LinkedIn is the second most-cited domain in AI answers. That changes where you publish

LinkedIn now ranks second among all domains cited in AI search answers. That is an unusual fact, because it means a rented surface is outcompeting almost every owned site for the thing every marketing team is currently chasing. The useful question is not whether to publish there. It is which content belongs there and which content must stay on your own domain, because getting that split wrong means competing with LinkedIn for your own words.

What the data actually says

The study to know is Semrush's, published on 10 March 2026 and run in collaboration with LinkedIn. It analysed 325,000 unique prompts submitted during January and February 2026 to ChatGPT Search, Google AI Mode and Perplexity, identified 89,000 LinkedIn URLs cited in the responses, and used LinkedIn's own data to describe what those URLs were.

Two findings do the work. LinkedIn is the second most-cited domain globally across the three tools, appearing in roughly 11% of AI responses on average. And that average conceals a spread wide enough to change strategy.

The inversion between Perplexity and the other two is the most actionable line in the dataset. A brand that publishes only from its company page is reaching the engine that cites LinkedIn least often, in the format that engine happens to prefer — and is largely absent from the citation pools of the two engines that cite LinkedIn nearly three times as often.

On format, LinkedIn's own figures, published in its 2026 guidance on AI visibility, put long-form articles and newsletters at around 60% of all AI citations of LinkedIn content, with feed posts accounting for somewhere between 15% and 28% depending on the platform. The length bands that get cited most are 500 to 2,000 words for articles and 50 to 299 words for feed posts. Reshares are rarely referenced at all.

Three further details from the study are worth holding on to, because they contradict the way most people optimise LinkedIn:

• Most cited posts carry moderate engagement, in the region of 15 to 25 reactions. Virality is not the selection mechanism.
• Around 75% of cited authors publish frequently — five or more posts in a four-week window. Consistency beats intensity.
• Between 54% and 64% of cited posts share knowledge or practical advice rather than news, opinion or announcements.

Read together, that describes a publishing profile that most brand social calendars are not built to produce: a named person, publishing useful specifics, often, at moderate reach.

The publishing-surface matrix

The decision this data forces is not "should we post on LinkedIn more". It is which content jobs belong on which surface, given that each surface earns a different thing. Citations, rankings and pipeline are three different outcomes, and the same article does not maximise all three.

The pattern to take from that table: anything you want to rank, convert or be the canonical source for goes on your domain first. Anything whose value is the author's credibility goes on LinkedIn, under a person's name. Those are different jobs, and treating LinkedIn as a distribution channel for the first category is how brands end up competing with LinkedIn for their own words.

The republishing rule set

Cross-posting is where this gets mishandled, usually by pasting a blog post into a LinkedIn article and hoping. Google does not penalise duplication as such; it picks one version as canonical and shows that. The risk is that it picks LinkedIn's copy, because LinkedIn has more authority than your domain does, and you then rank for your own article on somebody else's page with no conversion path. Six rules keep that from happening.

1. Own domain first, and confirm indexation before you adapt. Publish, check it is indexed and eligible for a snippet, then produce the LinkedIn version. Pages have to be indexed and snippet-eligible to be considered as supporting sources in AI answers in the first place, so this sequence protects the asset that does the most work.
2. Adapt, do not mirror. Rewrite the opening, cut to a single argument, drop the sections that exist for search coverage. The LinkedIn version should read like it was written for LinkedIn, because a near-duplicate is the case where a canonical gets picked against you.
3. Keep the citable block intact, and keep it on both. The definition, the figure with its source and date, the numbered method: those are what get lifted. They should be equally clean in both places. This is not the duplication that hurts you, and it is the part that earns the citation.
4. Link back in the body, not just at the end. One contextual link from inside the LinkedIn article to the fuller version on your site, placed where a reader would want it, not parked in a sign-off nobody reads.
5. Never republish a page you are actively trying to rank for a head term. If the page is your bet on a commercial query, write something adjacent for LinkedIn instead and link to it.
6. Attribute every figure with a source and a date, in both versions. Unattributed numbers are the least citable form of writing, because an engine has no way to stand behind them.

The anatomy of a citable passage

Every framework for AI visibility eventually reduces to whether a specific passage can be lifted out of your page and still be true, attributable and useful. That is a writing problem with a repeatable shape. Compare these two versions of the same claim.

The second version works because it has the five properties an extractable passage needs: a specific claim, a number, a named source, a date, and a stated condition or comparison. It survives being separated from the rest of the page, which is precisely what happens to it. The first version cannot be cited because there is nothing in it to attribute.

