How an AI builds a supplier list
"Which companies supply…" is the question closest to revenue. It's the moment an engineer turns a material choice into a shortlist of suppliers to call. So it matters a great deal where an AI's list comes from.
One question, two answers
In our pilot we asked ChatGPT and Claude: "Which companies supply underfill and mold compound materials for advanced semiconductor packaging?" Both produced plausible lists. Their sources had almost nothing in common.
“Which companies supply underfill and mold compound materials for advanced semiconductor packaging?”
Sources behind each engine's answer
Table view
| Supplier pages | Market research & press wires | Trade press, databases & reference | Total | |
|---|---|---|---|---|
| Claude | 8% (1) | 75% (9) | 17% (2) | 12 |
| ChatGPT | 100% (17) | 0% (0) | 0% (0) | 17 |
Claude, asked to cite its sources, drew 9 of its 12 citations from market-research report teasers and press-release wires: the "underfill materials market" pages that report sellers publish to rank in search, and syndicated press releases announcing them. ChatGPT cited 17 pages, every one of them a supplier's own product or press page.
To its credit, Claude flagged the problem in its own answer: "Most of the sources below are market research reports, which vary in quality. Treat their rankings and market-share numbers as rough estimates." Most readers will remember the list, not the caveat.
The pattern wasn't confined to semiconductors. For "Who are the leading suppliers of TPU for footwear?", 5 of Claude's 20 sources were report teasers or press wires, and one of ChatGPT's citations was a supplier's news page from a 2016 trade show.
Why teasers win
Report teaser pages are built to rank. They name the companies in a market, often with rankings and market-share figures, in plain HTML with the exact phrases an AI's search will use. Supplier websites often don't have a page that plainly says "we make underfill for advanced packaging, here are our grades and what they're for." When the model searches for a list of suppliers, the page that is literally a list of suppliers wins.
The model's own queries add to this. Asked who supplies footwear TPU, Claude searched for the companies it already knew, BASF, Covestro, Lubrizol, Huntsman and Wanhua, by name. A supplier the model doesn't already know has to appear on the pages it retrieves, and those were often teasers.
What suppliers can do, without gaming it
The tempting answer is to buy your way onto the teaser pages. We'd advise against it. The rankings are opaque, the pages are low quality, and an engine that learns to discount them will discount you with them.
Better:
- Publish the page the question is looking for. One clear, crawlable page per product family that states what you supply, for which applications, with grade names and key properties. That gives an engine something better than a teaser to cite.
- Keep your newsroom current. Engines cite old press pages. Retire or update announcements whose facts have changed, and make sure your current capabilities are stated somewhere permanent, not only in a launch press release.
- Earn credible coverage. Trade publications and technical papers carry weight with engines. A substantive article about your technology is worth more than a paid listing.
- Watch both engines. They built their lists from different worlds. Winning one doesn't mean you appear in the other.
This was one run per engine on one question; lists and sources vary between runs. The paper sets out the method and limitations.
Method, data and limitations: What AI Tells Engineers →