Small Shops Just Got Better AI Search Without a Big Tech Budget

When you type something vague into a shop's search bar — "something warm for a rainy hike" — the AI behind it has to guess which of thousands of products you mean. Big marketplaces like Amazon solve this with huge, expensive systems. Small shops often can't afford that, so they try a shortcut: paste the entire product catalog into a modern AI model and ask it to pick. New research shows that shortcut works much better if you tidy the catalog first.

The paper, called RPTune, comes from researchers who studied seven real merchants across different retail categories, testing 100 tricky conversational searches per shop. Their key insight is that AI models don't read long lists evenly — they pay more attention to some parts than others. So the team built a helper that reorders and prunes products, putting likely matches where the AI actually looks. It learns from the AI's own wrong answers and improves over time.

They also gave the AI extra training on these cleaner lists, using feedback tied to each specific catalog rather than generic examples. The results: tidying alone boosted search accuracy by up to 31.4 percentage points, and the extra training added roughly 10 more points on average. It worked on both paid, closed AI models and free, open ones.

Why should you care? If you run a small shop, this points to better search without an engineering team or a giant budget. If you shop at one, fewer frustrating "no results found" moments and more of what you actually wanted. The catch is that it's a research paper, not a product you can buy today — and it only helps catalogs small enough to fit inside an AI's memory.