About half of 20,548 US manufacturers weren't recommended by ChatGPT, Gemini, or Google's AI when a buyer asked for their service near their town.
When a buyer asked ChatGPT, Gemini, Google AI Overviews, or Google AI Mode to recommend a company for a manufacturer's service near its own town, about half of US manufacturers (51.7%) weren't named by any of the four (Alkali, analysis of 20,548 US manufacturers and 47,000 AI answers, collected September 26 to 28, 2026).
Recommended means the answer named the company or linked its website, and a company counts as recommended if any of the four assistants did. Most of the companies left out aren't unknown to AI: asked about them by name, ChatGPT, Google AI Mode, and Google AI Overviews cited their websites 75% to 86% of the time. Asked to recommend someone, they named other companies instead.
| Metric | Value | What it means |
|---|---|---|
| Not recommended by any of the four assistants | 51.7% | Roughly 44% to 52% across the question wordings we tested |
| Named by ChatGPT | 41.9% | ChatGPT names about 12 companies per answer, so more get in |
| Named by Google AI Overviews | 25.5% | Of searches where an AI Overview appeared (91% of them) |
| Named by Gemini | 23.6% | Gemini usually names one or two companies and then asks a follow-up question |
| Named by Google AI Mode | 22.2% | Two or three companies per answer |
| Named by all four | 9.7% | The assistants rarely agree on who to recommend |
| Not recommended by any assistant, when not on Google's first page | 67.8% | vs 1.0% when the company's own site ranked first in Google's regular results |
| Google's own map-pack #1 left out by Google's AI | about 60% | AI Mode 60.6%, Gemini 62.4%, AI Overviews 59.6% where one appeared |
| Named for a certified-supplier search, certification stated on the site | 41.2% | vs 14.5% when the site doesn't mention it |
| Not recommended by any assistant, towns with 50+ companies in the source database | 79.2% | vs 34.7% in towns with one or two |
| Not recommended by any assistant, metal stamping companies | 26.9% | vs 68.6% of custom packaging companies |
| Google showed an AI Overview, question-style search | 91% | vs 30% for keyword searches and 17% for "near me" |
B2B buyers, including the ones who buy from manufacturers, are already using these tools. In a Gartner survey of 646 B2B buyers, 45% said they used AI during a recent purchase [1]. In a Semrush survey of US B2B professionals, 92% of those who use AI at work said it had shaped which vendors they considered [2]. And Forrester found buyers ranked generative AI and conversational search as a more meaningful source of information than vendor websites, product experts, or salespeople [3].
The open question for a manufacturer is whether the company shows up when they do.
We asked ChatGPT, Gemini, Google AI Overviews, and Google AI Mode the question a buyer would ask: "Recommend a company for [service] near [town], [state]." Across 20,548 US manufacturers in 17 industries, 51.7% weren't named by any of the four.
The result held when we checked it. Asking 1,202 of the same questions again a day later gave nearly identical rates, and the same named-or-not result for 85% to 93% of individual companies. Rewording the question to "What are the best [service] companies in [town]?" moved the figure a few points, from 48.0% to 43.6% for the 1,621 companies in that check. Across the wordings and samples we tested, roughly 44% to 52% of manufacturers weren't recommended.
This measures one kind of search: a buyer looking for a specific service near a specific town. A manufacturer that sells mostly to customers across the country may be found in other ways, and this study doesn't measure those.
Invisible doesn't mean unknown. We took 1,000 companies that none of the four assistants recommended and asked each assistant about them by name. ChatGPT found and cited the company's own website 83% of the time, Google AI Mode 86%, and Google AI Overviews 75%. Gemini, which leans on Google Maps, linked the company's site 18% of the time. Fewer than 2% of answers said the assistant couldn't find the company.
So for most of these companies, the problem wasn't that AI couldn't find them. Asked to recommend someone, it named other companies instead.
Only 9.7% of companies were named by all four assistants, and 16.1% by exactly one. Ask the same assistant the same question twice and the full list it names overlaps by only 27% to 61%. The lists reshuffle, but the companies that were invisible mostly stayed invisible.
