystal ball — it's a tool for smarter experimentation and iteration.

"AI isn't your marketing crystal ball yet — at least, not yet."