In B2B, the person asking ChatGPT is rarely the person who later signs the contract - and that changes everything. Usually the asking is done by someone told to “look into it”: a specialist, an executive assistant, a department head who picked up the topic on top of their own job. That person has no settled view of the market, knows none of the company names and has no intention of reading ten proposals. They want to walk back into a meeting with three names. That is why AI visibility for B2B companies is a different job from visibility for an online store - what is at stake is whether you make that first, short list at all.
This is not an introduction to the subject. How AI-generated answers work and where they get their knowledge I covered separately in the post on SEO for AI, ChatGPT and AI Overviews. Here I assume you know that, and focus on what looks different in a B2B company: a longer buying process, several people on the other side, a narrow niche and projects you are often not allowed to write about.
Who is actually asking, and what
A consumer asks AI about a product. A B2B buyer asks about a category, about risk and about whether the thing can be done at all. Those are entirely different queries and entirely different answers.
In practice I see four question types where a B2B company can get named:
- Market scan - “who builds B2B ordering systems for wholesalers”. The model usually names three to six firms. That is the list you want to be on.
- Feasibility check - “can WooCommerce be connected to our ERP, and how long does it take”. The model answers on the substance and cites pages that addressed exactly that.
- Candidate check - “what do you know about company X”. Asked after the first call, to verify who they are talking to.
- Comparison - “X or Y, how do they differ”. The most merciless one, because the model lines you up against a competitor using whatever both of you wrote publicly about yourselves.
Notice that only one of those questions contains your company name. The other three are won by whoever has content answering a narrow industry question head-on, not by whoever has the prettiest homepage.
AI visibility for B2B companies starts with a list of questions, not content
The most common mistake I see: a company starts “writing for AI” without knowing which questions it wants to answer. The result is generic articles no model has any reason to cite, because a hundred identical ones already exist.
The order should be the other way round. Sit down with the person who answers the phone and reads incoming enquiries, and write out the questions customers genuinely ask before they have ever heard of you. Not keywords, full sentences: “will you do a run of 200 units”, “how many weeks from order”, “do you integrate with our ERP”, “what if we have an old system and do not want to replace it”. In most companies that list runs to twenty or thirty items and takes an hour to produce, because everyone knows them by heart.
That list is both your content plan and your test set. Ask the models the same questions and see who shows up instead of you. How to do that properly so the result is comparable a quarter later is laid out step by step in the AI visibility audit - the rule here is simple: same questions, fresh session, all on the same day.
One service, one page - give the model something to quote
A typical B2B site has a “Services” tab with six offerings described in two sentences each. To a model that is one page about nothing in particular. You cannot pull an answer to “who does X for industry Y” out of it, because neither X nor Y is there, only “comprehensive solutions”.
Splitting that up - one service or one problem per page - is the single strongest change I know of in this area. Each of those pages should answer one question from your list and state plainly: who it is for, what it covers, how the process runs, how long it takes, what you need from the client at the start and what the service does not include. That last point is underrated - the boundary of scope is information models happily quote, because it answers a genuine worry.
Write headings the way a customer phrases a question, not the way an org chart names a department. “Connecting a store to a warehouse system - how it works and how long it takes” can be quoted. “Our services” cannot. It is the same logic I described for a B2B website that generates leads: a site built around a decision rather than a presentation works better on people and on machines alike.
Specifications, timelines and terms - the currency of B2B credibility
Models assemble answers from whatever can be lifted out of the text as a fact. In B2B a fact is a specification, a timeline or a term of business - and that is precisely the part companies hide behind “details agreed individually”.
I understand where that comes from: nobody wants to be pinned to a price before a conversation. But there is a lot of room between a full price list and total silence. You can write that implementation usually takes six to ten weeks. You can write which systems you integrate with, which file formats you accept, what order sizes you handle, what the smallest sensible project budget is and how fast a quote comes back. Every sentence like that is a ready-made piece to cite, and at the same time a filter that screens out enquiries from outside your range.