The assistants also lean on different sources. A company's own website carries most of the weight with ChatGPT (77.5% of its citations) and is also the largest single source for Google AI Mode (45%). Google AI Overviews lean on directories and listing sites (62.7% of its citations), and Gemini mostly cites Google Maps. Across all four, the most-cited listing sites were MapQuest, Yelp, Facebook, Yellow Pages, and Thomasnet.
Companies that already show up on Google were far more likely to be recommended by AI. Only 1.0% of companies whose own website ranked first in Google's regular results for the buyer's search were missing from every AI answer, and 5.1% of those ranked second to tenth. Among companies that weren't on Google's first page at all, which was 74% of the sample, 67.8% were invisible to AI too. Part of that link is built in for Google's own AI, which draws on Google's results. But ChatGPT, which doesn't, showed the same pattern: it named 92% of the companies ranked first in Google's regular results and 27% of companies that weren't on the first page.
One exception: Google's own AI regularly skips the business Google ranks first in its map results. In 8,295 searches that showed a map pack, Google AI Mode left out the top business 60.6% of the time and Gemini 62.4%. Where an AI Overview appeared, it left it out 59.6% of the time. ChatGPT left it out 27.0% of the time. Ranking in the map isn't the same as being recommended.
How a buyer searches also decides whether they see AI at all. Google showed an AI Overview on 91% of the question-style searches in this study, but on only 30% of short keyword searches like "cnc machining dayton oh" and 17% of "[service] near me" searches. Even the regular results often don't show a supplier first. Google's top organic result was a directory or listing site rather than a supplier's own website 66% of the time for question-style searches and 52% for keyword searches.
Gemini and Google's AI name two or three companies per answer and ChatGPT about a dozen, so the more companies a town has, the more get left out. In towns with one or two companies in the database we drew from, 34.7% weren't recommended by any of the four assistants. In towns with 50 or more, 79.2% weren't.
Company size mattered less than you'd expect. Companies with $1 million to $10 million in revenue and those with $10 million to $100 million, together about 70% of the sample, went unrecommended at almost the same rate: 52.6% and 51.6%. The smallest (under $1 million) and the largest ($100 million to $1 billion) were left out more often, 61.9% and 63.9%, possibly because the largest sell nationally rather than to their own town.
Results varied widely by industry. 26.9% of metal stamping companies weren't recommended by any of the four assistants, 27.9% of cable assembly and wire harness makers, 28.0% of laser cutting companies, 32.3% of electronics manufacturers, and 33.4% of injection molders. At the other end, 58.2% of welding companies weren't recommended, 59.1% of general metal fabricators, 65.5% of heat treaters (55 companies), and 68.6% of custom packaging companies. Industries with fewer than 100 companies in the sample, like heat treating, are less reliable.
Some of that gap is how crowded each market is, but not all of it. Compared with similarly crowded markets, metal stamping companies were left out 17.5 percentage points less often than expected, and custom packaging companies 19.8 points more often.
| Industry | Companies | Not recommended by any assistant | Left out vs. similarly crowded markets |
|---|---|---|---|
| Metal stamping | 264 | 26.9% | 17.5 pts less often |
| Cable assemblies and wire harnesses | 140 | 27.9% | 13.7 pts less often |
| Laser cutting | 182 | 28.0% | 13.7 pts less often |
| Electronics manufacturing | 263 | 32.3% | 13.3 pts less often |
| Plastic injection molding | 928 | 33.4% | 12.0 pts less often |
| Contract manufacturing | 338 | 33.7% | 8.9 pts less often |
| Sheet metal fabrication | 333 | 39.6% | 4.1 pts less often |
| Structural steel fabrication | 1,183 | 40.1% | 6.9 pts less often |
| 3D printing | 202 | 45.5% | 3.8 pts more often |
| CNC machining | 6,642 | 48.9% | 7.0 pts less often |
| Powder coating | 106 | 49.1% | 8.3 pts more often |
| Industrial automation | 2,061 | 56.7% | 4.5 pts more often |
| Welding services | 1,573 | 58.2% | 12.0 pts more often |
| Tool and die | 56 (small sample) | 58.9% | 18.3 pts more often |
| Metal fabrication | 4,978 | 59.1% | 5.2 pts more often |
| Heat treating | 55 (small sample) | 65.5% | 24.7 pts more often |
| Custom packaging | 1,244 | 68.6% | 19.8 pts more often |
Raw state rankings partly reflect how many crowded towns a state has: for example, 66.0% of Texas manufacturers weren't recommended, against 24.7% in Iowa. Compared with similarly crowded markets, a gap remains. Manufacturers in Iowa were left out 16.9 percentage points less often than expected, Wisconsin 11.3 points less, and New Hampshire 10.1 points less. Texas manufacturers were left out 7.0 points more often, New Jersey 6.3 points more, and Colorado 6.1 points more. States with fewer than 100 companies in the sample aren't included, and this data can't say why the remaining differences exist.
