LinkedIn is now AI search's top cited source for professional queries

For most of the last year, the GEO advice for LinkedIn amounted to a shrug: have a presence, keep it current, move on. Profound's data says that advice expired sometime around December. LinkedIn didn't just get cited more. It got cited differently, and the difference tells you exactly what to publish next.

Methodology

Profound published the full dataset and writeup as part of its ongoing AI-visibility tracking work. The firm built this dataset from two sources layered together: a seven-day rolling average of real ChatGPT user prompts, and a synthetic basket of professional queries run consistently across six platforms, ChatGPT, Gemini, Google AI Overviews, Google AI Mode, Microsoft Copilot, and Perplexity. Citation tracking ran GDPR- and CCPA-compliant from November 15, 2025 through February 15, 2026, a three-month window, and produced 1.4 million tracked citations in total. That combination matters: real prompts capture what people actually ask, and the synthetic basket makes the platform comparison apples-to-apples instead of an artifact of which queries happened to get sampled that week.

The professional-query framing is doing real work here too. This isn't a claim that LinkedIn beats Wikipedia or Reddit on every kind of prompt an AI engine answers. It's a claim about a specific query category, business, careers, and industry-expertise questions, where LinkedIn now outperforms every other domain tracked, across every platform tracked. Outside that category, the study says nothing either way.

Two things about the setup are worth flagging before the numbers, because they change how much weight the findings deserve. First, running the same synthetic query basket across all six platforms is what makes the cross-platform comparison meaningful; without it, a rank difference between ChatGPT and Perplexity could just as easily reflect different query mixes as a real citation-behavior difference. Second, a rolling seven-day window on the real-prompt side smooths out single-day noise, a spam wave, a viral post, a platform outage, without smoothing out the three-month trend the study is actually built to detect. Neither choice eliminates every confound, but both are the right calls for the question being asked.

The headline shift

The rank movement alone is the kind of number that gets a domain into a keynote slide: from roughly 11th place in ChatGPT citations in mid-November to roughly 5th by mid-February, more than doubling citation frequency in three months. Rank changes that fast for a domain LinkedIn's size don't usually happen because of one algorithm update. They happen because the underlying content on the platform changed enough, in volume or in kind, to give retrieval systems something new to pull from.

Read the table left to right and the story is a swap, not just growth. Profiles went from the single largest citation type to roughly half their starting share, a drop of nearly twenty points. Feed posts and articles together picked up almost exactly what profiles lost. LinkedIn as a domain got cited more, but the specific pages doing the work moved from static bios to dated, claim-bearing posts, which is a meaningfully different kind of authority signal for a GEO strategy to chase.

It's tempting to read "LinkedIn is the #1 cited domain for professional queries" as a platform-level win and stop there. That's the wrong altitude. Domain-level rank is a useful headline, but it's an aggregate of page-level decisions an engine makes one citation at a time, and this dataset is unusually good at showing what those individual decisions favored during the window. A company that reads only the rank number and concludes "we should be more active on LinkedIn" will likely keep doing what got profiles cited in 2024. A company that reads the composition shift underneath the rank number gets a specific, buildable instruction instead.

What changed inside LinkedIn's citations

A profile is a snapshot: a headline, a job history, an About section that rarely changes month to month. It's useful to an engine answering "who is this person" but nearly useless for answering "what's the current thinking on X." A feed post or a long-form article is the opposite. It's timestamped, it makes a specific claim, and if it's any good it's structured enough for an engine to lift a self-contained fact without needing the rest of the page for context. That's close to the exact list of signals we've tracked in our own citation work, extractable structure, demonstrated authority, and machine access, and it's a reasonable explanation for why the mix moved the direction it did.

There's a second explanation that isn't mutually exclusive with the first: volume. LinkedIn's own reporting throughout the study window described record levels of member posting activity, and more posts in the index simply means more candidate material for an engine's retrieval layer to surface. A generative engine doesn't need every post to be excellent. It needs enough well-structured, recent posts on a topic that at least one of them clears the bar when a query comes in. Rising post volume and rising post quality both push in the same direction here, and this dataset can't fully separate how much credit belongs to each.

Why feed posts and articles are winning over profiles

Put yourself in the retrieval system's position for a second. Asked something like "what's the current thinking on outbound multichannel sequencing," an engine is not well served by a profile that says someone has fifteen years in sales leadership. It's well served by a post from three weeks ago that states a specific position, backs it with a number, and reads as a self-contained answer. That is, not coincidentally, also the exact content shape the six link types AI engines actually trust piece we published in June described as the pattern across every trusted source category, not just LinkedIn. Community and first-person expertise content keeps winning across platforms because it matches the shape retrieval systems are built to reward: current, specific, and self-contained.

This lines up with a pattern we've now seen across several unrelated citation datasets this year, not just LinkedIn's: static, institutional pages are steadily losing ground to dynamic, individually-authored content everywhere an engine has a choice between the two. It happened with corporate About pages losing share to founder interviews. It's happening with product marketing pages losing share to comparison content written by practitioners rather than vendors. LinkedIn profiles versus posts is the same pattern showing up on a new surface, which is part of why this dataset reads as more than a LinkedIn-specific curiosity.

