Original research by Bagga, Farias, Korkotashvili, Peng, Wu.
30-second rundown:
- DO rewrite product descriptions using iterative prompt optimization. Gut instinct alone will cost you ranking positions and money.
- DO laser-aim your product descriptions at user intent and position your product as superior to competitors.
- DON’T assume your best rewriting instincts are correct. The research found that even expert-written heuristics were outperformed by a machine that optimized itself through trial and error.
Bagga, Farias, Korkotashvili, Peng, and Wu — sleepless academics out of Columbia and MIT — dragged 7,000 real consumer queries from Reddit’s BuyItForLife community.
They matched those queries to Amazon product listings, asked GPT-4o to rank the products based on their descriptions, tested 15 rewriting strategies, and then unleashed a self-correcting optimization algorithm on the wreckage.
This is what they found:
1. Write for the Actual Human, Not the Keyword.
If your listing says “stylish footwear for every occasion,” you are invisible.
If it says “durable leather slip-on with reinforced stitching built to outlast two years of hard weekly use,” you are a candidate.
Lead with the problem your product solves. Use the language your customer reaches for at midnight when they are frustrated and finally ready to buy.
2. Pick a Fight. Then Show the Receipts.
The best-performing strategy positioned products as superior to competitors without naming them, hitting +1.61 ranking positions after optimization.
Stop describing your product and start arguing for it.
Then produce evidence: a star rating, a warranty, a material spec.
AI ranking engines are like suspicious bartenders. They have heard every story.
Show them the receipt or go home.
3. Structure Is a Weapon.
Every optimized rewriting prompt, regardless of starting point, converged on scannable formatting: headers, bullets, a strong claim up front.
The algorithm has no patience for literary atmosphere. Neither does the customer.
The generative engines are ranking your products right now, and for the first time somebody has actually mapped the terrain.
Clear intent, competitive positioning, structure: the algorithm wants exactly what your best customer wants.
Original research by Bagga, Farias, Korkotashvili, Peng, Wu.
