75% of LinkedIn's AI Citations Come From Employee Profiles, Not Company Pages

New research from Meltwater and LinkedIn analysed 9.5 million AI citations across six large language models. The headline finding reframes what an employee advocacy programme actually is: 75% of every LinkedIn citation surfaced inside an AI answer came from an individual member profile. Only 25% came from a Company Page.

If your B2B brand is invisible in ChatGPT, Gemini or Google AI Overviews, the problem is unlikely to be your Company Page. It is that nobody at your company is publishing in their own name.

This post breaks down what the study found, why the profile-to-page ratio is so lopsided, and the specific content formats that get cited. It is written for B2B marketing leaders, content teams and anyone running (or about to run) an advocacy programme.

The Numbers

The study, authored by Meltwater Chief Product Officer Chris Hackney and published in May 2026, tracked citations across ChatGPT 5, Google AI Mode, Google AI Overviews, Gemini 3.5 Pro, Microsoft Copilot and Claude Sonnet 4 over a four-week window, covering B2B categories from SaaS to financial services.

• Total AI citations analysed: 9.5 million
• LinkedIn citations from individual profiles: 75%
• LinkedIn citations from Company Pages: 25%
• LinkedIn's share of all AI citations: 0.53%, second only to YouTube at 1.52%
• Growth in LinkedIn citation share: up 26% across tracked models in four weeks
• User-generated platforms as a share of all citations: 47.5%
• Company websites as a share of all citations: 18.7%

LinkedIn now ranks first for AI citations in the AI & Data Science and Marketing & Advertising categories, second in Leadership & Strategy, Sales & Revenue and Financial Services, and third in Technology & SaaS and HR & Talent. For most B2B categories, LinkedIn sits inside the top five sources an AI model reaches for.

That is a structural shift, not a blip. We covered the early signals of it in why LinkedIn content now shows up in ChatGPT. This study is the first dataset large enough to tell you what to do about it.

Why Profiles Beat Pages Three to One

Three things are happening at once.

Models are trained to prefer attributable expertise. A named person with a job title, an employer and a track record reads as a source. A Company Page reads as marketing. When a model has to pick which of two LinkedIn posts to cite in an answer about vendor selection, the one attached to a human with a credible role wins.

Profiles simply publish more. Most B2B companies post three to five times a week from one Page. A programme with 40 active employees publishing twice a week produces 80 posts in the same period, across 40 different networks, angles and areas of expertise. Volume and topical spread both matter to retrieval.

Distribution feeds indexing. Posts that earn engagement travel further and get crawled more reliably. Our own data on posting frequency and impressions found accounts posting three or more times a week grew impressions 32%, while less frequent posters declined 6%. The same consistency that grows reach is what keeps a profile in the retrieval pool.

Notably, the study found the citation sweet spot is not mega-influencers. Creators with 1,000 to 10,000 followers accounted for 40% of citations, and those with 10,000 to 100,000 accounted for 38%. Your Head of Solutions Engineering with 4,000 connections is exactly the profile AI models cite.

What Gets Cited: The Format Data

Not all content performs equally. Here is the breakdown of cited LinkedIn content by format:

• Text posts: 72% of cited content
• Articles (LinkedIn Pulse): 12%
• Video: 11%

And by content type, ranked by share of citations:

1. "Best X" listicles: 54%
2. Side-by-side comparisons: 50%
3. "How to choose" guides: 33%
4. Educational explainers: 17%
5. Thought leadership with supporting data: 8%

The last line is the uncomfortable one. Opinion-led thought leadership, the format most executive ghostwriting programmes default to, is the weakest performer in the set. Models cite content that helps a reader make a decision, not content that positions a point of view.

Originality and recency matter too. 72% of citations went to original content rather than reshares, and 48% went to posts published in the previous three months while only 12% went to content over a year old. A library of evergreen assets does not protect you here. Citation share decays.

The Structural Traits of Cited Posts

The study examined the top-cited articles and found a consistent set of formatting characteristics:

• Bullet points or numbered lists: 100%
• Clear H2/H3 heading structure: 92%
• Named companies, tools or categories: 75%
• Quantitative data or hard numbers: 67%
• Comparison frameworks: 50%
• Explicit decision guidance: 33%
• Year included in the title: 25%

Every single top-cited article used bullets or numbered lists. Not most. All of them.

This is the same principle as structuring LinkedIn articles for AI citation: models extract passages, not documents. A post written as five unbroken paragraphs offers nothing clean to lift. A post with a labelled heading, a five-item list and a named comparison offers a ready-made answer.

What This Means for Employee Advocacy Programmes

For years, advocacy has been justified on reach and cost-per-click against paid social. That case still holds. But this research adds a second, harder-to-replace outcome: advocacy is now the primary mechanism through which a B2B brand enters AI-generated answers.

That changes three things in how you run the programme.

1. Coverage matters more than volume

If a model is answering "best employee advocacy platform for enterprise," it needs a cited source that names vendors and compares them. If nobody at your company has published anything on that topic, you are not in the consideration set, regardless of how many posts your Page ships.

Start by listing the 25 to 50 high-intent prompts a buyer in your category would actually type into ChatGPT. Then map which of your people could credibly answer each one. That map, not a content calendar, is the real plan.

2. Ownership beats rotation

The study's recommendation is to assign topic ownership to named internal experts and have each publish two to three posts a week. That is very different from a rota where everyone reshares the same company post. Reshared content accounted for only 28% of citations.

