By Dan Nestle, Co-Founder & Chief Product & Intelligence Officer, Lilypath
Someone is going to make a decision about your client this week — maybe to run a story, take a meeting, send an RFP, or buy a product. Until recently, they would have Googled the name; now, they ask an AI. ChatGPT or Claude tells them who your client is in one confident, sourced paragraph, composed and finished before your client has said a word. They read it and walk away with an impression. Your client was never in the room for it. Neither were you.
I walked a room of communicators through this recently in a session that was meant to cover a lot of ground, yet it kept bending back to one subject: LinkedIn. We talked about content. Posting do’s and don’ts. But several questions were variations on a theme: what is the actual connection between a LinkedIn profile and what ChatGPT or Claude says about a person?
It’s a sharp question, and the answer changes how comms professionals should think about the platform.
From the Click to the Citation
Let’s start with what shifted. For two decades, being findable meant ranking. You competed for the click against a full page of results, and the reader did the choosing. Then, soon after the launch of ChatGPT in late 2022, the answer era arrived. Ask a model a question now, and it returns a single synthesized response, with no page of options to weigh, which shifts the whole game from ranking for a click to getting cited in the answer.
We’re already edging into a third phase, where autonomous agents act on a person’s behalf, doing the shortlisting, and the contest becomes whether your client gets chosen at all.
LinkedIn sits at the center of this for anyone in communications. Across the major AI platforms, it is the single most-cited source when a model is asked about a professional. Research by Profound, covered in Axios this spring, found that citation frequency has roughly doubled since late last year. A personal site, a Forbes mention, a podcast appearance, a Substack — all of it is out there. But when a model is asked, “Who is this person?” it turns to LinkedIn first.
A Different Machine Underneath
For twenty years, we’ve talked about “the LinkedIn algorithm,” as if there were just one. That algorithm is real, and it still asks the same thing every time someone opens the app. What should I show this person next? Its audience is humans scrolling, and its horizon is the next few hours.
A second system runs alongside it now, and it’s the one that answers the room’s question. Call it the citation engine. Its readers are other machines, LinkedIn’s own ranking models and, increasingly, public LLMs that pull LinkedIn as their first source for professionals. Its job is to answer one question. Who is this person?
It works on a much longer horizon, over months and years, and it optimizes for coherence: how well the profile, the posts, and the engagement history all tell the same story. That system is the bridge between a LinkedIn profile and what ChatGPT says about your client.
Under the hood, that work is split across several models, each with a narrower task. One of them reads the prose in your About section and your job descriptions and infers your domain expertise. It weighs the words far more than the skills list you’ve curated. If the headline says you’re a strategic consultant, and the About section mostly talks about comms, the system quietly files your profile under the comms.
What ties all of LinkedIn’s AI models together is the input. Each one reads the same text: the headline, the About section, the job descriptions, the posts, the comments. The words the feed reads to rank a post are the words the public models read to describe your client, two audiences who never meet, working from one set of prose.
A coherent profile gets read consistently everywhere it’s touched, and an incoherent one gets misread just as inconsistently, across every surface where the AI is shaping a decision. Get it right once, and it gets read correctly in both places.
What This Changes for the Work
Some of what follows stings.
Reach on any single post has collapsed; a typical post now lands with roughly half the audience it would have reached a couple of years ago, which makes chasing impressions one post at a time a slow way to lose. The audience that has actually grown is the one that follows a coherent identity over time. Substantive comments now outweigh likes by a wide margin, and a save signals more than any other interaction. Drop an external link into the body of a post, meanwhile, and the whole thing gets throttled.
The bigger shift is who the platform favors. Company pages have been squeezed down to a sliver of the feed while personal profiles command most of it. For communicators who’ve spent careers building brand channels, that’s a hard turn: the person now beats the logo, and the AI almost never cites a company page when answering a question about a human being.
So the discipline changes. The work is to know what the models say about your client today, to find where the profile contradicts itself, and to change the words that throw the systems off. All of that can be measured and fixed, which makes it a practice in its own right.
My Lilypath co-founders and I are building what we have come to call Authority Intelligence: reading how AI interprets a professional’s authority, then managing it on purpose. The work now is making sure that when the machine gets asked about your client, it already has something true and coherent to say. Lilypath is how you see what it’s saying, and what you can do to change it.
The platform rewards coherence over volume, one steady signal held long enough to become unmistakable.
That’s the part worth sitting with. In an AI-mediated world, a professional identity is infrastructure. It’s load-bearing, the thing every system reads before anyone gets near your client’s actual ideas. And staying silent doesn’t keep your client safely neutral. Refuse to shape what the machine reads, and it writes the description anyway, from whatever fragments it can find, and hands it to the room your client will never enter.