If AI can't explain you, your competitor gets the recommendation: how to get your company recommended by ChatGPT
Buyers are asking AI for a shortlist before they ever visit a website. A company the model cannot describe is a company that does not exist.
On April 16 an agency founder pitched me an llms.txt service. I have heard a lot of pitches. This elevator line stuck:
If you cannot be understood by these AI systems. It's like you don't exist and your competitors are the ones showing up.
He was selling a file. The sentence is bigger than the file.
Discovery moved: buyers ask the model first
A week earlier, on April 10, a founder building in beauty tech told me the same thing from the other side of the market: product discovery is moving into LLMs. People who used to search, scroll and compare are asking a model to do it for them.
The agency founder described what that looks like. A Google search is a few words. An AI search is a long, conversational question, closer to how you would ask a colleague. "We are a 40-person B2B company selling to banks, who should we use for outbound?" The model answers with names. If yours is not one of them, you were never in the running, and you will never know it happened.
That is the scary part. Lost search traffic shows up in your analytics. Lost AI recommendations do not show up anywhere.
People refuse to be sold to, but they need help buying
The beauty-tech founder put the buyer psychology in two lines that apply to B2B just as well:
they refuse to be sold to.
But they all need help buying.
That is exactly why people ask a model. It feels like advice, not a pitch. Nobody is trying to close them. So brands have to move from selling to helping customers buy: being the clearest, most consistent, most widely cited answer to the question the buyer is asking.
What makes a company understandable to a model
From the agency founder's description, AI systems weigh three things when deciding whether to mention you: clarity, consistency and how citable you are.
Clarity means a model can say in one sentence who you serve and what you do. If your homepage says "the future of intelligent revenue orchestration", the model has nothing to repeat.
Consistency means the same description shows up everywhere. Your site, your LinkedIn, your founder's bio, directory listings, interviews. If you describe yourself five different ways, the model averages them into mush.
Citability means other people talk about you in a way the model can quote. That is the one founders underestimate.
llms.txt: cheap, but not magic
Here is where I part ways with the pitch. An llms.txt file is a plain-text summary of your site for AI systems. It is cheap to make and I think you should have one.
But public analysis points out that no major AI platform has confirmed it reads llms.txt, and that third-party citations across the web carry more weight. A file on your own server is you describing yourself. A model trusts what other people say about you more than what you say about yourself, the same way a buyer does.
A related idea I saved back in February: the pages chatbots seem to cite most are comparisons and "best X for Y" lists, so AI visibility is won by being the clearest entry in those comparisons, not by stuffing keywords. I treat that as a hypothesis, not a law, but it matches what I see. Models repeat what the web already says in a structured way.
Why a show creates the citations models read
If the game is third-party mentions, the question becomes how to create a lot of them, in your own words, consistently.
A show does that as a side effect. Every episode is a page, a transcript, a video description and a set of clips, each describing what you do in the same language. Every guest shares it with their audience, which puts your name on pages you do not own. Every guest's company links to the conversation. And the transcripts are exactly the long, conversational text that models learn from.
Fifty episodes is fifty consistent, citable, third-party-adjacent descriptions of who you are. No single file does that.
Fix your own one-liner first
Before you chase citations, make sure there is something clear to cite. Write one sentence that says who you serve and what changes for them. Put it at the top of your homepage, in your LinkedIn company description, in every founder bio and in your llms.txt, word for word.
Then write the comparison page nobody in your category has written well: you against the obvious alternatives, honest about where each one fits. Models repeat structure. Give them a clean, fair comparison and they have something to quote when a buyer asks "what are my options?".
Check what the model says about you today
Open ChatGPT, Claude and Perplexity. Ask each one the question your buyer would ask, in full sentences, with your category and your customer in it. Write down who shows up.
Then ask each model directly what your company does. If the answer is wrong, vague or missing, fix clarity on your own site first, then go build the citations.
The companies that will win AI recommendations in two years are the ones creating clear, consistent public conversations now. See how a show builds your public footprint.
Frequently asked questions
Does llms.txt help you show up in ChatGPT?
It is cheap and worth having, but no major AI platform has confirmed it reads llms.txt. Third-party citations across the web appear to carry more weight than a file describing yourself.
How do B2B buyers use AI to build a vendor shortlist?
They ask long, conversational questions that include their size, industry and need, and the model answers with named companies. If you are not named, you were never considered.
What makes AI models cite a company?
A clear one-sentence description, the same description used consistently everywhere, and plenty of third-party pages that mention you. Interviews, guest appearances and comparison pages help most.