75% of LinkedIn's AI citations come from people, not brand pages

LinkedIn published a playbook this month telling brands how to get their content cited by AI assistants, built around a framework it calls the Credibility Stack. It is unusual for a platform to document this explicitly, and the timing is not accidental: independent research has been converging on the same finding all year, and it is an awkward one for marketing teams. The LinkedIn content AI systems quote is overwhelmingly published by named individuals, not by the company page your social budget funds.

What the research actually found

Three separate datasets are worth separating, because they were built differently and they answer different questions.

Meltwater analysed 9.5 million AI citations across 16 B2B categories using its GenAI Lens product and reported in May 2026 that LinkedIn ranks second only to YouTube among cited sources. Its platform-level finding is the widely quoted one: LinkedIn, Reddit and YouTube together account for 47.5% of AI citations, against 18.7% from company websites and 15% from peer review sites. Within LinkedIn specifically, about 75% of cited content came from individual members and about 25% from company pages, and 51% of citations came from accounts with fewer than 10,000 followers.

Semrush approached it from the URL side, analysing roughly 89,000 LinkedIn URLs that appeared as citations in AI search, and looked at what characteristics those URLs shared.

LinkedIn itself published guidance in September 2026 organised around the Credibility Stack, establishing a foundational presence, then extending it through staff profiles and relevant creators. Its reported figures point in a consistent direction: long-form articles in the 500 to 2,000 word range plus mid-length posts of roughly 50 to 299 words account for the largest share of citations, LinkedIn articles make up somewhere between half and two-thirds of cited LinkedIn content across the major AI platforms, and 54% to 64% of cited posts are ones that share knowledge or practical advice. Reshares are almost never cited.

A platform publishing research that recommends using that platform more deserves scepticism, which is exactly why the Meltwater number matters: it was produced independently, by a media intelligence company with no stake in LinkedIn's engagement figures, and it reached the same conclusion.

Why individuals outrank company pages here

The finding looks strange until you consider what a language model has to decide when selecting a source. It needs a reason to believe the text is grounded in something. A LinkedIn profile supplies structured, machine-readable evidence of that: a job title, an employer, an industry, a work history, all attached to a named human writing about their own field. A company page supplies a brand name and a marketing department.

The follower-count finding is the corroborating detail. If citation selection were driven by popularity, the 51% share coming from accounts under 10,000 followers would be inexplicable. It is not popularity being rewarded. It is specificity, the thing a practitioner writes about their own work that a brand account, writing for everyone, systematically rounds off.

Who should publish what

The useful conclusion is not "post from personal accounts instead." Company pages still do jobs that individual profiles cannot, and a brand that abandons its page loses the surface that ad campaigns, recruitment and basic verification depend on. The allocation below reflects what each surface is actually good at, given the citation data.

This is essentially an argument for employee-led publishing with editorial support behind it, which is a different operating model from a social calendar, and one we have written about in the context of employee-generated content and creator networks.

What makes a passage quotable

Across the studies, the structural characteristics of cited content are consistent and unremarkable: clear formatting with bullet points and numbered lists, strong descriptive headings, named entities, and quantitative data. None of that is a trick. It describes writing that can be lifted out of its surroundings and still make sense, which is precisely what an AI system needs to do when it uses you as a source.

Practically, that means each substantial piece should contain a handful of passages that survive being extracted:

1. A direct answer in the first two sentences under each heading. Not context, not a wind-up, the answer, then the reasoning. A paragraph that takes four sentences to reach its point cannot be quoted in one.
2. Specific numbers with their source and date attached in the same sentence. "Meltwater analysed 9.5 million AI citations in May 2026 and found 75% came from individual profiles" is quotable. "Recent research suggests personal accounts perform better" is not, because a system that repeats it has attributed nothing.
3. Named entities. Companies, products, standards, people, dates. Entities are how a model connects your passage to the rest of what it knows; a paragraph of unattributed abstractions connects to nothing.
4. Conditions stated explicitly. "This applies to B2B categories; the same pattern has not been demonstrated for consumer brands." Bounded claims are safer to cite, and systems selecting sources are, in effect, managing risk.
5. A conclusion that does not require the surrounding paragraphs. If your key sentence only makes sense after two paragraphs of setup, it will not be extracted, however good the setup is.

