LinkedIn Citations: Which Page Types AI Models Pick
LinkedIn citation page types differed sharply by AI engine in a May to June 2026 sample. Here is how to read the shares without mistaking them for causes.
LinkedIn is several kinds of website under one domain. A company page is an identity record. A regular post captures a moment in the feed. A Pulse article is a longer, bylined page. When an AI answer cites linkedin.com, the domain alone does not reveal which of those surfaces supplied the information.
The Promptwatch LinkedIn page-type report separates them. Its measurement window runs from May 18 through June 17, 2026 and covers LinkedIn citations found in ChatGPT, Google AI Mode, Google AI Overviews, and Perplexity. In the combined view, Pulse articles accounted for 37.67% of LinkedIn citations, regular posts for 32.19%, and company pages for 13.35%.
Those percentages use all LinkedIn citations in the window as the denominator, including categories not displayed in the main chart. They do not mean that 37.67% of AI answers cited a Pulse article. They also do not say how often LinkedIn appeared compared with other domains. This report starts after a LinkedIn citation has already been observed, then classifies the page.
The pooled result hides four different patterns
ChatGPT's LinkedIn citations leaned much more toward company pages than the combined result suggests. Company pages represented 23.84% of its LinkedIn citations during the May 18 to June 17 window. Pulse articles were 8.77% in that engine-specific view.
Google's two surfaces looked different. Pulse articles made up 44.85% of LinkedIn citations in AI Mode and 42.25% in AI Overviews. In Perplexity, ordinary posts led at 41.88%, while Pulse articles accounted for 32.46%.
This is why "publish more on LinkedIn" is not a measurement plan. A company trying to improve how ChatGPT describes its business would inspect its company page and the exact pages ChatGPT already cites. A publisher studying AI Mode or AI Overviews would have more reason to examine long-form article coverage. A Perplexity-focused test could include regular posts. Each is an engine-specific hypothesis drawn from an observed source mix.
None of the percentages proves that changing format will cause a citation. The engines may have received different prompts. Page age, topic, author identity, external references, indexability, and current retrieval conditions can all differ. The report classifies selected pages; it does not randomize otherwise identical content across LinkedIn formats.
Citation share is conditional on LinkedIn appearing
There are two denominators worth keeping apart. Page-type share asks, "Of the LinkedIn citations we observed, what portion went to this type?" LinkedIn response rate asks, "Of all eligible AI responses, what portion cited any LinkedIn page?" A page type can gain share while LinkedIn as a whole appears less often.
The wider social citation report addresses the second layer by comparing social domains within monitored citation data. It is a live page, so any exact value from it needs a snapshot date. It should not be merged into the fixed May 18 to June 17 page-type window without labeling the periods separately.
Pooled data needs another caution. The combined LinkedIn view reflects the citations collected across the included engines. It should not be read as four engines casting equal votes. If one engine contributes more LinkedIn citations to the pool, its pattern can have more influence on the combined distribution. The per-engine panels avoid that ambiguity.
Even within an engine, percentages can conceal concentration. Twenty citations to twenty separate articles are different from twenty citations to one article repeated across similar prompts. Save exact URLs and count unique pages as well as total citation events.
Audit LinkedIn as a set of source surfaces
Begin with prompts for which professional identity, company facts, or practitioner commentary could reasonably matter. Separate branded prompts from category questions. Branded prompts are more likely to retrieve identity pages, so mixing them into a topical content cohort can make company pages look stronger than they are for discovery.
For each run, retain the response and classify any LinkedIn URL. A workable set of fields includes engine, prompt, observed date, page type, author or organization, publication date when available, and whether the cited page supports the statement beside it. Also note whether the answer names your brand. A citation to an employee's post can influence an answer without creating a clear company mention.
Review rates by engine before looking at a combined number. Then compare at least two measures:
- page-type share among all LinkedIn citations
- percentage of eligible responses containing that page type
The first describes the mix inside LinkedIn. The second describes coverage across your monitored answers. Unique URL count and repeat citation rate add context without pretending to be substitutes for either denominator.
Keep the prompt set and engine configuration stable during a test. If prompts are added, version the cohort and report old and new prompts separately for an overlap period. Otherwise a new batch of company-research questions could create an apparent rise in company-page citations without any retrieval change.
Match the page to the information task
The report gives a sensible starting point for content maintenance. Company pages should contain accurate, current company facts because they are plausible identity sources and had a notable share in ChatGPT's LinkedIn citations during the study window. Long-form articles can answer a complete professional question on one URL, which makes them worth testing on the Google surfaces represented in the data. Regular posts deserve their own Perplexity cohort because they led that engine's observed mix.
This is not a call to duplicate one article in every LinkedIn format. Duplication makes attribution harder, and it can leave several pages with conflicting facts. Choose the format that fits the information, keep claims sourced, and observe which exact URL is retrieved.
The AI search KPI guide published on August 11, 2026 draws a useful line between mentions and citations. A mention names a brand. A citation links to a source. LinkedIn measurement should preserve that distinction, especially when an employee, company, and third-party commentator can all appear in one answer.
For a practical tracking setup, Promptwatch can retain prompt responses and break citations out by source. Our Promptwatch review covers the product beyond this use case. The reason to use a tracking platform here is mundane but useful: page type, engine, prompt, and date stay attached, so the team does not have to reconstruct a trend from screenshots.
The May 18 to June 17, 2026 data supports one firm conclusion: LinkedIn citation mix varied substantially by engine. It does not establish a universal best page type. The next step is to test the relevant format against your own prompt cohort and keep its denominator visible.