TL;DR
- AI Mode cited LinkedIn on 23 of 100 professional answers, Gemini zero.
- Professional prompts multiplied LinkedIn's AI Mode share five times over consumer prompts.
- Feed posts and Pulse articles carried the citations, profiles took none.
- ChatGPT tagged every LinkedIn citation it returned with utm_source.
- Freeze one prompt set and one denominator before tracking anything.
Google AI Mode cited linkedin.com in 23 of 100 professional-intent answers we collected on 25 August 2026. Gemini cited it in zero of the same 100. On a consumer prompt set two days earlier, AI Mode’s count was 9 of 100. Which LinkedIn URLs get cited is just as lopsided: feed posts and Pulse articles carried almost every citation, and not one engine cited a profile page. LinkedIn’s AI visibility is a property of the prompts you measure with and the engine you measure on, and this study puts numbers on both.
Five engines, 100 professional prompts
We sent 100 professional-intent prompts to ChatGPT, Google AI Mode, Gemini, Perplexity and Copilot, US geo, on 25 August 2026. The prompts cover career moves, hiring, B2B tool choice, salary negotiation, workplace practices and professional certifications. No prompt contains the word “LinkedIn”. Fifty ask what practitioners say, fifty ask for a fact or a procedure. Here is LinkedIn’s share of the citation list each engine attached.
| Engine | LinkedIn share of citations | Answers citing LinkedIn | Citations counted |
|---|---|---|---|
| Google AI Mode | 4.4% | 23 of 100 | 896 |
| Perplexity | 1.4% | 7 of 56 | 562 |
| ChatGPT | 1.3% | 4 of 100 | 306 |
| Copilot | 0.9% | 4 of 100 | 1,063 |
| Gemini | 0% | 0 of 100 | 196 |
AI Mode is where LinkedIn lives. It returned 896 citations on this prompt set and 39 pointed at linkedin.com, which made LinkedIn the third most cited domain in the run behind reddit.com and youtube.com. The engines disagree from there. Gemini answered the same 100 questions with 196 citations and zero LinkedIn URLs. We keep finding this shape of result: four engines given the same question return four different source lists, and LinkedIn is one of the domains that swings hardest.
The split inside the prompt set matters as much as the engine. In AI Mode, LinkedIn took 6.9% of citations on the “what do people say” half and 1.9% on the factual half. In ChatGPT and Copilot every LinkedIn citation sat in the experience-seeking half, and the factual half produced none. That is the same phrasing effect that drove ChatGPT’s Reddit citation collapse in August 2026, now visible on a second domain.
The same engines on consumer prompts
Two days earlier we ran a 100-prompt consumer set through the same five engines for the Reddit study: product picks, travel, health, home troubleshooting. Recomputing LinkedIn’s numbers from those 500 raw responses gives the contrast.
| Engine | Professional prompts | Consumer prompts |
|---|---|---|
| Google AI Mode | 4.4% of citations, 23 of 100 answers | 0.9% of citations, 9 of 100 answers |
| ChatGPT | 1.3%, 4 of 100 | 0.7%, 2 of 100 |
| Copilot | 0.9%, 4 of 100 | 0.1%, 1 of 100 |
| Perplexity | 1.4%, 7 of 56 | 0.2%, 1 of 100 |
| Gemini | 0%, 0 of 100 | 0%, 0 of 100 |
Same engines, same country, same week. Moving the prompt set from consumer to professional multiplied LinkedIn’s AI Mode share by five and its answer count by 2.6. Any single “LinkedIn share of AI citations” number you see quoted is downstream of a prompt basket someone chose, which is the reason prompt research has to precede any AI visibility measurement. A tracker built on hiring and career prompts and one built on shopping prompts will disagree about LinkedIn every week, and neither has a bug.
Which LinkedIn URLs actually get cited
Every LinkedIn URL in the professional run falls into a handful of path types. Feed posts live under /posts/, long-form member articles under /pulse/, profiles under /in/, company pages under /company/.
| URL type | Google AI Mode | ChatGPT | Copilot | Perplexity | All engines |
|---|---|---|---|---|---|
| Feed posts (/posts/) | 26 | 0 | 0 | 6 | 32 |
| Articles (/pulse/) | 23 | 3 | 10 | 10 | 46 |
| Profiles (/in/) | 0 | 0 | 0 | 0 | 0 |
| Company pages (/company/) | 0 | 0 | 0 | 0 | 0 |
| Other (news, learning, advice) | 8 | 1 | 0 | 0 | 9 |
The content people publish is what gets cited. Feed posts from named individuals, recruiters arguing about one-page resumes, a compensation consultant explaining salary bands, and Pulse articles with a thesis in the headline. Profile pages and company pages took zero citations across the whole professional run. Profound’s tracking of real ChatGPT usage found the same direction of travel: within LinkedIn citations, profiles fell from 33.9% to 14.5% between November 2025 and February 2026 while feed posts rose from 20.9% to 26.0%, in their study of millions of real ChatGPT queries. Our snapshot points the same way, with profiles at zero. This is the mechanism behind founder-authored content earning AI citations at multiples of corporate content: the engines cite the post, and the post carries the person.