It's also worth naming what this data does not say. It doesn't say your LinkedIn profile stopped mattering entirely, only that its share of the citation pie shrank as posts and articles grew. A profile still functions as the credential an engine checks once it has already decided to cite the person behind a post. Losing citation share is not the same as losing relevance; it's a division-of-labor change, where the profile establishes who you are and the post is what actually gets quoted.

The volume story and the structure story reinforce each other in a way that's easy to miss. More posting activity gives an engine more raw material to choose from, but raw material alone doesn't explain why the choices skewed toward posts and articles specifically rather than, say, comments or company-page updates. The likelier read is that posts and articles are simply the LinkedIn content types built to make a single, complete point, which happens to be exactly what a retrieval system is trying to extract when it answers a professional query. Comments are fragments of someone else's thread. Company pages are institutional and rarely make a falsifiable claim. Posts and articles are closer to a self-contained mini-essay, and that shape travels well into an AI answer.

What this means if you sell to B2B buyers

For a company selling into a buying committee, this data reframes what "having a LinkedIn presence" should mean operationally. A company page with occasional updates and a handful of employee profiles is the tier-1 professional-query citation strategy from 2024. The data says the surface that's actually earning citations now is individual, dated, claim-bearing posts and native long-form articles from people the company employs, not the institutional account.

This also changes who inside a company should own the work. GEO for most surfaces, your own site, comparison pages, docs, is fundamentally a content and technical SEO function. GEO for LinkedIn is closer to an internal comms and executive-visibility function: it lives or dies on whether real people at the company are willing to post their actual opinions under their own names, on a schedule, in public. That's a harder org problem than shipping a content calendar, and it's exactly the reason most companies still default to the institutional company page. The company page is easier to control. It's also, per this dataset, the losing bet.

There's a buying-committee angle specific to B2B that makes this shift matter more than a general-purpose citation study would. Enterprise buyers researching a vendor decision are disproportionately likely to run a "who's saying what" query, comparing named practitioners' takes rather than product marketing copy, precisely the professional-query category this study covers. If an AI engine's answer to "what do people in this space actually think about X" increasingly pulls from LinkedIn posts, then having zero employees posting is not a neutral absence. It's ceding that specific answer to whichever competitor's team does show up in it.

Where the data has limits

This is a single-domain study, and it's worth being precise about what a single-domain study can and can't establish. It shows LinkedIn's citation composition changed and shows the direction and magnitude of that change. It does not, on its own, prove that publishing more feed posts causes an engine to cite your company, and it doesn't isolate LinkedIn's shift from broader changes in how these six platforms retrieve and rank sources generally during the same window. The professional-query framing is also narrower than it sounds in a headline: it's LinkedIn's rank for business, career, and industry-expertise prompts specifically, not a claim about AI search citations in general, where domains like Wikipedia, Reddit, and G2 still dominate different query categories entirely.

The study also doesn't report anything about downstream outcomes, click-through, conversion, or pipeline attributed to a citation. A citation is visibility, not revenue, and treating the two as interchangeable is the same mistake we've flagged in our own AI visibility measurement coverage: the mention is the leading indicator, not the result.

One more limit worth stating plainly: three months is a real trend, but it is not a permanent state. Platforms adjust retrieval logic, LinkedIn's own algorithm changes what gets distributed into feeds, and a shift this sharp over one quarter could partially mean-revert or could keep compounding. Treat the direction, content-type composition moving toward posts and articles, as the durable signal, and treat the specific percentages as a snapshot of one window rather than a permanent ratio to build a five-year plan around. Re-run this same audit against your own citation tracking every quarter, not once, and watch whether the direction holds before committing headcount to it.

The audit to run this quarter

Pull your company's LinkedIn output from the last ninety days and sort it into the same three buckets Profound tracked: profile updates, feed posts, and native long-form articles. If the split looks like November 2025, mostly static profile activity with sporadic posting, that's the gap. Assign two or three named people inside the company to publish a specific, numbered claim on LinkedIn weekly, in the post composer and the native article tool, not just as a link to your blog.

Set a floor before you start, not after: at minimum, one native long-form article a month per named contributor, plus weekly posts that each make one falsifiable claim, a number, a named comparison, a verdict, rather than a general observation about the industry. General observations don't get cited because they don't say anything an engine can lift and attribute. A post that states "reply rates fell below 1% in most verticals this quarter" is citable. A post that says "outbound is evolving fast" is not, no matter how many people like it.

Then check the effort against a generative engine optimization or content marketing plan built for the B2B SaaS buying committees who are increasingly asking AI engines these exact professional questions before they ever open your site. The domain-level ranking is Profound's finding. What you publish on it next, and whether it's shaped like a claim an engine can quote or a bio nobody's asking about, is yours.

Profound - "LinkedIn is the most-cited domain for professional queries in AI search" - 2026
https://www.tryprofound.com/blog/linkedin-is-the-most-cited-domain-for-professional-queries-in-ai-search
Media Copilot - coverage and corroboration of the Profound dataset - 2026
https://mediacopilot.ai/linkedin-top-cited-ai-search-profound/

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.