LinkedIn citations have a follower floor. Clicks don't.
Meltwater's 2026 LinkedIn study puts 65% of AI citations on accounts over 10k followers plus company pages. Read with the 75/25 cut, it moves budget.
Two figures published this week describe the same system from opposite ends. Read together, they settle an argument B2B marketing has been having for a year.
On 12 November, Robert Yanik, AI search marketing lead for LinkedIn Ads, published the platform's read on Meltwater's 2026 analysis: 75% of LinkedIn citations in AI chatbot answers come from individual member profiles, and 25% from Company Pages. Three days later, Social Media Today reported another cut of the same study: accounts with more than 10,000 followers, together with company pages, account for 65% of those citations. Accounts under 10,000 followers hold the other 35%.
"People get cited more than pages" is the headline everyone will quote. The second cut is the one that moves money. Held together, the two say the citation layer has a follower floor and the click layer does not.
Two cuts of one study
| Cut | Figure | What it actually says |
|---|---|---|
| Profile vs. page | 75% individual profiles, 25% Company Pages | The citation surface is people, not logos |
| Reach vs. page | 65% over 10k followers plus company pages, 35% under 10k | Most of that surface belongs to people with reach |
Subtract one from the other and you get a number neither publication printed: company pages 25, individual accounts above 10,000 around 40, individual accounts below 10,000 around 35. That middle figure is my arithmetic, not a Meltwater finding, and it should be read that way. But it is the most useful way to hold the two cuts together, because it kills the comfortable reading that "individual profiles win" means any individual profile will do.
The company page is a quarter of the pool and the only asset on that list you fully control. Large individual accounts are the biggest single block. Everything under 10,000 followers shares something close to a third.
The advice that breaks
For two years the standard LinkedIn advice has been post more, post consistently, let the ranking learn you. That advice made sense when the reward was reach. It gets expensive when the reward is a citation, because a citation is a vote on credibility, and the study says credibility is being read off identity markers a posting schedule cannot manufacture.
Yanik is direct about what the models weigh: a verified identity with employer history, skills, and a professional network is more evaluable than a post from a brand. He also notes that C-suite and VP-level articles earn higher citation rates per article than non-executive content, and that profile metadata such as title, company and industry works as a direct authority signal. That is a different game from impressions. A junior account with a forty-post month is still a junior account.
None of this shuts the small account out. A third of the citation pool is real and reachable. The point is that the third is won on specificity: one defensible claim, argued at length, in a place with a stable URL. Volume competes for the 65. Specificity competes for the 35, and the 35 is contested far less.
Why a model prefers a person with a job title
An anonymous forum post and a post from a named person with a verified employer are not the same input. The model is trying to answer "who would know this," and a LinkedIn profile answers that question in structured form: who they work for, what they have done, who they are connected to, what they list as expertise. A logo answers none of it.
LinkedIn's own guidance leans on the same idea from the other side. Its credibility-stack argument is that brand content sets the point of view, employees humanize it, creators scale it and customers validate it, and that corroboration across those voices is what makes a claim citable. That is a useful correction to the 75/25 headline: people are cited more often, but a person citing a claim that exists only on their own profile is a weak signal. The person is the carrier; the brand page is where the claim lives long enough for someone else to echo it.
Which is why "should we invest in the company page or the founder" is a badly formed question. They are different instruments, and the mistake is choosing between them instead of assigning each one a job.
Split by job
The operating fix is an allocation, not a preference, and it is usually the first thing we restructure in a social media engagement where LinkedIn is one channel among four. Each row below has a different asset, a different cadence and a different metric. The failure mode is running one row's cadence against another row's metric.
| Job | Asset | Format | Cadence | Metric |
|---|---|---|---|---|
| Demand capture | Exec or employee profile, amplified with Thought Leader Ads | One argument per post, no round-ups | Weekly | Click-through, cost per qualified conversation |
| AI citation | Company page plus two or three named experts | Long-form article with one defensible claim | Monthly | Presence in a fixed prompt set |
| Validation | Customer or partner voice | Short, specific outcome | Quarterly | Whether sales actually uses it |
| Niche reach | Creator partner | Native to their audience, not your template | Per agreement | Qualified traffic, not impressions |
Two notes on reading that table. The citation row is deliberately slow: an article that states a position the model cannot find anywhere else is worth more than five posts restating the category. And the demand-capture row is the one place where the personal profile beats the page on measured click behavior, which is the reason Thought Leader Ads outperformed company page creative for us in the first place. Keep that row narrow, too: the LinkedIn premium only pays when the audience is tight, and the same rule applies here. Clicks reward a face. Citations reward a track record. Those are not always the same face.