The same thing works on the human on the other end. Whoever is doing the research usually has one disqualifying criterion in mind and is scanning the text for it. If they cannot find it, they move on - exactly like the model does.
Proof when everything you do is under NDA
This is the most common blocker in B2B: “we cannot write about our projects, clients will not allow it”. Honestly, that is a real constraint, not an excuse. But it does not close the road, it changes the format of the proof.
What you can publish without breaking a confidentiality agreement:
- An anonymous but specific write-up - industry, scale, the problem, what was done, how long it took. No client name, but with real numbers from your side of the work.
- A description of the process - what implementation looks like week by week, what happens at each stage, who on your side owns what. No NDA covers that, and it is one of the most frequently cited things you can publish, because it answers “what will this look like for us”.
- Technical lessons from projects - “in migrations off system X, Y is what usually breaks, and here is how we work around it”. That demonstrates competence more convincingly than a client logo in the footer.
- A sentence about the NDA itself - stating outright that most projects are confidential and that you can walk through one full example in a call. That reads as more credible than a site with no specifics at all.
And one thing worth saying plainly: do not invent numbers you do not have. Models cross-check your claims against other sources, and a B2B buyer will verify them in the meeting anyway. One made-up figure costs more than ten missing ones.
Traces outside your own website
Your own site is one source. Models build a picture of a company from several at once - and in B2B the outside ones often carry more weight, because there are few industry sites and plenty of directories and trade media.
The things that genuinely leave a trace: company profiles in trade directories and chambers of commerce, entries in supplier databases, expert comments in industry media, talks and conferences written up on the organiser’s site, a LinkedIn profile with a real description of your services instead of a slogan, and a Google business profile with current details. What matters is that the company name, address and description read the same everywhere - inconsistency between sources is the simplest way to make a model treat the information as unreliable and leave it out.
The technical layer is the quickest part to tick off: check that AI bots can reach the site at all, and describe your company and services with structured data so a machine knows what is a name, what is a service and what is an answer to a question. I explained that in plain terms in the post on schema.org structured data.
Where to start if you have one week
Measure first, write second. List twenty buying questions from your sales side, put them to the models and note who appears instead of you. Then pick the three questions where your absence hurts most, and write three separate pages full of specifics: process, timelines, terms, boundaries of scope. The rest can wait - three good pages do more here than twenty generic ones.
If you want to see how your company currently shows up in model answers, and which gaps on your site are costing you a place on that short list, send me your website address and two sentences about who you sell to. I will come back with a concrete list of fixes, not talk about potential.
Frequently asked questions
Is AI visibility different in B2B than for an online store? The foundation is shared, but the stakes sit elsewhere. For a store, the model usually answers about a product and a price; in B2B it answers about a category of suppliers and about whether something can be done. So B2B work runs mainly through content describing your process, scope and terms of business, not through product pages.
I cannot publish case studies because of NDAs. Can I still get cited? Yes, just not with client logos as the proof. What works is an anonymous write-up with numbers from your side, a week-by-week description of how implementation runs, and technical lessons from projects. Confidentiality protects the client’s data, not your knowledge of how such projects are run.
How long before a B2B company starts appearing in AI answers? This is quarters of work, not weeks, because the engines update their picture of a market slowly and draw on many sources at once. A sensible rhythm is one measurement per quarter using the same question set - only the second measurement tells you whether the work is doing anything.
Is it worth writing a separate page per service when I only take on a dozen clients a year? That is exactly why it is worth it. At a dozen clients a year, one extra enquiry off a good shortlist can change the year, and a narrow niche works in your favour: competition for those questions is thin, because almost nobody in the industry has written a concrete answer to them.
Want your company to show up in Google and in ChatGPT answers? I work on technical SEO and AI visibility from the code side, not the reporting side. Tell me which questions matter to you and I will quote it.