The practical test: cut one paragraph out of your draft, read it on its own, and ask whether it would answer a question for someone who never sees the rest. If it needs the paragraph above it to make sense, it will not be selected. This is the same discipline as writing for featured snippets, applied to a longer unit, and it is the core of what we cover in AEO and GEO for AI search.

Who should publish, and how often

Given that individual members account for 59% of LinkedIn citations on both ChatGPT Search and Google AI Mode, and that around three quarters of cited authors post five or more times in four weeks, the operating model follows fairly directly. Two or three named people publishing consistently will outperform a company page publishing more.

That is a sustainable schedule for a small team: four short posts, one article and one owned-site piece a month, from one or two named people. It is considerably less volume than most social calendars demand, aimed at a different outcome.

What this does not solve

Two honest limits. First, a citation is not a visit. Being named in an AI answer builds recognition and occasionally sends a click, but it will not show up in your analytics as a channel you can reconcile, and a strategy justified purely on citation counts is hard to defend at budget time. Measure it as a brand-awareness input, alongside branded search volume and direct traffic, not as an acquisition channel.

Second, none of this substitutes for your own site. LinkedIn's authority is why it gets cited, and that authority is not yours: a rented surface can change its terms, its algorithm or its indexation policy without consulting you. The durable asset is still a well-built site with content people cite, which is the argument we made in the llms.txt piece and it has not changed. LinkedIn is a leverage layer on top of owned content, not a replacement for it. Where earned media fits into the same picture is covered in AI search and earned media, and both feed the same underlying goal: being the source an answer engine can safely stand behind.

If you want help building the owned layer that all of this leans on, that is what our content production work is for.

Frequently asked questions

How often is LinkedIn cited in AI search answers?

LinkedIn is the second most-cited domain globally and appears in roughly 11% of AI responses on average, according to a Semrush study run with LinkedIn and published on 10 March 2026, based on 325,000 unique prompts from January and February 2026. The average hides a wide spread: ChatGPT Search cites LinkedIn in 14.3% of responses, Google AI Mode in 13.5% and Perplexity in only 5.3%.

Do LinkedIn articles get cited more than feed posts?

Yes, by a wide margin. LinkedIn's own 2026 figures put long-form articles and newsletters at around 60% of all AI citations of LinkedIn content, with feed posts accounting for between 15% and 28% depending on the platform. The length bands cited most often are 500 to 2,000 words for articles and 50 to 299 words for feed posts. Reshares are rarely referenced at all.

Should we publish from our company page or from personal profiles?

Both, weighted towards people. Individual members account for 59% of LinkedIn citations on ChatGPT Search and 59% on Google AI Mode, the two engines that cite LinkedIn most. Company pages account for 59% of LinkedIn citations on Perplexity, which cites LinkedIn least. A brand publishing only from its company page is therefore strongest on the weakest engine and largely absent from the two strongest.

Does republishing blog posts on LinkedIn hurt SEO?

Not through a duplicate-content penalty, because Google generally selects a canonical version rather than penalising duplication. The real risk is that it selects LinkedIn's copy, since LinkedIn has more domain authority than most business sites, leaving you ranking for your own article on a page you do not control and cannot convert from. Publish on your own domain first, confirm it is indexed, then publish an adapted rather than identical version on LinkedIn with a contextual link back.

Do I need a lot of followers or viral posts to get cited by AI?

No. The Semrush study found most cited posts carry moderate engagement, in the region of 15 to 25 reactions, so virality is not the selection mechanism. What does correlate is publishing frequency, with around 75% of cited authors publishing five or more times in a four-week window, and content type, with 54% to 64% of cited posts sharing knowledge or practical advice rather than news or announcements.

How do I write a passage an AI answer engine will cite?

Make it survive being separated from the page. A citable passage has five properties: a specific claim, a number, a named source, a date, and a stated condition or comparison. Test it by cutting one paragraph out of your draft and reading it alone. If it needs the paragraph above it to make sense, it will not be selected. Vague claims such as "long-form content performs better" cannot be cited because there is nothing in them to attribute.

The takeaway

The split is simple once you accept that citations, rankings and pipeline are different outcomes. Anything you want to rank for, convert from or be the canonical source of goes on your own domain first: definitions, original data, frameworks, pricing, case detail. Anything whose value is a person's credibility goes on LinkedIn under that person's name: practical how-tos as monthly articles, points of view as weekly posts. Two named people publishing consistently at moderate reach will beat a company page publishing more, because that is what the citation data describes. And write passages that stand alone, with a number, a source and a date, because the ones that cannot be lifted will not be.