Your buyers have started asking AI assistants which company to use before they ever type a search into Google. The assistant answers with a shortlist of names. If your company is not on that shortlist, you are out of the deal before you knew it existed.
Most industrial and B2B companies have no idea where they stand in those answers. This guide shows you how to find out in about ten minutes, what your results mean, and what actually drives the answer.
Why this matters now
A client asked me a question last year that almost nobody was asking the year before: "When someone asks ChatGPT to recommend a company like ours, do we show up?"
It is the right question, and the reason has less to do with AI than you would expect. 6sense's 2025 B2B Buyer Experience Report found that "94% of buying groups ranked preferred vendors before first contact," and that "they ultimately purchased from that preliminary favorite 77% of the time." The shortlist is built before anyone talks to you, and the name at the top of it usually wins.
I want to be careful with the AI part of this, because the same report cuts against the easy version of the story. Yes, 94% of buyers used large language models during the process. But 6sense found they used them mainly "to summarize reviews or analyze data," and that buyers still averaged 16 interactions with the winning vendor, unchanged from the year before. Their own write-up of that number is titled "Why ChatGPT Isn't Killing Your B2B Demand Gen, Yet." So this is not a claim that AI has replaced how buying happens.
The narrower claim is the one that matters. A shortlist forms early, it decides most deals, and one of the places buyers now assemble it is a chat window that returns three or four company names. If you are not one of them, you were never in the running, and nothing in your analytics will record that you lost.
Traditional marketing measurement does not see any of this. Companies track keyword rankings and ad spend. Nobody is checking the answer the model actually gives. Which means a competitor can quietly own the AI shortlist in your category for months before it shows up as lost revenue.
The 10-minute AI visibility check
You can run this yourself today. No tools, no budget.
Step 1. Write down the five questions a buyer asks right before they would contact you.
Not your product name. Not your brand. The problem, phrased the way a real buyer phrases it: "who does X for Y." If you build industrial conveyor systems, the question is something like "what company makes heavy-duty conveyors for food processing plants." If you sell industrial water treatment, it is "water treatment systems for manufacturing facilities." Ask your sales team what prospects actually say on first calls. Use those words.
Step 2. Ask those exact questions in ChatGPT, Perplexity, and Google's AI Overviews.
Fresh session each time. No leading. Do not mention your company name first, because the model will happily talk about any company you name. You want the unprompted answer, the one your buyer sees.
Step 3. For each answer, record three things.
- Are you named at all?
- Who gets named instead?
- What does the model say about them that it does not say about you?
That third question is the whole game, and it is the one most people skip.
Reading your results
You are named, described accurately. You are in the consideration set. Now the work is widening the gap: more third-party citations, clearer proof, so a competitor cannot displace you.
You are named, described wrong or thinly. The model knows you exist but cannot make your case. This usually means your expertise lives in places a model cannot lift: PDFs, image-heavy pages, vague homepage copy.
You are not named, a competitor is. Study what the model says about them. It is usually specific: what they do, who they serve, why they are credible. Then trace where that language comes from. In a recent audit I ran, the winning competitor was not the biggest company in the category. The biggest brand won on sheer scale, a different mechanism. The consistent winner was the firm whose expertise was stated clearly across the sources the model reads.
One finding from that audit is worth sitting with. In several answers I could see the engine's own research trail: it searched the client by name, checked their reputation and history, weighed them against the competitor, and still left them out of the answer. The client was not unknown. The AI investigated them and could not find enough usable proof to recommend them. If you think your problem is awareness, it may actually be legibility.
Nobody in your category is named. The model answers generically. This is the open-field scenario, and it is an opportunity: the first company in the category to become clearly citable tends to own the answer.
What actually drives the answer
The model is not ranking pages the way Google ranks pages. It is reasoning over what the web says about you, and three things dominate:
1. Who cites you. Third-party mentions carry more weight than anything on your own site. This is off-page SEO doing double duty, and the practical version of that work is here. Trade publications, industry directories, association pages, review platforms, press coverage. The model treats independent sources as evidence and your homepage as a claim.
2. Whether your expertise is stated in a form a machine can lift. Clear sentences that say what you do, for whom, with what proof. A page titled "Conveyor Systems for Food and Beverage Manufacturing" that plainly describes the service can be quoted. A homepage that says "engineering tomorrow's solutions today" cannot.
3. Whether the model can tell what you do at all. Entity clarity. Consistent naming, a specific category, unambiguous descriptions across your site, your LinkedIn, your directory listings. Companies that describe themselves five different ways in five places dilute their own identity in the model's reading.
Most industrial company websites fail on all three, and not by accident. They were built for a human skimming a homepage, not a model summarizing an answer. The design brief was impressions, not liftability.
One thing worth saying plainly, because a lot of people are being sold the opposite. There is no technical file or markup that gets you into an AI answer. Google's own guidance on AI features is explicit: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary," and "you don't need to create new machine readable files, AI text files, or markup to appear in these features." It goes further and says "there's also no special schema.org structured data that you need to add."
So if someone quotes you for an llms.txt file or a schema package to fix your AI visibility, they are selling you a deliverable, not a result. The three drivers above are slower and less tidy, and they are what actually moves the answer.
An honest word on limits
Two disclosures, because overclaiming is how agencies get fired.
First, AI answers are not deterministic. The same question can produce different shortlists on different days, and the models update constantly. A single check is a snapshot, not a scoreboard. Run your five questions monthly and watch the trend, not any single answer.
Second, nobody outside these AI companies knows the exact weighting of any of this. What I describe above is what my audit work has shown so far: companies with strong third-party citations and liftable expertise get named, companies without them do not. Treat it as a well-supported operating model, not a published algorithm.
What to do with a bad result
The fix is real work, not a plugin, and it falls out of the three drivers: earn citations in the sources the model reads, restate your expertise in liftable form, and clean up your entity signals. Which lever matters most depends on what your check revealed, which is why the diagnosis comes first.
If you run your five questions and do not like the answer, that is the situation my AI Visibility Audit exists for. I benchmark whichever competitor keeps winning the answer, work out which of the three levers is actually doing it, and hand back a written plan in priority order. But run the ten-minute version first. It costs nothing, and it tells you whether you have a problem worth solving.
I run paid media for companies whose ad spend has to prove itself in revenue. Increasingly that means making sure the AI layer sitting above the ads knows my clients exist. If you want a second pair of eyes on what the models say about you, book an intro call.