A Meltwater study of 9.5 million AI citations, run with LinkedIn, found LinkedIn is the second most-cited source in AI answers behind YouTube — and that three-quarters of its citations trace to individual member profiles rather than company pages, with more than half from creators under 10,000 followers.
2026-08-17 · by Moe Ameen
Meltwater, working with LinkedIn, analyzed roughly 9.5 million AI citations across more than a dozen B2B categories and six major AI platforms — including ChatGPT, Google's AI Mode and AI Overviews, Gemini, Copilot, and Claude — to see how professional content gets pulled into AI-generated answers. The headline number was already known: LinkedIn ranked as the second most-cited source of any kind in those answers, trailing only YouTube. The study was published May 12, 2026; its findings resurfaced widely in mid-August 2026, which is why they are being passed around again now.
The sharper finding is about who on LinkedIn gets cited. Roughly 75% of LinkedIn's citations came from individual member profiles — people's own posts and articles — while about 25% came from company pages. Follower count was not the gate: about 51% of the cited creators had fewer than 10,000 followers, which suggests the models were rewarding clarity, expertise, and usefulness rather than audience size. In other words, a mid-sized expert with a specific, well-structured post was more likely to be quoted than a large brand page publishing polished but generic messaging.
The cited content shared a consistent shape. In Meltwater's sample, essentially all of the most-cited posts used bullet points or numbered lists, about 92% had clear H2/H3 structure, around 75% named specific entities like companies or tools, and roughly 67% included statistics or hard data. Original content dominated over reshares, and fresh material — posted within the last few months — made up a large share of citations. The practical instruction Meltwater drew from it: identify your internal experts and help them turn what they know into public, structured, decision-led content — how-to guides, "how to choose" frameworks, comparisons, and ranked lists.
These are figures from a Meltwater–LinkedIn collaboration, so treat them as directional and platform-adjacent rather than independent, and note that AI engines and their exact citation behavior change often. The durable takeaway is structural and holds up against separate third-party studies: on LinkedIn, individual expert voices — not the company page — are the primary surface AI answer engines cite, and how a post is structured matters more than how many followers its author has.
The finding reframes the job: LinkedIn AI visibility is won by a *bench of individual experts* publishing structured, decision-led content, not by one polished company page. That is a supply-and-coordination problem, and it maps directly onto how [Kompozy](/) is built. Kompozy runs an [AI Influencer persona pool](/glossary/persona-brief) — many individual voices plus a brand identity, each with its own governed [Persona Brief](/glossary/persona-brief) — so a founder, a head of product, and three subject experts can each maintain a distinct, on-brand publishing stream instead of everything funneling through one account. From a single source — a talk, a call recording, a rough point of view — it generates net-new [Text Posts, long-form articles, and document carousels](/glossary/output-buckets) per voice, not just repurposed clips, and schedules them to LinkedIn and the other primary social platforms on [Autopilot](/glossary/autopilot) behind a per-post review gate.
The specific edge is that the content Kompozy produces already has the cited shape. Because the format the study rewards is explicit — lists, H2/H3 structure, named entities, hard data — you can hold every persona to that fingerprint in its brief, so scaling volume across a team means more structured, extractable expertise rather than more generic filler. The tactical move this week is a reaction post: "New data says 75% of LinkedIn's AI citations come from people, not company pages — here's what that means for your team." Feed the study into Kompozy and it becomes a blog explainer, a carousel of the individual-vs-company split, quote cards of the sharpest numbers, and a batch of captioned shorts, published everywhere in a day. Kompozy can't force an engine to cite you, but it removes the reason most teams never build the expert bench the data says gets cited: sustaining structured, original output across several individual voices by hand is the part that breaks by week three.
Individual profiles, by a wide margin. Meltwater's 2026 study of about 9.5 million AI citations, run with LinkedIn, found roughly 75% of LinkedIn's citations in AI answers came from individual member profiles and only about 25% from company pages. The practical read for B2B teams is that arming internal experts to publish structured, first-hand content beats leaning on a single corporate page.
Fewer than you might think. In Meltwater's study, about 51% of cited creators had under 10,000 followers, indicating that AI engines rewarded clarity, specific expertise, and usefulness over audience size. A well-structured, original post from a mid-sized account can be the source an engine quotes — reach mostly affects human distribution, not whether a model extracts your text.
Structured, decision-led, original content. In the Meltwater sample, essentially all of the most-cited posts used bullets or numbered lists, about 92% had clear H2/H3 headings, roughly 75% named specific entities, and about 67% included statistics or hard data. Formats that performed included how-to guides, "how to choose" frameworks, comparisons, and ranked lists — content shaped like an answer.
Yes. Meltwater's analysis ranked LinkedIn as the second most-cited source of any kind in AI answers, behind only YouTube, across more than a dozen B2B categories and six major AI platforms. Because so much of that citation flows to individual profiles, a well-structured personal LinkedIn presence is one of the higher-leverage places to publish content you want AI engines to repeat.