"Will AI recommend us?" Guessing does not answer that. So we measured it.
How we measured
Twenty English-language category terms typical of Chinese export suppliers — robotics, moulds, machining, castings, extrusions, PCB, lighting, solar, hydraulics, packaging, valves, sheet metal, silicone, batteries. Queried in the US market on desktop, through a search data API that returns the AI answer, recording every source cited.
Date: 1 October 2026. One sampling round. Raw data kept locally.
One caveat up front: this is a sample, not a trend. AI answers change daily and vary with phrasing. Read the numbers below for structure, not for exact proportions.
Finding 1: More than half the terms got no AI answer at all
Of 20 terms, one request failed, leaving 19 valid. Of those:
- 11 produced no AI answer
- 8 produced one
Almost six in ten procurement queries returned an ordinary list of results with no generated answer attached.
And which ones lacked an answer? Precisely the most typical procurement terms:
injection molding supplier · aluminum extrusion manufacturer · hydraulic cylinder supplier · packaging machine manufacturer · industrial valve supplier · precision sheet metal fabrication · steel fabrication services · plastic injection molding company
This matters more than anything else here: not every term has an AI position to compete for. Some categories have no battlefield yet. Before investing, test a handful of core terms — it costs almost nothing and saves a great deal of wasted effort.
Finding 2: Half the "citations" do not point to content sites
Across the 8 terms with an AI answer, there were 104 citations in total.
Sorted by destination:
- 44 pointed to Google's own shopping cards and map interfaces
- 42 pointed to third-party websites
In other words, more than four in ten of the citation numbers you see inside an AI answer do not lead to someone's content page — they lead to a search-engine feature.
The most extreme case: one battery-category term returned 24 citations, 20 of which were Google's own shopping entries.
Why this matters for judgement: if you track only the total "am I being cited" number, you will overestimate the opportunity for content sites, and you can be misled by something as crude as a rising citation count. Strip out the house interfaces. What remains is the real content competition.
Finding 3: Third-party citations are extremely dispersed
42 third-party citations, spread across 41 different domains. One domain appeared twice. Every other appeared once.
There is no incumbent. Ask the same question ten times and you may get ten different names.
For anyone not yet in the game, that is the most useful thing in this report: the seats are empty.
Finding 4: What gets cited is a peer supplier's own website
The detail worth dwelling on.
For the castings term (stainless steel castings), all 6 third-party citations were foundry websites: barron-industries, lincolnfoundry, milwaukeeprec, rlmcastings, stainlessfoundry, waukeshafoundry.
Specific factories. Not industry portals, not encyclopaedias, not marketplaces.
For the PCB term (pcb assembly services), all 6 citations were PCB service providers: pcbway, pcbunlimited, sierraassembly, protoexpress, pcbnet, pcbassembly. One of them, pcbway, is a Chinese brand.
The pattern is clear: in this category, AI cites the service provider's own website when it explains the work clearly. The bar is not brand size. The bar is whether there is a clear, quotable page.
That also dismantles a common assumption — that AI only cites big platforms. On procurement questions, the opposite is true.
Finding 5: Citation quality is uneven
For one term (oem electronics manufacturer), the citations included established manufacturers like IBM and Zebra, alongside:
- a job board
- a financial news site
- a Hong Kong ETF fund page
The last three have nothing to do with finding an electronics contract manufacturer. They were cited anyway.
What that tells you: the mechanism is immature. It is nowhere near selective. For anyone willing to produce real content, that is the window — the bar today is far lower than it will be.
Three things to do now
1. Test your own terms. Take five to ten core category terms, ask AI each one, and note whether it answers and whom it cites. Terms that get answers deserve investment. Terms that do not should get ordinary SEO and a better website first.
2. Strip the search engine's own interfaces out of your citation counts. What is left is your real competition.
3. Build one product page that can be cited. Castings and PCB both showed the same thing: what gets cited is a site that states products, specs and capability clearly. You do not need to be the biggest name in the field.
Limits of this measurement
Stated plainly, so the numbers are not misused:
- One day, one round: AI answers vary with time and phrasing
- One search engine: other AI engines draw on different source pools
- US market, desktop only: mobile and other countries may differ
- Citations only, not the answer text: whether AI names a brand in its prose is a separate question
So do not treat these figures as "AI citation rates." Treat them as a structural observation: which terms have a battlefield, and who is standing in it.
To turn "can be cited" into something concrete on your site, see GEO · AI search optimization. To have me run a round against your own core terms and tell you which deserve investment, ask for a free acquisition audit.
Two related reads: tracking the visitors those AI answers send, in how to tell whether an inquiry came from ChatGPT; and choosing terms that match how buyers actually speak, in stop building keyword lists from your internal product names.
FAQ
How was this measured?
Twenty English-language category terms typical of Chinese export suppliers, queried in the US market on desktop through a search data API, pulling the AI answer and recording every cited source. Collected on 1 October 2026, one sampling round.
Why did more than half the terms get no AI answer?
That is the finding, not an inference. The triggers for AI answers are not publicly documented and vary by category and phrasing. Test your own terms before investing — some categories have no AI battlefield yet.
What does the dispersion of citations tell us?
That nobody has locked the space up. 42 third-party citations spread across 41 different domains, with almost no site cited more than once. For anyone not yet in the game, that is good news: the seats are empty.