LLMs are picking winners in your niche
Why AI search tools recommend some experts and skip others, and what LinkedIn presence has to do with it.
PostHog published something worth reading last month. Their traffic from LLM referrals grew 41x in two years, and it converts better than almost any other source they have. The uncomfortable corollary: not all of their products get recommended at the same rate. Some are well-known to the models; others are effectively invisible. We think the same asymmetry is playing out right now across B2B founders and operators, and the gap is widening.
LLMs recommend people and products based on the citation inventory they were trained on and the sources they can retrieve in real time. For B2B founders and operators, LinkedIn is one of the highest-signal surfaces feeding that inventory. Operators with consistent, specific, citable content on LinkedIn are far more likely to show up when a buyer asks an AI tool "who should I talk to about X." Those without it are not. The window to build that presence before recommendations calcify around a short list of recognizable names is shorter than most people think.
How LLMs decide who to recommend
When someone asks ChatGPT, Claude, or Perplexity "who is the best fractional CMO for B2B SaaS" or "which consultant should I hire for enterprise pricing strategy," the model is not running a Google search and surfacing whoever has the most backlinks. It is drawing on two things: what was in its training data, and what it can retrieve right now from high-authority sources.
Training data is the longer game. It reflects what has been written about you, what you have written, and where that writing lives. Retrieval is the shorter game. Perplexity and ChatGPT's browsing mode pull live content from sources they trust. LinkedIn is one of them. So is your company blog. So are publications that quote you.
The PostHog team found that their best-known products get recommended consistently, while newer or less-documented products get skipped entirely. The models know what they know. If you have not created a clear, consistent, retrievable body of work that answers the questions your buyers are asking, the model will recommend someone who has.
This is not a future problem. Across the founders and operators we work with, we are already seeing inbound come in with the note "an AI tool suggested I reach out to you." That was rare 18 months ago. It is no longer rare.
Why LinkedIn specifically
There are three reasons LinkedIn matters more than most operators expect for LLM visibility.
First, LinkedIn posts are indexed and retrievable. Perplexity pulls them. ChatGPT's browsing mode pulls them. When a model is trying to answer "who are the credible voices on go-to-market for vertical SaaS," it is pulling from content that exists in retrievable form, and LinkedIn posts meet that bar in a way that a Slack message or a podcast appearance does not.
Second, LinkedIn has a trust signal that general web content lacks. The platform attaches professional identity to content. A post about pricing strategy written by someone whose profile says "VP of Product at three Series B companies" carries a different signal than an anonymous blog post. Models that care about source authority register that.
Third, volume creates citation density. One good post is not enough. The operators who show up in AI recommendations tend to have published consistently on a narrow set of topics over a meaningful period of time. The models see a coherent body of work and treat it as authoritative. One-off posts, even excellent ones, do not build that.
We wrote about how LinkedIn has become a source signal for AI search tools in more depth here, and the same dynamics we described then are accelerating now.
The citation inventory problem
Here is the practical issue: most B2B founders have not built a citation inventory on LinkedIn. They have posted occasionally, usually about company news or surface-level commentary on trends. When we run content audits, we consistently find that founders have expertise that is nowhere in their public record.
A founder who has spent five years building and selling to mid-market professional services firms knows specific things: how those buyers evaluate, what objections they raise at what stage, which integrations matter and which do not. That knowledge is an asset. If it only lives in sales calls and private Slack channels, it does not exist for the purposes of LLM recommendations.
The fix is not complicated, but it is consistent. You need a body of public work that answers the questions your buyers are already asking AI tools. That means publishing on LinkedIn with enough specificity that a model ingesting your content can accurately describe what you do and who you are for.
Vague is the enemy here. "I help founders scale" is not citable. "In eight engagements with Series A SaaS companies, the CAC problem turned out to be a segmentation problem four times" is citable. The model can do something with the second sentence. It cannot do anything with the first.