These are the same properties that make a passage useful to a human skimming on a phone, which is why this is not really an AI-specific technique. It is the answer-first structure that also underpins answer engine and generative engine optimisation on your own site.

What this data does not say

Three caveats worth holding, because most coverage of these studies omits all of them.

Citation share is not the same as being the best source. LinkedIn, Reddit and YouTube's combined 47.5% partly reflects that they are large, well-structured, openly accessible and consistently formatted, properties that make content easy to retrieve and quote. A more authoritative analysis sitting behind a login or a paywall is not cited, and that says something about crawlability, not quality.

Citations are not traffic, and traffic is not revenue. Being quoted in an AI answer may generate a click, a brand impression with no click, or nothing measurable at all. Treat AI citation as a brand visibility outcome, and be sceptical of anyone selling it as a lead generation channel with a clean attribution story, because that story does not currently exist.

You do not own any of it. Every one of these findings describes performance on a platform whose ranking and distribution rules can change without notice, and where your best-performing content is attached to an employee who may leave. That is the strongest practical argument for the mirroring recommendation above: the article publishes on LinkedIn where it can be found and cited, and a canonical version lives on a domain you control, so the asset survives both the platform and the personnel. Independent research has separately found that AI systems cite earned media far more than brand-owned pages, a pattern we covered when Muck Rack analysed 25 million AI citations, and the same lesson applies: visibility you rent should always have an owned counterpart.

Frequently asked questions

Do AI assistants cite LinkedIn content?

Frequently. Meltwater's May 2026 analysis of 9.5 million AI citations across 16 B2B categories found LinkedIn is the second most-cited source in AI answers, behind only YouTube. LinkedIn, Reddit and YouTube together accounted for 47.5% of citations, compared with 18.7% from company websites.

Do company pages or personal profiles get cited more?

Personal profiles, by a wide margin. Roughly 75% of LinkedIn citations in Meltwater's dataset came from individual member profiles and about 25% from company pages. Profile metadata such as job title, employer and industry gives an AI system evidence that the author has relevant expertise; a company page supplies a brand name without that grounding.

Do you need a large following to be cited?

No. In Meltwater's analysis, 51% of citations came from members with fewer than 10,000 followers. Citation selection appears to track specificity and usefulness rather than audience size, which is why practitioners writing about their own narrow area outperform larger generalist accounts.

What LinkedIn format gets cited most?

Long-form articles. LinkedIn's own September 2026 guidance reports that articles in the 500 to 2,000 word range account for the largest share of cited content, making up between roughly half and two-thirds of cited LinkedIn material across major AI platforms. Mid-length posts of about 50 to 299 words also perform. Reshares are rarely cited at all.

What is LinkedIn's Credibility Stack?

It is the framework in LinkedIn's September 2026 playbook for improving the odds of being cited by AI assistants. It works outward from a foundational brand presence, then extends credibility through staff profiles and relevant creators, on the basis that AI systems weigh identifiable human expertise heavily when selecting sources.

Does an AI citation actually drive business results?

Treat it as brand visibility rather than a lead channel. A citation may produce a click, an impression with no click, or nothing you can measure, and there is currently no reliable attribution path from an AI answer to a booked enquiry. That is a reason to publish a canonical version on a domain you own, where the traffic you do earn can be measured and converted.

The takeaway

The finding underneath all of this is older than AI search: specific, attributable expertise from a named person outperforms anonymous brand copy. What has changed is that the machines now selecting sources read profile metadata, prefer extractable passages, and reward quantitative specificity, so the gap between the two has become measurable. If your organic LinkedIn effort runs entirely through the company page, you are publishing into the 25% share and competing for the least-cited quarter of the platform.