Somewhere between asking a colleague for a recommendation and typing a query into Google, your next buyer opened a chat window instead. They described their problem in a sentence or two, asked for three vendors worth a look, and got an answer in about four seconds. Your website was not part of that conversation unless the model already knew your name.

That is the uncomfortable part. Your buyer did not skip research. They moved it somewhere your analytics cannot see.

I have watched this shift play out with B2B marketing teams for a while now, and the teams handling it well treat AI answers as a real channel with real inputs, not as a novelty. If that sounds like your team, you can track brand visibility in chatgpt the same way you would watch rankings or share of search, because the mechanics underneath are surprisingly similar.

What the buyer actually does inside the chat window

The pattern rarely starts with a brand name. Buyers arrive with a mess: a tool that does not talk to their CRM, a finance lead who wants the contract consolidated, a founder who saw a competitor’s pricing page and panicked. They type something like “what is a good alternative to the platform we use for pipeline reporting” or “which vendors handle SOC 2 evidence collection for a 40 person startup.”

The model answers with a short list, a sentence or two of reasoning, and often a caveat. That caveat matters more than people think. If the answer says “these three are commonly mentioned, though pricing varies widely,” your buyer now trusts the list but distrusts the details, so they go verify. What they verify first is usually your comparison page, your pricing page, and your G2 profile, in that order, usually within the same browser session. So the shortlist gets built in one place and stress tested in another. Marketing teams that only watch the second place are reading half the story.

Why some brands keep showing up and others never do

Models do not invent vendor names out of thin air. They assemble them from patterns across the public web: review sites, comparison articles, forum threads, press coverage, documentation, and the collective weight of how often a name appears next to a specific problem. Three things push a brand into those answers. Topical proximity means your name shows up near the exact problem language buyers use, not just your category label. Third party corroboration means other sites mention you in a relevant context, because a model weighing ten sources will trust the name that appears in eight of them. Freshness means recent content carries more weight than a page from four years ago, which is why companies that stopped publishing in 2023 are quietly disappearing from answers now.

Here is my honest take: most B2B companies are losing this fight on third party corroboration, not on their own website. They have a beautiful product page nobody cites. According to Pew Research Center, a large share of Americans now usechatbots for information seeking, which means the pool of buyers encountering your category this way keeps widening whether or not your content strategy has caught up.

What marketing teams should measure, and what they should ignore

Raw mention counts are close to useless on their own. A brand mentioned forty times but never recommended is not winning anything. What matters is recommendation share, meaning how often you appear when someone asks for options in your category, and sentiment, meaning whether the framing helps or hurts.

Build a small prompt set that mirrors how buyers actually talk. Skip the tidy category terms your product team prefers. Write fifteen to twenty prompts the way a frustrated operations lead would type them at 9pm, then check them monthly across the major assistants. With a chatgpt visibility tracker, you can track four things:

● Whether you appear at all in the answer

● Where you land in the order of names

● Which sources the model cites when it mentions you

● Which competitors appear and where they sit

That fourth item is where the useful arguments live. If a competitor gets named in nine of twenty prompts and you get named in three, you have a specific gap to close, and it is usually a content gap, not a product gap.

The Friday afternoon audit that takes ninety minutes

You do not need a big program to start. I would run this before touching any budget.

1. Write your twenty buyer prompts in a shared doc. Plain language, no jargon.

2. Run each one in ChatGPT, Gemini, and Perplexity. Copy the answers into the doc, messy is fine.

3. Highlight every vendor name the model mentions. Count yours.

4. Look at which third party pages the models lean on. Note the domains.

5. Pick the three prompts where you are absent but should not be, and find the top cited page for each.

6. Ask yourself one question per page: could we earn a mention here with a better data point, a clearer definition, or a case study with real numbers?

That last step is the whole game. Models repeat what the web says, so if the most cited page in your category is a competitor’s comparison chart, your job is to become the source that page cites. When you disclose your own data or claims publicly, keep it accurate, because the Federal Trade Commission holds advertisers to substantiation standards and AI answers will happily repeat an overclaim back to your buyer.

Where AI answers fit in your funnel reporting

Do not try to attribute AI answers to pipeline in a clean numeric way yet. The path is too messy and you will burn weeks chasing it. Instead, treat it as a leading indicator. If recommendation share climbs across a stable prompt set, your downstream branded search and direct traffic should follow within a quarter or two.

Keep the measurement honest by documenting your method. NIST publishes general guidance on measurement and evaluation in its AI risk management framework, and the core idea applies here: any evaluation is only as trustworthy as the consistency of how you run it. Same prompts, same cadence, same recording method, or the numbers mean nothing.

What to do first

Pick one assistant, one buyer prompt where you are missing, and one page you could improve this month. Fix that, rerun the prompt, and see what changes. Small loops beat big programs in this channel because the underlying sources shift fast.

The teams I would bet on are not the ones with the biggest AI budgets. They are the ones who started a boring monthly habit a year ago and can now tell you exactly which prompts they own and which they have never cracked. Which of your buyer prompts would you fail today?