What the models are actually trained to look for
AEO (answer engine optimization) is the emerging field PostHog references. We do not love the term because it has already attracted the same keyword-stuffing instinct that ruined early SEO. But the underlying principle is sound: if you want to be recommended by AI tools, you need to write content that answers specific questions accurately, in depth, in a format that is easy to cite.
That translates to a few practical things on LinkedIn.
Specificity beats breadth. A founder who has 40 posts about pricing strategy for B2B SaaS will be recognized by models as an expert on that topic. A founder who has 40 posts across 15 topics will be recognized as someone who posts a lot. Own a lane.
First-person data beats general commentary. "According to Gartner..." is a worse citation anchor than "across the 12 companies I've worked with in this segment." Models are trained on a lot of Gartner paraphrasing. They are trained on less specific practitioner data. First-person specificity stands out.
Frequency creates the training signal. PostHog's 41x growth in LLM traffic did not come from one good piece of content. It came from a consistent volume of useful, citable writing over two years. The same principle applies to individuals. A single excellent post in March will not make you a recommended expert in November. Twelve good posts per month for twelve months might.
We covered the compounding logic of consistent LinkedIn presence in detail in our piece on how Google's no-click data changes the distribution math. The argument there applies with even more force to AI-driven recommendations, where there is no second page of results.
The short list problem
Here is what makes this urgent. LLMs do not return a long tail of results the way Google does. When a buyer asks an AI tool for a recommendation, they get three to five names, occasionally fewer. The model is not going to say "here are 47 options, sorted by relevance." It is going to say "based on what I know, these are the people worth talking to."
That is a radically different competitive dynamic than traditional search. In traditional search, being on page two is mediocre but not catastrophic. In AI search, not being in the top three to five is functionally invisible. The buyer closes the chat, contacts whoever was recommended, and never thinks about anyone else.
The short list is already forming in most niches. Not because one player has locked it up, but because a small number of operators have built enough of a citation inventory that the models default to them. The operators who are not on that list yet have a narrowing window to build their way onto it.
This is not hypothetical. Run the prompt yourself. Open ChatGPT or Claude, turn off memory, and ask: "who are the best consultants for [your specific niche]?" See who shows up. If you are not in the results, ask a follow-up: "what would I search to find someone who does what [your name] does?" If the model does not know your name, that is your baseline.
What to do this week
The practical starting point is the same one PostHog recommends for products, applied to people.
Run three to five prompts that your buyers would realistically use. "Best [your role] for [your niche]." "Who should I hire if I need help with [your core problem]." "Recommended [your category] consultant." Do this across ChatGPT, Claude, and Gemini. Note where you appear, where you are missing, and where you are misrepresented.
Then look at your LinkedIn content from the last six months and ask honestly: does this add up to a citable body of work on a specific topic? If the answer is no, you know what to build.
The good news is that the bar for individual operators is lower than it is for products. PostHog is competing against dozens of well-funded, well-documented analytics platforms. You are probably competing against a handful of operators who have not yet realized this game is being played. Most of them are posting sporadically about general trends. You do not need to outpublish them. You need to out-specify them.
Forty posts over six months, all on the same specific problem you solve, written with enough first-person data and practitioner detail that a model can accurately describe what you do. That is the investment. The return is showing up on a short list that your buyers consult before they ever open a search engine.
“
The models are not going to return a long tail of results. They are going to return three to five names. That short list is forming now.
”
The founders we see winning on LinkedIn right now are not the ones with the most followers. They are the ones building the kind of public record that makes a model confident enough to recommend them by name. That is a different game than follower growth, and it is the one worth playing.
Frequently asked
LLMs draw on two things: content from their training data and real-time retrieval from trusted sources. For individual experts, that means the models look at what you have written publicly, where it is hosted, and how consistently you have addressed a specific topic. LinkedIn posts, blog articles, and quotes in publications all feed that inventory. Operators with a consistent body of work on a narrow topic are more likely to be recommended than those who post occasionally across many topics.
