What AI tells a footwear materials engineer

Footwear is where Reference Grade's co-founder spent eight years developing polymers for consumer products, so it was the first segment we put to the test. We asked ChatGPT and Claude six questions a footwear materials engineer might ask, plus one about who supplies footwear TPU and one about how to manufacture a lattice midsole, and read every answer and source.

The midsole question

The first question was the classic one: "What foam should I use for a high-rebound running shoe midsole: PEBA, expanded TPU, or supercritical EVA?"

The answers were good. Both engines landed on the same framing. Claude, asked plainly, put it this way: "Use PEBA for maximum rebound at minimum weight, such as a race shoe. Use expanded TPU (eTPU) if durability and consistent performance over many miles matter more than weight." It added sensible advice about test methods: compare cyclic energy return at running-relevant loads, not just ball rebound.

Claude's plain answer named no supplier and no brand at all. The decision it helped the engineer make was a material-class decision. The supplier decision comes later, and whoever owns the clearest explanation of the class is best placed when it does.

Who got named

Across all 24 footwear-related answers (eight questions, three engine arms), one supplier's TPU brands appeared in 10. The brand most associated with PEBA foam appeared in none, even though both engines recommended PEBA for maximum rebound. In our pilot that brand only came up when the question was about medical tubing.

That's the gap between winning the class and winning the brand. The engines know PEBA, but they didn't connect it to its best-known supplier when it mattered.

Who got cited

On the six core footwear questions, supplier pages were 31% of the citations. Academic papers were 28%, and patents another 17%: a stack of foam and midsole patents, read by the engine as descriptions of the materials.

The rest is telling:

• Running media and retailers. RunRepeat, Running Warehouse, Marathon Handbook and Outside were cited for foam properties: consumer reviews explaining polymer science to engineers.
• Footwear and foam manufacturers' blogs. Several footwear and foam manufacturers publish search-oriented articles on "PEBA vs. TPU vs. EVA". On the midsole question, those pages were retrieved and cited, while only one major polymer supplier had a page cited, by one engine.
• Report teasers. Asked who supplies footwear TPU, Claude drew 5 of its 20 sources from market-research teasers and press wires, and ChatGPT cited one supplier's news page from a 2016 trade show.

What this means for footwear material suppliers

• Publish the class comparison, with your data. The PEBA vs. eTPU vs. supercritical EVA question is asked constantly, and today it's answered with numbers from blogs and running sites. A supplier that publishes energy return, density, compression set and durability under stated test methods, including where its own material is not the right choice, gives engines something better to quote.
• Connect your brand to the class. If engines recommend your material class without naming you, publish pages that make the link explicit: footwear applications, grade names and test results, in HTML.
• Mind your patents. Your midsole patents are being read as product descriptions. Plain-language pages for the commercial grades give engines the current story.
• Keep the newsroom current. A 2016 announcement shouldn't be the page that answers a 2026 sourcing question.

This was a single-run pilot; individual answers vary between runs. The paper sets out the method, the limitations and all 52 questions.