Measure it with five prompts, not a dashboard
There is no LinkedIn metric called "AI citations," so every team working on this ends up building the same small instrument. It takes an hour a month.
Pick five prompts a real buyer would type. One should name your category without naming you, two should describe the problem you solve, one should ask for a shortlist, and one should name your brand so you can see what the model says about you rather than to you. Run them in two or three assistants on the same day each month. For every answer, record which LinkedIn account or page is cited, the URL, the model and the date.
Three rules make it usable. Count a citation only when the surfaced URL is a LinkedIn post or article, not a profile the model merely mentioned. Compare the same prompt set month over month, because answers move and a new prompt set resets your baseline. And treat a single answer as noise: the signal is the same voice appearing across several prompts and persisting.
We run this ledger on client engagements as a check on whether the work is landing, not as a promise about any particular model. The useful finding is rarely "we are not cited." It is which voice gets cited when we are, because that decides whether the next month of content goes to the page or to a person.
The cheapest fix is a profile field
Before buying anything, spend an afternoon on the fields the model actually reads. On every profile you intend to use:
- Title states the function, not an internal grade. Head of Performance Media beats Growth Ninja.
- Company is selected from the page entity, not typed as free text. A typed string does not link the person to the company.
- Industry is filled in and consistent with the page.
- Experience has dated entries. A profile with three companies and no dates is a weaker signal than one with a history.
- Skills include the specific capability you want associated with the claim, not twelve generic ones.
- Verification is done wherever the platform offers it. Identity is the input being weighted.
None of that is content work, and it is consistently the part nobody does. A team will spend a quarter on a posting calendar and skip the ten minutes it takes to make sure the founder's profile links to the company entity it is supposed to belong to.
What not to buy
Follower growth as a goal. The 10,000 line is where citations cluster, which makes it tempting to buy your way across it. But the study describes citations following authority that happens to correlate with audience size. A purchased audience changes the number and not the correlation, and the model is reading title, history and corroboration, not a follower count.
Comment volume. On 15 November, LinkedIn's chief product officer, Hari Srinivasan, said new classifiers now flag suspected AI-generated comments as less valuable and rank them lower, with a reported drop of more than 20% in views of those comments during testing. Comment pods have been a target for longer. Any strategy whose unit of work is "more comments" is now working against the platform's ranking.
A ghostwritten cadence that sounds like a press release. What is being evaluated is a person with a demonstrated track record. A posting schedule attached to a voice nobody recognizes produces the format without the signal, and readers spot it faster than any classifier does.
If nobody on the team is near 10,000
Then the plan is narrow, and it is still worth running.
Put the defensible claim on the company page, in long form, with a date and a named author. Give two people with real, narrow expertise a monthly article each, on subjects where they can be wrong in public. Get one external placement so the claim exists somewhere you do not control, which is the corroboration the credibility-stack argument is pointing at. Measure all of it with the five-prompt ledger and give it a quarter before you judge it.
What you should not do is run a follower-growth campaign and wait for citations to arrive. Citations and clicks are two different jobs on the same platform, and the profile audit that follows from them is mostly unglamorous field work, the same discipline that makes a citation profile readable to an assistant in the first place. One budget can serve both jobs. It just cannot be aimed at a single page and expect to hit both.
Frequently Asked Questions
How many LinkedIn citations in AI answers come from personal profiles?
LinkedIn's own AI search lead cites Meltwater's 2026 analysis: 75% of LinkedIn citations in AI chatbot responses come from individual member profiles and 25% from Company Pages. A second cut of the same study adds that accounts with more than 10,000 followers plus company pages account for 65% of citations, leaving 35% for smaller accounts.
Do you need 10,000 LinkedIn followers to be cited by ChatGPT or Gemini?
No, but reach is doing more of the work than most B2B teams assume. Accounts above 10,000 followers and company pages together hold about 65% of citations in Meltwater's data, which leaves roughly a third of the pool for smaller accounts. That third is reachable with specific long-form arguments rather than posting volume.
How do you measure whether a brand shows up in AI answers?
Fix a short set of buyer prompts, run it against two or three assistants on the same day each month, and record which LinkedIn account or page is cited, along with the model and the date. Judge the trend across three runs. One answer is noise; the same voice appearing across several prompts is a signal you can act on.



