The Conversation You Can't See: How B2B Buyers Choose Vendors Inside ChatGPT

Nov 14, 2026

The short answer: To win a buyer in AI search, your company needs to be the source AI cites at every stage: when the buyer is mulling the problem, when they are gathering details on solutions, when they compare options, and when they ask for the best company to use. You won't see this buyer conversation in your analytics, but you can match your content, third-party citations, and thought leadership to steps in the buyer's journey. This guide shows what it looks like, where companies drop out, and how to stay in it.

The invisible buyer journey

The journey that it takes an AI search engine to recommend your brand starts many levels higher in ChatGPT, Claude, Gemini, Perplexity, or Copilot. A buyer typically starts the conversation asking about a business problem and how to solve it. The AI search can use query fan-out to provide many levels of detail so that the buyer has a full understanding of the problem and the ways to solve it. Research shows the buyer journey often starts in AI. In a March 2026 G2 survey of 1,076 software buyers, 51% said they now start their research with an AI chatbot.

Many B2B companies approach visibility in AI search and the buyer journey in the same way they would SEO, with the company matching the AI criteria and showing up in the recommended answer. The reality is your company should have the AI presence that will associate your brand with the early-stage queries as well as the best vendors to solve the problem.

It helps to look at how a B2B buyer researches solutions in AI. The person rarely opens ChatGPT or another AI search engine and types a company's name. Typically, they start with a problem, ask follow-up questions for days or weeks, and only ask for vendor names once they understand the category. They are also working in an AI search engine that has a layer of customization that goes far beyond Google. The AI search engine can remember the buyer's location, preferences for vendors, other queries for similar problems, and this detail also feeds into the answers that AI gives.

A typical B2B buyer conversation moves through four stages:

  1. Confused. "Our auditors flagged something we don't understand. What does it mean?"

  2. Learning. "What are the options for fixing this? What do companies like us usually do?"

  3. Comparing. "What's the difference between approach A and approach B? What does each cost?"

  4. Choosing. "Which vendors are best for a company our size?"

Research backs up this query model. In a 2026 Semrush survey of 622 U.S. B2B professionals, 66% said they regularly use AI to research vendors. Of those, 59% use it to understand a problem or category, 61% to compare vendors, and 53% to ask for recommendations.

What makes this process frustrating is that companies get insight into none of it. These conversations are private, with no keyword report, no click, no visit to your site from the early searching.

How AI researches B2B questions: query fan-out

When a buyer asks AI a question, the engine doesn't run one search. It typically breaks the question into many smaller questions, searches for each one at the same time, and combines what it finds into one answer. This is called query fan-out, and it is what makes AI search so powerful. Rather than getting a single list of answers, you get contextual content. With that said, simple questions, like a definition or a date, usually need little or no fan-out. Complex buying questions trigger the most.

Google describes query fan-out this way for AI Mode: it breaks "your question into subtopics" and issues "a multitude of queries simultaneously on your behalf." Google uses it in AI Mode, Deep Search, and some AI Overviews. Other AI search tools work in a similar way, though most don't show their searches.

Here is what one buyer question might fan out into:

The buyer asks: "Do we need CMMC Level 2, and how hard is it to implement?"

AI might also search for:

  • What CMMC Level 2 requires

  • Who needs Level 2 versus Level 1

  • How long certification takes for a small company

  • What certification costs

  • Common reasons companies fail an assessment

  • Providers that help small manufacturers

Why this matters for you: Each of those smaller searches has its own winning authoritative sources that it cites. A broad page about "cybersecurity compliance" will match all of these searches loosely but may appear in none. A page on "how long CMMC Level 2 takes for a small manufacturer" or "should a mid-sized defense contractor use a CMMC compliance software platform or hire a managed service provider (MSP)?" will have the specificity to get cited for a matching query. Fan-out also means AI answers questions the buyer never typed, so you can be cited for a question that was not on the buyer's list. To note, having a specific page on its own does not guarantee that AI will cite you. A company has to build that authority across a number of signals.

Where companies drop out of AI search

If you look at most B2B website content, it is geared to answer only the last question: "Who should I hire?" Product pages, feature lists, and "Why choose us" pages can all be strong, but they won't match the earlier questions.

That leaves the first three stages to someone else. When the buyer is confused or learning, AI cites sources that explain the problem best: an analyst, a trade publication, a Reddit thread, or a competitive guide. By the time the buyer asks for names, AI already trusts those sources, and it tends to recommend the companies providing this type of content or mentioned by one of the third-party sources it used in an earlier query stage.

This matters because the shortlist forms early. G2 found that about one-third of software buyers purchased from a vendor they had never heard of before their research. That cuts both ways. A smaller company can win a deal it would never have been in, and a well-known company can lose one without ever knowing it was in play.

The fix is not more brand-centered content. Instead, the content should focus on your buyer's problem, written well enough that AI cites you from the first question.

What the conversation looks like by industry

The four stages are the same everywhere, but the questions are not. Below is an example conversation for each industry we work with. Each one starts far from a vendor name, which is exactly where most companies have no content.

Cybersecurity

The buyer: IT director at a 300-person manufacturer that sells to the U.S. Department of Defense.

The buyer's AI chat: IT director, defense manufacturer
1. Confused

A customer says we need CMMC Level 2. What is that?

What you need on your site: A plain-English explainer of CMMC levels for small manufacturers.

2. Learning

How long does CMMC Level 2 take for a company with no security team?

What you need on your site: A realistic timeline, with what you've seen at real clients.

3. Comparing

Should we use a managed security provider or hire in-house?

What you need on your site: An honest comparison, including when in-house is the better choice.

Three questions in, the buyer still hasn't asked about a vendor.

4. Choosing

Which providers help small manufacturers get CMMC certified?