We crawled the homepages of 18,697 companies in the study and compared the ones AI recommended with the ones it didn't. The differences weren't technical. They were basic.
Companies whose page title named their service were recommended 66% of the time, against 42% for those that didn't, and only 28% of sites did it. Naming the city on the page went with 54% versus 35%, naming the state 53% versus 33%, a street address 55% versus 40%, a phone number 52% versus 37%, and a certification such as ISO 9001 or AS9100 60% versus 45%. The same pattern held, somewhat weaker, among companies that don't show up on Google's first page, so it isn't only a side effect of ranking.
The things often sold as AI fixes barely registered. LocalBusiness schema markup went with 54% versus 47%, Organization schema with 49% versus 48%, and a blog or news section made no difference at all. Neither did a longer homepage. The companies AI recommends tend to say plainly what they do, where they are, and what they're certified for. These are associations, not proof of cause, but every one of them points the same way.
Two other common pieces of visibility advice didn't track with AI recommendations either. Thomasnet shows up among Google's AI citations, but having a profile made no difference: about 11% of companies had one whether AI recommended them or not (11.1% versus 10.7% in a 3,000-company sample). And among the businesses Google showed in its map results, more Google reviews didn't mean more recommendations. ChatGPT named 62% to 69% of businesses in each review band below 200, and 51% of those with 200 or more, possibly because heavily reviewed businesses tend to sit in more crowded markets. Both come from narrower checks, so they deserve less weight than the findings above.
We asked a second set of questions in 200 CNC machining and metal fabrication markets, including "Which [service] companies near [town] are ISO 9001 or AS9100 certified?" Companies whose websites stated a certification were named 41.2% of the time. Companies whose websites didn't were named 14.5% of the time, roughly a third as often. That group includes companies that aren't certified at all, so it isn't a clean test of the website alone.
But a company that holds a certification and doesn't say so on its site is giving AI less to go on in exactly the searches where that certification would set it apart.
Start by asking the question a buyer would. Type "Recommend a company for [your service] near [your town]" into ChatGPT and Google, and see whether your company comes up and who comes up instead. Use a private window and ask more than once, because the list reshuffles between runs.
Then check the basics on your homepage. Does the page title name the service you want to be found for? Does the page name your city and state, and show your address and phone number? If you hold ISO, AS9100, ITAR, or other certifications, are they stated in plain words on the page, not only in a logo? None of this is complicated, and in this data the companies AI recommended had these basics far more often than the ones it passed over. If you'd rather have it done for you, our complimentary AI Visibility Check runs the same four-assistant test on your services in your town and comes back with where you stand, what's working, what's holding you back, and the strategy we'd follow.
Who ran this: the study was designed and run by Alkali, led by founder Nick Baudoin. Alkali builds websites for B2B companies and offers AI visibility work, including the complimentary check mentioned above, so we have an interest in the topic. That's why the data sources and method are published here in full, so the numbers can be checked.
Companies: 20,548 US manufacturers and industrial service companies across 17 industries (CNC machining, metal fabrication, industrial automation, welding, packaging, structural steel, injection molding, and others) in all 50 states and Washington, DC, sourced from DiscoLike's database of US B2B companies. Companies with more than 1,000 employees or more than $1 billion in revenue, and companies without a usable location, were excluded. Companies offering the same service in the same town were grouped into one search, giving 12,023 searches.
