We asked an AI assistant to recommend suppliers, the way a buyer might, for 1,917 US industrial B2B companies. Each question used that company's own category and state. The assistant named the company in 41% of its answers.
The conditions matter as much as the number, so they come first. It was one assistant, Claude Haiku, allowed up to two web searches per question, asked between July 30 and August 1, 2026. Each company was asked about once, and answers vary between askings, so 41% describes the group, not the odds for any single company. The companies came from a list of established US industrial businesses rather than a random sample of all of them. And the method shaped some of what came back, which the next section explains.
What we asked, and how that shaped the answers
Each question matched what the company does. A manufacturer's read: "I'm looking for a company that manufactures warehouse conveyor systems in Wisconsin. Who are the best companies to consider?" A distributor's asked for "a supplier or distributor of factory automation components in Georgia." Services businesses and manufacturers' rep agencies each had their own wording too.
Two instructions were added to every question. The assistant was told to name specific companies. It was also told to label each company it named as either national or global, meaning large brands, public companies and household names, or local and regional, meaning independent firms a buyer in that state would realistically call. Every question also named a state, which tends to pull answers toward companies in that state.
We asked 2,000 questions. For 83 the answer didn't include a real list of companies, so those are left out, leaving 1,917.
Bigger names were usually in the answer
Seven in ten answers, 69.8%, included at least one company other than the one we asked about that the assistant labelled national or global. Among answers that left the company out, the figure was 78.0%.
When another national or global name was present, the company we asked about was named less often: 34.1% of the time, against 57.0% when none appeared. The study doesn't show why, and several other things moved with it. National names turned up more often in answers for distributors (88.3%) than for manufacturers (65.0%). In 241 answers the assistant labelled the company we asked about as national itself. And answers that listed six or more companies named it less often (28.8%) than shorter answers (50.2%). Within those longer answers, the gap between having a national name present or not nearly disappears, at 28.3% against 30.5%. So treat it as a pattern worth knowing about, not proof that big names push smaller suppliers out.
The answers weren't all big brands. Most individual names were labelled local or regional, which is what you'd expect when every question names a state, and a typical answer mixed the two. Asked about a Wisconsin maker of hydraulic cylinders and pumps, the assistant named Perfection Hydraulics, Hader Industries, Hydraulic Solutions of Wisconsin, Morse Hydraulics and Simplex, alongside The Oilgear Company, which it labelled national. It didn't name the company we'd asked about.
One caveat on all of this: "national" and "local" are the assistant's own labels, applied using a definition we gave it, not a measured company size.
Does anyone actually buy this way?
Fair question, and the honest answer is that we have two data points rather than a pattern.
One industrial client closed a deal we could trace back to ChatGPT. Another, a company that produces physical marketing materials for CPG brands, keeps seeing inbound traffic with ChatGPT as the referrer, though nothing there has been traced to a closed deal yet.
Two companies is an anecdote, not a trend. What it does settle is a narrower question: whether a buyer can find and contact a company this way at all. They can, because it has happened.
Why a company might not get named
This study didn't test why one company was named and another wasn't, so this isn't a finding from it. It draws on separate research and on what we see when we look at sites.
When an assistant searches the web to answer a question like this, it often relies on what companies' own websites say, alongside directories and other pages. Our separate study of more than 55,000 B2B websites found that 82% aren't structured for AI search to read them: only 9% use structured data, and about half keep their specifications inside PDFs rather than on the page. Copy like "quality solutions for a variety of industries" doesn't name a product, a material or a state, so it gives an assistant little to match against a specific question. The same vagueness keeps companies out of ordinary search results, as covered in why isn't my website showing up on Google.
That's a plausible explanation for some of the 59%, and it wasn't measured here. Making a site readable to these tools is usually called answer engine optimization, and there's a practical walkthrough in how to get cited by ChatGPT.
Why does ChatGPT recommend our competitors but not us?
In our September 2026 study of 20,548 US manufacturers, the companies that ChatGPT, Gemini, Google AI Overviews or Google AI Mode recommended differed from the ones that didn't in several ways. Three of them are worth checking here: two you can compare against a competitor, Google and what the homepage says, and one you can't change, how crowded your town is. Being findable usually wasn't the gap, since ChatGPT, asked by name about companies no assistant had recommended, found and cited their own site 83% of the time.
Each of the three below comes from one kind of search, a buyer asking for a service near a specific town, and each is an association, not proof of cause.
- Are they on Google's first page for that question, and you aren't? 74% of the companies in the manufacturer study weren't on Google's first page for the buyer's search, and 67.8% of those were left out by all four assistants. For companies whose own site ranked first, it was 1.0%. ChatGPT, which doesn't use Google's results, showed the same pattern.
