AI Visibility Report #005 · B2B procurement · Taiwan

AI is already shortlisting suppliers — and mid-size ones are not on the list

Almost everything written about AI search is about consumer brands. So we asked the other question: when a buyer asks an AI engine for an industrial supplier, what comes back? We put 30 real procurement questions — membrane filtration, fleet monitoring systems, bicycle components, freight forwarding, functional knit fabric — to four engines, about five Taiwanese suppliers. 120 scored answers.

By Hsiang Lee · Scanned Sep 13–14, 2026
Recommendations
2
out of 120 checks, across five suppliers
Own-site citations
2
their websites were read twice in 120 answers
But the questions get answered
67%
of answers cite real sources — just not theirs

Every question was generated from the company's own homepage and checked against its actual business before scanning.

Finding 01 — Five suppliers, four engines, two recommendations.

The five are ordinary, real Taiwanese B2B companies: a membrane-filtration maker, a vehicle monitoring and inspection system vendor, a bicycle parts supplier, a freight forwarder, and a functional knit fabric mill. None is a household name; all of them are the kind of firm a procurement manager is supposed to find.

Supplier (category)RecommendedOwn site citedAnswers with no sources
Membrane filtration2 / 2419
Fleet monitoring systems0 / 2406
Bicycle components0 / 2417
Freight forwarding0 / 2409
Functional knit fabric0 / 2409

By engine: ChatGPT 0 of 30, Google AI Overviews 0 of 30, Claude 0 of 30, Perplexity 2 of 30. This is not one engine's blind spot.

Finding 02 — The questions are not going unanswered. They are being answered with the giants.

This is the part that matters. AI did not shrug at "which supplier should I use" — it answered, confidently, with the biggest names in each industry. Across the 120 answers the most-named companies were:

Named by AITimesIn answer to
DHL Taiwan / FedEx Taiwan / UPS Taiwan15 eachfreight forwarding for SMEs
Toray Industries8membrane filtration
中華郵政 (Chunghwa Post)5freight forwarding
Nitto Denko · Veolia4 · 3membrane and water treatment
KMC · Giant · Merida4 eachbicycle components

A buyer asking an AI assistant for a freight forwarder is handed DHL, FedEx and UPS. That is not a wrong answer — it is a safe one, assembled from the companies with the most written about them. The mid-size specialist that would actually quote the job never enters the conversation.

Finding 03 — In B2B, the sources are encyclopedias and directories, not vendors.

We logged every cited source. The single most-cited domain across all five categories was Wikipedia — 36 times. After it came a fleet-management vendor's site, the government vehicle registry, a shipping comparison site, a manufacturer directory, a national magazine, a B2B search engine, and — revealingly — a job board.

The five suppliers' own websites were cited twice in 120 answers. In our consumer studies the pattern was that community threads and listicles out-cite official sites; in B2B it is worse and stranger: reference works and directories out-cite everyone, and the vendors are barely in the corpus at all.

For a consumer brand, the path into AI answers runs through the places buyers discuss products. For a B2B supplier, on this evidence, it runs through the places that catalogue companies — encyclopedic entries, industry directories, trade bodies, registries. That is a different content strategy from the one every GEO article currently recommends.

Finding 04 — A third of B2B answers cite nothing at all.

40 of the 120 answers — 33% — arrived with no sources attached, written from what the model already knew. In our 13-brand consumer study the equivalent figure was 16%. B2B questions send engines to their memory twice as often, and memory is exactly where a mid-size supplier has never been written into.

Method, and what this does not show

Five Taiwanese B2B suppliers, six procurement questions each, four engines, one pass per cell: 120 scored answers, scanned 13–14 September 2026. Questions were generated from each company's own homepage, then checked by hand against what the company actually sells before the scan ran — a mismatch would have produced a false zero, so this step matters. Each answer was scored on the same rubric: is the supplier named, is it recommended, and which sources were cited.

Four limits. One: five companies is a small sample and they are not randomly drawn — they are mid-size specialists found via two agencies' public case pages, which is exactly the segment we wanted to test, but it means this measures that segment, not "B2B" as a whole. Two: a single pass per cell cannot separate a real gap from run-to-run drift; we saw it directly here — rescanning the membrane company a day later moved it from 0 recommendations to 2. Three: we count the sources an engine discloses, so "no sources" means "no sources shown". Four: Gemini was excluded, and Google AI Overviews was queried through Taiwan-located search, which gives it a local advantage the chat engines lack.

What we are confident in: across five unrelated industrial categories and four engines, these suppliers were essentially absent while the largest players in each category were named repeatedly. What we are not claiming: that AI is wrong to name DHL, or that any of these five companies is a worse supplier. The finding is about who is legible to an AI answer, not who is good.

Is your company in the answer, or is your biggest competitor?

Paste your site — one real buyer question across five engines, free, no email. The full scan runs your procurement questions weekly, with the sources behind every answer.

Run your free scan →