Questions and assistants: each search asked "Recommend a company for [service] near [town], [state]." We collected answers from ChatGPT and Gemini as shown in their public interfaces, and Google AI Overviews and Google AI Mode as returned by Google for the town, between September 26 and 28, 2026, using DataForSEO. That produced 46,988 answers. ChatGPT was asked with web search turned on, and Google's answers were requested for the company's town. A company counted as recommended when an answer cited its website or named it, allowing for spelling variants. Samples were checked by hand: all 12 name-only matches and all 20 spelling-variant matches checked were correct, and company names pulled from answers were about 97% accurate in a 75-answer check. We also checked 40 companies counted as not recommended that looked like possible misses against the full answer text: the misses were spelling variants, about 1% to 2% of all not-recommended results, which the final matching catches, but a few misses may remain, so the not-recommended figures may run slightly high. In about 9% of searches no AI Overview appeared, so only three assistants could name the company; 56.6% of those companies weren't recommended, against 51.2% where an Overview appeared.
Checks: 1,202 searches were asked again the next day; 1,000 were reworded as "What are the best [service] companies in [town]?" and also run on Google as "[service] near me"; 2,000 were also run on Google as short keyword searches; 1,000 companies no assistant recommended were asked about by name; 600 certification, reshoring, and rush-job questions were asked in 200 markets; 3,000 companies were checked for a Thomasnet profile; and Google review counts were taken from its map results. Industry comparisons are measured against the rate expected given how many sample companies competed for the same search; state comparisons also account for how many companies in the source database are in each town. Both are reported in percentage points. Website signals come from our own crawl of 18,697 company homepages; 1,851 more blocked automated visits. All figures are computed from this data and reported in aggregate. No individual company is named.
Limits: this is a snapshot of answers collected September 26 to 28, 2026, and AI answers change over time. It covers US companies only, drawn from a single commercial database (DiscoLike) that may not represent every US manufacturer, and one main kind of search per company: a buyer looking for a service near a specific town, answered without any personal search history. It doesn't measure how national buyers find suppliers. The website, industry, and state findings are associations, not proof of cause; the industry and state comparisons account for how crowded each market is, but not for other differences between companies.
Alkali, “How often AI assistants recommend US manufacturers,” analysis of 20,548 US manufacturers and 46,988 AI answers, 2026. https://www.byalkali.com/research/manufacturer-ai-visibility/
You're welcome to cite or quote these figures with attribution (CC BY 4.0).
Sometimes. When we asked four AI assistants to recommend a company for a manufacturer's service in its own town, 51.7% of 20,548 US manufacturers weren't named by any of them. ChatGPT named the company 41.9% of the time, Google AI Overviews 25.5%, Gemini 23.6%, and Google AI Mode 22.2%.
Usually not because it doesn't know you exist. Asked about unrecommended companies by name, ChatGPT found and cited their websites 83% of the time. The companies it recommends tend to state plainly on their homepage what they do, where they are, and what they're certified for.
In our data, strongly. Only 1.0% of companies whose own website ranked first in Google's regular results for the buyer's search were missing from every AI answer, compared with 67.8% of companies that weren't on Google's first page.
Not much, in this data. Companies with LocalBusiness schema were recommended 54% of the time versus 47% without it, and Organization schema made almost no difference. Naming the service in the page title (66% versus 42%) and the city on the page (54% versus 35%) mattered more.
If you hold them, yes. When buyers asked AI for ISO 9001 or AS9100 certified suppliers, companies whose websites stated a certification were named 41.2% of the time, compared with 14.5% for companies whose websites didn't.
It varies a lot by industry. Only 26.9% of metal stamping companies, 27.9% of cable assembly and wire harness makers, and 28.0% of laser cutting companies weren't recommended by any of the four assistants, compared with 58.2% of welding companies, 59.1% of general metal fabricators, and 68.6% of custom packaging companies. Part of that gap comes from how crowded each market is, but metal stamping still did better than expected for how crowded its markets are, and custom packaging worse.
From our own measurement, using a company sample sourced from DiscoLike: 12,023 buyer-style searches covering 20,548 US manufacturers, answered by ChatGPT, Gemini, Google AI Overviews, and Google AI Mode between September 26 and 28, 2026, plus a crawl of 18,697 company homepages. The method is described above.
We'll ask ChatGPT, Gemini, Google AI Overviews, and Google AI Mode about your services in your town, then send you where you stand, what's working, what's holding you back, and the strategy we'd follow. Complimentary, and no call required.
Check My Company ↗