- Does their homepage say what they do and where, when yours doesn't? Where the page title named the service, the company was recommended 66% of the time, against 42% where it didn't. Only 28% of the sites did this. A city named on the page went with 54% against 35%. A title like "CNC Machining in Dayton, Ohio" hands an assistant the words in the buyer's question. "Precision Solutions Since 1978" doesn't.
- The one you can't change: how crowded your town is. Gemini and Google's AI usually list one to three companies, ChatGPT about a dozen. Where a town had one or two companies in the database the study drew from, 34.7% went unrecommended by all four. Where it had 50 or more, 79.2% did. Being good at the work doesn't get you onto a three-name list by itself.
If the competitor is a bigger national name, that may not be the reason: in the test this post is built on, which asked Claude Haiku rather than ChatGPT, companies were named less often when one appeared, but within longer answers that gap nearly disappears.
In the manufacturer study, three things often sold as fixes didn't track with being recommended, or only barely. A blog section made no difference, and neither did a Thomasnet profile in a smaller check. Structured data, mentioned above, showed only a small one: LocalBusiness schema went with 54% against 47%, so plain words on the page went with a much bigger gap than the markup did.
What to check first. Search the question your buyer asked on Google and look for the competitor the assistant named. If their site is on the first page and yours isn't, that's likely an SEO problem before it's an AI one, and it's the place to start. The five-minute test further down covers the AI side.
The reason this is hard to notice
You can't see it happen. There's no record of a buyer who asked an assistant, got an answer that didn't include you, and contacted someone else. It doesn't reach your CRM or your phone. From the inside it looks like a week when fewer people were looking.
A test you can run in about five minutes
You don't need a tool for a first read. You need the question a buyer would ask, an assistant that searches the web, and somewhere to write down what comes back. The first pass, with one assistant, takes about five minutes. The extra prompts and the log below build a fuller picture over time.
- Write the buyer's question, not your company's name. Use the specific thing you make and the state or town you sell from, the way someone who has never heard of you would ask it. If your name is in the question, the assistant will talk about you whatever it would otherwise have said.
- Ask it in a private window while logged out. Use an assistant that searches the web, such as ChatGPT, Gemini or Perplexity, and search the same question on Google, where it may bring up an AI Overview (not every search shows one), or ask it in Google's AI Mode. Signed in, an assistant can carry saved memory, custom instructions and past chats into a new conversation, and if you've ever asked one about your own company, that history can tilt the answer toward you. A temporary chat is a weaker fallback: ChatGPT's can still apply your memory and custom instructions unless you start it unpersonalized.
- Notice whether you're in the answer, who is, and whether any of them are much larger than you.
- Then ask twice more, once with different wording and once naming an industry you serve instead of a product. Answers vary, so one answer tells you little.
- Check your analytics for referrals from chatgpt.com, perplexity.ai and gemini.google.com. Some visits sent by an assistant show up there, though many arrive with no referrer at all, so an empty report doesn't mean nothing is happening. Clicks from Google's AI Overviews and AI Mode arrive as ordinary Google search traffic, so they won't show up as a separate line.
Prompts to copy
Replace each bracket with something specific: a product rather than a category, and a real town rather than a region. In the second prompt, [your service] means the work you do for customers, such as CNC machining or powder coating.
- I'm looking for a company that makes [the specific thing you make] in [your state]. Who are the best companies to consider?
- Recommend a company for [your service] near [your town], [your state].
- Which [kind of company you are, such as machine shops] in [your state] work with [an industry you serve] companies?
The first follows the wording behind this post's test. The second is the one our study of 20,548 US manufacturers asked, and it's closer to how a buyer looks for local work. The third is the industry version from step 4.
A simple results log
Write one line per answer. A spreadsheet works, and so does a page in a notebook. These seven columns are enough:
- Date
- Assistant
- Exact prompt
- Were we named?
- Who was named instead
- Any much larger than us?
- Sources it cited
The sources column is often the most useful. It shows which directories, lists and competitor pages the assistant leaned on, and an answer that cites nothing may not have searched the web at all. Ask the same prompts again at least once a month, and add every answer as a new line. In the 20,548-manufacturer study, when 1,202 of its questions were asked again a day later, 85% to 93% of companies got the same named-or-not result, so a single flip between two runs can be noise. A few months of lines show whether you're named more often or less, and whether some assistants name you while others don't.
If you'd like a second opinion, our AI Visibility Check puts buyer questions like these to ChatGPT, Gemini, Google AI Overviews and Google AI Mode, starting with your own town, and reviews what your homepage gives them to work with. It's complimentary, and the report is yours to keep either way.