What you need on your site: Proof: case studies, certifications, and named client outcomes.

Where cybersecurity companies drop out: Their sites talk about threats and platforms. The buyer is asking about compliance deadlines and budgets.

SaaS

The buyer: VP of Finance at a growing software company whose billing has outgrown spreadsheets.

The buyer's AI chat: VP of Finance, growing software company
1. Confused

Why doesn't our revenue in the books match our bookings?

What you need on your site: An explainer on revenue recognition for subscription businesses.

2. Learning

When should a SaaS company move off spreadsheets for billing?

What you need on your site: The signs it's time, drawn from what your customers told you.

3. Comparing

Should we use our ERP's billing module or a separate tool?

What you need on your site: A fair comparison, including when the ERP module is enough.

Three questions in, the buyer still hasn't asked about a vendor.

4. Choosing

What billing software works for a company with 500 customers?

What you need on your site: Customer stories with company size, setup time, and results.

Where SaaS companies drop out: Their content lists features. The buyer is trying to explain a problem to their CFO.

Watch your pricing: At the comparing stage, buyers ask AI what each option costs. If AI shows old or wrong prices, you can lose the deal before anyone visits your site.

Fintech

The buyer: Chief Risk Officer at a regional credit union seeing more payment fraud.

The buyer's AI chat: Chief Risk Officer, regional credit union
1. Confused

Why are we seeing more authorized push payment fraud?

What you need on your site: A clear explainer of the fraud pattern and why it's growing.

2. Learning

What do regulators expect credit unions to do about it?

What you need on your site: A plain summary of the guidance, with what it means day to day.

3. Comparing

Is real-time fraud scoring worth it for an institution our size?

What you need on your site: Honest cost and benefit by institution size, with real numbers you can share.

Three questions in, the buyer still hasn't asked about a vendor.

4. Choosing

Which fraud detection vendors work with credit unions?

What you need on your site: Named credit union clients, integrations with core banking systems, and results.

Where fintech companies drop out: Their sites are written for big banks. The buyer wants proof that the vendor understands an institution their size and their regulators.

AI/ML

The buyer: Head of Engineering at a healthcare software company putting an AI assistant into its product.

The buyer's AI chat: Head of Engineering, healthcare software company
1. Confused

Our AI assistant sometimes makes things up. How do we catch that before users see it?

What you need on your site: An explainer on why models hallucinate and how teams detect it.

2. Learning

What do we need to log and test to use AI safely with patient data?

What you need on your site: A practical checklist for healthcare teams, based on real deployments.

3. Comparing

Should we build our own testing tools or buy a platform?

What you need on your site: A build-versus-buy breakdown with time and cost from real projects.

Three questions in, the buyer still hasn't asked about a vendor.

4. Choosing

Which AI monitoring tools are used by healthcare companies?

What you need on your site: Healthcare client stories, compliance details, and security documentation.

Where AI/ML companies drop out: Their content is written for other engineers who already know the category. The buyer often has to explain the risk to legal and compliance first.

How to be in the whole conversation

To be cited at every stage, you need an in-depth answer for each question your buyer asks, on your own website, written so both AI and people want to use it.

  1. Collect the real questions. Pull them from sales calls, support tickets, onboarding notes, and the emails prospects send. Write down the buyer’s exact words, not your internal terms.

  2. Sort them by stage. Put each question under Confused, Learning, Comparing, or Choosing. Expect the first two columns to be the thinnest.

  3. Go deep, not broad. One page that fully answers “How long does CMMC Level 2 take for a small manufacturer?” beats a general guide to cybersecurity. AI breaks a buyer’s question into many smaller ones, and specific pages match those closely.

  4. Answer first. Put the direct answer in the first two or three sentences. Then explain, give examples, and add detail.

  5. Add what only you know. Client results, your own data, real timelines, mistakes you’ve seen. AI skips content any company could have written.

  6. Write it for a person, too. Clear structure helps AI pull the right passage, but the page still has to be something a buyer would read and share with their boss. Content that is useful to people is what AI is looking for. There is no separate version “for the machine.”

  7. Make each section stand on its own. AI often lifts one section, not the whole page. Each one should make sense without the rest.

  8. Keep facts current. Prices, product names, and certifications change. Outdated details at the comparing stage cost deals.

How to check where you stand

You can test this yourself in about an hour, with no special tools.

  1. Write the buyer’s conversation. Use the example for your industry above, or write 8 to 10 questions of your own across the four stages.

  2. Run it as one conversation. Open ChatGPT, Perplexity, Gemini, and Claude. Ask the questions in order, in a single chat, the way a buyer would.

  3. Note every source cited. At each stage, write down which companies, publications, and sites AI mentions or links to.

  4. Find your gaps. Mark the stages where you appear and where you don’t. Note which competitors show up early.

  5. Repeat monthly. AI answers change. Run the same conversation each month to see if your new content is being picked up.

If you show up only at the Choosing stage, or not at all, that gap is where to start writing.

FAQs

Why can’t I see these conversations in my analytics? AI conversations happen inside ChatGPT, Perplexity, and other tools. Unless the buyer clicks a link to your site, nothing reaches your analytics. Many never click until they have a shortlist.

Do I still need SEO? Yes. Many AI tools pull from search indexes, so a site that can be crawled and indexed is still the foundation. GEO builds on SEO; it doesn’t replace it.

Should I write separate content for AI and for people? No. Structure your content so AI can pull clear answers from it, but write it for a real buyer. Content people find useful is what AI wants to cite.

How long until we see results? It depends on your industry and how much useful content you already have. Run the self-check above monthly to track progress.

Do we need to be a big company to show up? No. AI cites the most useful answer, not the biggest brand. A smaller company with specific, in-depth answers can be cited ahead of larger competitors.

Sources