A study published in May 2026 analysed 9.5 million AI citations and reported that LinkedIn is the second most cited source in AI answers, behind YouTube. The finding is real, but the number that matters sits next to it: LinkedIn’s citation share in that dataset was 0.53 per cent, against YouTube’s 1.52 per cent. It is second place in a field where almost nobody holds much share, measured inside B2B prompts, over four weeks, by a company that sells the tool used to measure it. Read that way, the study is still useful. It just does not say that LinkedIn is where AI visibility is won.
We wrote about this question in September, before any numbers existed, and said the honest answer was to test it on your own topics over eight weeks. Numbers exist now. They change what we would look for in that test. They do not remove the need for it.
What the study measured
Meltwater ran prompts across six systems using its own GenAI Lens product: ChatGPT 5, Google AI Mode, Google AI Overviews, Gemini 3.5 Pro, Copilot and Claude Sonnet 4. The prompts were drawn from B2B categories. The window was four weeks. The output was 9.5 million citations, scored for which domains the systems referenced.
Two things follow from that design, and both are easy to lose when the finding gets repeated.
The first is scope. The prompt set was B2B, so the ranking describes B2B questions, not AI search generally. Where the scope widens, LinkedIn moves. Ahrefs tracks citations across a broad set of United States queries on all topics and has Wikipedia at 10.1 per cent, Reddit at 6.2 per cent and YouTube at 2.3 per cent, with LinkedIn outside the top fifty domains entirely. Peec AI analysed 30 million cited sources across five systems and put LinkedIn third overall. Three studies, three different places for LinkedIn. The scope is doing the work, not the platform.
The second is who benefits. Meltwater sells the measurement tool. The version of the write-up that most people read sat on LinkedIn’s own marketing blog, next to a button that sells LinkedIn ads. Neither party is lying, and the underlying data looks carefully collected. But when a vendor’s study finds that the vendor’s category matters more than people thought, the claim deserves the same scrutiny you would give a case study.
Worth knowing if you compare the two versions: the LinkedIn-hosted post lists Gemini 2.5 Pro where Meltwater’s own write-up says Gemini 3.5 Pro, and it carries figures for CEOs at 8.2 per cent and founders at 7.5 per cent that do not appear in Meltwater’s version at all. We have not used those two figures below.
The findings we would act on
Set the ranking aside. The composition of the citations is the part with practical value, because it says something about what gets quoted rather than how much.
Personal profiles, not the company page
Across the systems analysed, 75 per cent of LinkedIn citations came from individual member profiles and 25 per cent from company pages. More useful still: 51 per cent of cited creators had fewer than 10,000 followers, and accounts in the 1,000 to 10,000 band accounted for 40 per cent of citations.
That matches what we see on client work. A language model assessing whether a passage is worth quoting is reading the passage and the signals around the author, not the follower count. A production manager in Chakan with 1,800 connections who writes clearly about injection moulding tolerances is a better citation candidate than a company page posting festival greetings. This is the same mechanism behind experience and expertise signals on your own site, applied to a borrowed platform.
Structure, in specific proportions
Among the 24 most cited LinkedIn articles in the study, the shared features were consistent:
| Feature | Share of the most cited articles |
|---|---|
| Bullet lists or numbered items | 100% |
| Clear H2 or H3 headings | 92% |
| Named companies or tools | 75% |
| Hard numbers and data | 67% |
| Comparison or evaluation frameworks | 50% |
| A “how to choose” decision guide | 33% |
| A year in the title | 25% |
By format, “best X” listicles made up 54 per cent of the most cited content and side-by-side comparisons 50 per cent. Thought leadership supported by data came last at 8 per cent. None of this is new to anyone who has read our note on writing passages an assistant can lift cleanly, but it is the first time we have seen the proportions put to a number across a large sample.
Recency is doing real work
Posts published in the previous three months accounted for 48 per cent of citations. Anything older than twelve months accounted for 12 per cent. Original posts took 72 per cent against 28 per cent for reshares.
The practical reading is that a LinkedIn presence does not accumulate the way a website does. A page on your own domain can earn citations for years. A post decays. If you intend to be quoted from LinkedIn, that means a publishing rhythm, not a campaign.
What the study cannot tell you
Four weeks is a short window, and a 26 per cent rise in citation share inside it is as likely to reflect model updates during the period as a durable trend. There is no per-brand data, so nothing here shows that a given company posting more got cited more. Crawler access is still unresolved: a great deal of LinkedIn is only partly visible without signing in, and the study does not establish what each system can actually read. And a B2B average says nothing about your category. LinkedIn ranked first in AI and Data Science and in Marketing and Advertising. If you sell CNC machining or run a hotel, you are not in those categories.
Which is why the eight-week test still stands. The study tells you what the winning posts look like. It does not tell you whether assistants cite LinkedIn for the ten questions your buyers actually ask.
Where this sits in our order of work
It does not move. Your website comes first, because it is the asset you control and the one that compounds. LinkedIn is a distribution and corroboration layer on top of it.
The reason is in the citation itself. When an assistant quotes a LinkedIn post, the link goes to linkedin.com. Your name appears in the sentence, which has real value as a brand mention, but the destination is not yours and the post will fade within a year. A page on your own domain answering the same question keeps earning. So the sequence we use with clients is unchanged: build the pages that answer buyer questions, publish the evidence only you have, then use LinkedIn to put that work in front of the people who will mention it elsewhere.
What the study does change is how we brief the LinkedIn part. We now ask clients to get two or three people with real operational knowledge posting under their own names, rather than routing everything through the company page.
What to do this month
- Pick two or three people in the business with genuine depth in one subject. Follower count does not matter. Operational detail does.
- Have each of them write one comparison or “how to choose” piece for buyers in your category, with real options named and real numbers.
- Put the answer in the first two lines, before the “see more” fold.
- Use headings and numbered lists inside long posts and articles. The most cited articles in the study did this without exception.
- Name specific tools, suppliers, standards and prices where you honestly can. Vagueness is what keeps a passage from being quotable.
- Publish the same substance as a page on your own site, and link to it.
- Track whether assistants start naming you, and check whether they cite your domain or linkedin.com. The difference matters.
Frequently asked questions
Should we shift budget from our website to LinkedIn?
No. The study measures how often systems cite linkedin.com, not how often they cite you. A domain with 0.53 per cent citation share across a B2B prompt set is not a reason to move spending away from the asset you own. Treat LinkedIn as the place your expertise travels, not where it lives.
Does the company page still matter?
Yes, for a different job. A complete company page helps systems resolve who you are and connect your name to the right entity, which is the work described in our note on why assistants fail to mention a company. It is identity infrastructure. The quoting happens mostly on personal profiles.
Do we need to post every week to stay cited?
On LinkedIn, the recency data suggests a steady rhythm matters more than volume in bursts. On your own site, it does not work that way: a well-made page holds up. That asymmetry is a good reason to keep the owned pages as the foundation.
Is the study trustworthy?
The data collection looks sound and the sample is large. The framing is promotional, and the headline rank depends entirely on the B2B scope. Use the composition findings, which are specific and checkable. Hold the ranking lightly.