Practically, this means giving each advocate a lane. Your CFO owns procurement and pricing questions. Your Head of Customer Success owns implementation and onboarding. Your engineers own integration and security. Multi-profile management and team leaderboards make that ownership visible and trackable rather than aspirational.

3. Authenticity is now a technical requirement, not just an ethical one

Generic AI-generated posts published under a real name fail twice. They fail with human audiences, and LinkedIn has started letting members report AI slop directly. Suppressed content does not get engagement, does not get distributed, and does not get cited.

The fix is not to avoid AI assistance. It is to use tooling that writes in each employee's actual tone of voice and keeps a human in the approval loop. Tone of voice matching only works when it is built on a person's real publishing history rather than a generic prompt.

A Practical Checklist for Getting Cited

Use this as the brief for every advocate post that is meant to earn AI visibility:

• Lead with a question or a decision. "How to choose an employee advocacy platform" outperforms "Thoughts on advocacy."
• Name the audience in the first two lines. "If you run a 50-person B2B marketing team" tells a model who the answer applies to.
• Use descriptive subheadings. Not "Part Two." Say what the section covers.
• Include at least one list, framework or checklist. This is the single non-negotiable trait.
• Name real tools, companies, categories or trends. 75% of cited posts do.
• Include hard numbers. An original observation from your own data beats a recycled industry stat.
• End with a recommendation. Tell the reader what to do, explicitly.
• Publish original, not reshared. And publish recently. Refresh core topics quarterly.

Pair this with a consistent cadence. Two to three posts per expert per week is the study's benchmark, and scheduling that in advance is what makes it survivable for people with day jobs. If you are weighing up tooling, our comparison of employee advocacy platforms covers the options.

How to Measure It

Traditional advocacy metrics (impressions, engagement, clicks) still tell you whether content is travelling. They do not tell you whether it is being cited. Run both:

• Distribution metrics. Impressions, engagement rate, and the split between in-network and out-of-network reach, pulled from LinkedIn's official API.
• Citation metrics. Run your priority prompts through ChatGPT, Gemini, Perplexity and Copilot monthly and log which sources get cited. It is manual, and it is currently the only honest way to track it.

Track them together. A post with strong distribution and zero citations usually has a structure problem, not a topic problem.

The Takeaway

The 75/25 split is the number to take to your leadership team. AI models are citing people, and your Company Page cannot substitute for them. Brands that treat employee advocacy as a nice-to-have amplification layer will be structurally absent from the answers their buyers now read first.

Brands that build a real programme, with named topic owners, structured content and a consistent cadence, get to be the source.

If you are building that programme now, our complete employee advocacy strategy guide covers the operational side, and our guide to the best LinkedIn tools for B2B marketing covers the stack.

About the Data

All figures in this article are drawn from research conducted by Meltwater in partnership with LinkedIn, analysing 9.5 million AI citations across ChatGPT 5, Google AI Mode, Google AI Overviews, Gemini 3.5 Pro, Copilot and Claude Sonnet 4 over a four-week period. The findings were summarised by Meltwater CPO Chris Hackney in Social Media Today and announced in Meltwater's press release. Additional coverage is available from Demand Gen Report.

Frequently Asked Questions

What percentage of LinkedIn AI citations come from individual profiles?

75% of LinkedIn citations in AI-generated answers come from individual member profiles, and 25% come from Company Pages, according to Meltwater and LinkedIn's analysis of 9.5 million AI citations across six large language models.

Why do AI models cite individual LinkedIn profiles more than company pages?

Three reasons. Named individuals with job titles and employers read as attributable, credible sources while Company Pages read as promotional. Employee networks collectively publish far more content across more topics than a single Page. And personal posts typically earn more engagement, which improves distribution and the likelihood of being retrieved.

How many followers do you need to get cited by AI?

Fewer than most people assume. Creators with 1,000 to 10,000 followers accounted for 40% of citations and those with 10,000 to 100,000 accounted for 38%. Mid-sized, credible professional accounts outperform both very small and very large ones.

Which content formats get cited most by AI on LinkedIn?

Text posts account for 72% of cited content, articles 12% and video 11%. By type, "best X" listicles (54%) and side-by-side comparisons (50%) lead, followed by "how to choose" guides (33%). Opinion-led thought leadership without data performs worst at 8%.

Does old LinkedIn content still get cited?

Rarely. 48% of citations went to content published in the previous three months, while only 12% went to content more than 12 months old. Citation share decays, so core topics need refreshing quarterly.

How do you structure a LinkedIn post to be cited by AI?

Use a question-based or decision-led title, name the audience early, use descriptive H2/H3 subheadings, include at least one bullet list or numbered framework (100% of top-cited articles did), name real tools and companies, include hard numbers, and close with an explicit recommendation.

Is LinkedIn the most cited source in AI answers?

No. YouTube leads with 1.52% of all citations and LinkedIn is second at 0.53%, ahead of Reddit (0.44%), Capterra (0.38%) and Medium (0.21%). LinkedIn does rank first within the AI & Data Science and Marketing & Advertising categories.

How can employee advocacy software help with AI visibility?

Advocacy platforms make the required cadence achievable by supplying tone-matched content ideas, scheduling posts across many employee profiles, and reporting on reach through LinkedIn's official API. Vulse is built specifically for LinkedIn advocacy, with tone of voice matching, multi-profile management and live analytics. You can see pricing or talk to the team.

Rob Illidge is the founder and CEO of Vulse, the LinkedIn-native employee advocacy platform. Vulse is built on LinkedIn's official Marketing Developer Platform API and is ISO 27001 certified.