One detail from the capture: all 4 LinkedIn URLs ChatGPT cited arrived with a utm_source=chatgpt.com tag appended, and the other engines appended nothing. ChatGPT stamps its outbound citations, so LinkedIn’s own analytics can attribute that referral traffic while AI Mode’s and Copilot’s arrivals blend into generic referral rows.
Four true numbers for one engine
Take AI Mode’s professional run and count it four defensible ways. LinkedIn took 4.4% of the 896 citations. It appeared in 23 of 100 answers. It appeared in 23 of the 77 answers that returned any citation, which is 29.9%. Count every URL field the engine exposes instead of the citation list alone and the share reads 4.3% of 1,318 URLs. All four are correct. A reader who meets “4.4%” in one report and “29.9%” in another is looking at the same run.
Published LinkedIn figures live on different denominators too. LLM Pulse’s live leaderboard had linkedin.com sixth at 3.43% on 24 August 2026, and its stated denominator is the percentage of AI answers that cited the domain at least once, with zero-citation answers excluded. Profound’s URL-type percentages are shares of citations. A per-answer number and a per-citation number can differ by a factor of seven on the same data, so check the denominator before you compare anything to anything. The same discipline applies when share of voice in AI answers becomes a KPI: pick one denominator, freeze it, and compare it only to itself.
Run-to-run noise sits on top of the denominator problem. When we asked ChatGPT the same questions four times, only 12% of sources appeared in every run, so treat any single-run share, including ours, as one draw from a distribution.
What to do with this
If you want LinkedIn citations for yourself or your brand
- Publish posts and Pulse articles in the formats that earn LinkedIn citations, since those are the URL types engines cite. A polished profile earned zero citations in 456 professional answers.
- Write for experience-seeking questions. LinkedIn citations concentrated in the “what do recruiters actually think” half of our prompts, and the factual half produced almost none.
- Expect the wins in Google AI Mode first. It cited LinkedIn on 23 of 100 professional answers while Gemini cited it on none, and both are Google.
If you measure AI visibility
- Freeze a prompt set before you track anything, and label it. A professional basket and a consumer basket returned LinkedIn shares five times apart on the same engine in the same week. AI visibility numbers without the basket attached are not comparable.
- Name the denominator on every figure. Share of citations, share of answers, and share of answers-with-citations are three different numbers.
- Track engines separately. Blending AI Mode’s 4.4% with Gemini’s zero produces an average that describes neither. Our analysis of 42,971 AI citations in Google AI Mode and our guide to how ChatGPT Search picks sources show how differently the two retrieve.
Method and limits
456 valid responses collected 25 August 2026, US geo, across ChatGPT, Google AI Mode, Gemini, Perplexity and Copilot: 100 per engine, except Perplexity at 56. The 100-prompt professional set was fixed before collection, split 50/50 between practitioner-opinion and factual questions, and no prompt names LinkedIn. The consumer comparison recomputes LinkedIn’s numbers from the 500 five-engine responses behind our Reddit citation study, collected 23 August 2026 with the same pipeline. Shares are of the citation list each engine attached to its answer. We count URLs on the linkedin.com host itself, which leaves out two business.linkedin.com citations, one in AI Mode and one in Perplexity. URL types come from the first path segment of each LinkedIn URL. We run this with our own open-source GEO/AEO tracker.
- Perplexity is a partial arm. 57 of its 100 prompts failed at capture behind an auth wall. A 50-record retry recovered 13, 44 stayed failed, and every Perplexity figure uses the 56 valid responses. The other four engines returned 100 of 100.
- One engine configuration per platform, one collection window, one run per prompt. Citation lists vary run to run, so shares carry noise, and the response-level counts are the sturdier numbers.
- Visible citations only. An engine can read a page and not cite it, and consumer ChatGPT answers that ran no web search attach few links or none. On this run ChatGPT triggered web search on none of the 100 professional prompts, and its citation lists came from answers written without live retrieval.
- Two prompt sets cannot cover professional intent. Ours skews to career, hiring and B2B operations questions in English, US geo.
- Counts reflect what each engine exposes in its citation fields, and Copilot publishes that list under a sources label. Engines differ in how many URLs they attach, which is why per-citation and per-answer numbers are both reported.
LinkedIn sits inside a wider pattern where a few community and video platforms are the most-cited domains in AI answers. On professional questions it joins that group. On consumer questions it barely appears, and the number you quote depends entirely on which corner of that split you measured.