Over five weeks this summer, we monitored how four AI assistants answered buyer-style questions for a design partner: an established, mid-market B2B food-ingredient supplier. We asked what a product developer or a buyer in procurement might type when sizing up suppliers. Questions about sugar alternatives and clean-label recipes. Questions about taste, aftertaste and how an ingredient performs, the details that decide a shortlist.
I expected to write about a number moving. That is not the story the data told. The story is a gap, and it barely moved at all.
Why I think this matters for ingredient suppliers
I want to be careful here, because it's easy to overclaim. My data measures what AI *answers*. It does not measure how many of your buyers now do their research in an AI assistant.
What I can say is narrower, and still useful. When buyers ask these tools "who makes a clean-label sweetener system without a bitter aftertaste," the answer names a few companies. If you're not one of them, you weren't on that buyer's list for that answer.
For a category built on trust, specs and long sales cycles, missing from that first answer is a real cost. It is also an easy one to miss.
How we measured it, plainly
Over five weeks, for each product area, we sent a fixed set of buyer questions to four AI assistants. We logged every answer and the sources it cited.
We split the questions into two kinds. **Branded** questions name the supplier ("tell me about *[this supplier]*"). **Unbranded** questions don't ("recommend a supplier for *[this problem]*"). Branded questions measure **Recognition**: does AI know you when you're named? Unbranded questions measure **Discovery**: does AI bring you up when nobody names you?
First, data quality, because most AI visibility numbers skip it. Not every monitoring run came back complete. Some returned partial data and a few failed. Midway, we also revised the questions, so early and later runs can't be compared directly.
So I used only complete runs, within a single product area and a single version of the questions. I also re-counted every mention from the raw saved answers. Everything below survives that filter.
Recognition is AI answering questions about you. Discovery is AI giving you one of the few places on its shortlist.
Finding 1: AI knew the company by name. It almost never recommended it unprompted.
When named, the supplier was recognized in nearly every check. AI knows the brand.
When not named, it nearly vanished. In most complete runs, it never came up on unbranded questions at all. It did surface in most product areas, but only a handful of times, against a large number of answers. Not "declining." Just barely there in the answer AI gives a buyer who hasn't heard of you yet.
That gap, recognized when named and missing when not, held in every product area and on every AI assistant.
Finding 2: Every unprompted mention came from one AI assistant.
I checked this finding hardest. It is easy to assume and wrong to state loosely. I pulled every run where the supplier came up unprompted and re-checked each one against the raw answers.
Every one of those mentions came from the one AI assistant that searches the live web while it answers. The other three never named the supplier unprompted, in any run.
Those three answer from what they already "know." On unbranded questions, they didn't bring this supplier up. If you treat "AI" as one thing, you miss this. For this supplier, being discovered was only possible on the assistant that searches the web.
Finding 3: An idea about sources, stated as an idea.
In almost every run where the supplier came up unprompted, the answer listed the supplier's own website among its sources. And in the runs I re-checked, only the web-searching AI assistant cited sources at all.
That is a correlation. It does not prove cause, and one design partner's data does not settle it. But it is a specific idea I want to keep testing. Unprompted mentions tracked whether the web-searching assistant had the supplier's own material to cite. When the supplier's own content was within reach, it sometimes came up. When the cited sources were third-party research, regulatory pages or a competitor's website, it didn't come up at all.
What changed, and what I would not conclude
Almost nothing moved, and I won't dress it up. Across comparable runs, the overall picture held flat.
The one thing that *did* move was surface-level. The competitor named first changed almost from run to run. A small group of large, long-established ingredient companies traded the top spot while the overall picture stayed put.
What I would **not** conclude is that anything trended, or that monitoring itself moved a number. We changed the questions midway, and many runs came back incomplete. The one small uptick I saw is tiny and muddied by that change. Claiming a trend here would be exactly the overreach I'm arguing against.
What this means if you supply ingredients
Three things I'd stand behind:
**Recognition is not Recommendation.** Being known is not being recommended. This supplier's Recognition was strong and its Discovery was close to nothing, at the same time.
**Be among the sources AI cites.** In the runs I re-checked, unprompted mentions came when the supplier's own content was cited, and only on the web-searching AI assistant. If your material isn't where that assistant looks for your category, you may not be recommended, however established you are. Worth testing on your own name.
**Watch the trend over time, not one scan.** A single scan can mislead you. The top competitor changing from run to run shows why. The pattern across weeks is the only honest read.
The close
This is why we run this as ongoing monitoring, not a one-time audit. A single snapshot would have said "recognized, competitors present" and stopped. Five weeks of complete runs showed something more useful and more uncomfortable: an established supplier that AI knew by name and almost never recommended, on the questions its buyers may be starting to ask. That gap can be measured, and you can only close what you can see.
*Written from a design-partner engagement. The company, its competitors and the cited sources are withheld. Findings come only from complete, comparable runs, re-checked against the raw answers.*