LinkedIn, ChatGPT & the 11% myth
Everyone repeats the same line: "LinkedIn is the second most-cited source by AI." I asked ChatGPT 50 questions, classified 458 citations, and cross-checked 130 domains with Ahrefs. Here's what the number really means — and what it leaves out.
For the past six months, one idea has been circulating in every LinkedIn post about AI visibility:
It's true — but it depends on which LLM platforms you use.
I've read a lot of studies, and it's hard to make sense of them all. So I decided to test it myself. The goal is to help you make better decisions when you think about your LinkedIn strategy, especially if you want to be more visible on ChatGPT.
What I did
In July 2026, I asked ChatGPT 50 questions in incognito mode, specifically about AI visibility — the 50 most-asked questions on Ahrefs for GEO in France. It's niche, but since it's my field, I could evaluate the answers with real precision: "what is GEO", "which agency should I choose", "how do I get visible on ChatGPT". I analysed all 50 responses line by line. 458 citations classified, 130 domains cross-checked with Ahrefs.
Here are my findings, as of right now. One important thing to keep in mind: this study only covers ChatGPT. Results differ on Perplexity or Gemini, for example. I'll publish a v2 of this article.
GEO, if you're new to the term: it's the art of appearing in AI answers — ChatGPT, Perplexity, Google's AI summaries. The equivalent of Google SEO, but for artificial intelligence. Let's start with ChatGPT. It's the most-used AI in France right now.
Out of the 458 citations ChatGPT used, LinkedIn appears only once. Reddit, thirteen times. YouTube, zero. How do you reconcile that with the famous "LinkedIn = 11% of AI citations"?
What the numbers from previous studies say — and what they leave out
The "11%" comes from a Semrush study published in March 2026, based on 325,000 prompts. It's a very serious study, with verifiable data. Like many studies, it has its limits. It measures ChatGPT Search — the mode with web search turned on. In that mode, LinkedIn is indeed cited in 14.3% of responses. That's already a bias. I wanted to test the opposite.
The mode I tested is standard ChatGPT, free version. No web search. The mode where you type your question and get an answer "from memory" (drawn from ChatGPT's internal database). And there, the story changes completely.
On the substantive questions — the ones that actually guide decisions — ChatGPT doesn't open the web. It answers from its internal database. Result: 35 out of 50 responses with no source at all. Zero citations. No LinkedIn, no Reddit, no YouTube, nobody. A 30,000-word manual recited from memory.
So when someone sells you "publish on LinkedIn and ChatGPT will cite you", ask the real question: in which mode? Because in the default mode — the one where you type your question without thinking about it — 7 out of 10 responses in my corpus triggered no web search at all. LinkedIn won't show up.
The numbers that do hold up
Once you get past the "magic LinkedIn" promise, some solid facts remain. Otterly analysed 1.31 million LinkedIn citations in June 2026. Three numbers to remember:
87.8% of cited LinkedIn URLs come from named individual authors. No company pages. No institutional content. People writing under their own name. For example, I'm not publishing this study on the Visible sur l'IA company page but through my personal profile.
70 to 74% of LinkedIn citations are Pulse articles — long-form articles, not short posts. On Microsoft Copilot, it's as high as 90%.
Short posts (50–299 words) still generate a share of citations — but only when they contain a complete, self-contained idea, not a status update.
And one number from Ahrefs that should make every SEO expert think twice: 80% of the pages most cited by AI don't rank in Google's top 100 (Ahrefs, 2025). In other words, being good at SEO predicts almost nothing about your AI visibility.
That's consistent with what I found on ChatGPT: the most-cited domain in my French corpus has a Domain Rating of... zero, and zero visitors. No SEO footprint whatsoever. Cited 8 times anyway.
What this changes for you
If you're an SMB, an agency, an entrepreneur — here's what I take away.
Every engine has its own citation graph
YouTube reigns on Perplexity and in Google's AI answers, disappears on ChatGPT. Reddit dominates everywhere except ChatGPT's standard mode. LinkedIn shines in ChatGPT's web search mode, fades in "memory" mode. Optimising "for AI" in general doesn't exist. You have to choose your engines and your surfaces.
And as an SEO professional with about ten years of experience, I always start with a solid website and a sound SEO strategy. You'll see results on AI at some point. These are not two separate strategies for your site.
It's people who get cited, not brands
87.8% of LinkedIn URLs cited by AI come from named individual authors (Otterly, 1.31 million citations). No company pages. No corporate posts. People writing under their own name.
For your brand, this shakes up the organisation: yes, long-form LinkedIn articles feed the AI — but mostly when they're published by individuals. A brand that wants to be cited needs to rely on employee ambassadors who share their experiences, their studies, their wins. Under their own name.
Important nuance: on Perplexity, it's the opposite — 59% of LinkedIn citations come from company pages. If you're investing, plan for both strategies. It can be really interesting to publish opinion pieces with employees from your company, and more corporate posts on your company pages, where you showcase the company's successes in a broader way.
Long-form articles for AI, short posts for your network
The right question isn't "should we publish on LinkedIn". It's: what type of content? The answer is clear: for AI, write long-form articles, under your own name, with a strong, verifiable point of view. Short posts serve your human network and the company page. Two different jobs.
Never start from "how do I write for AI"
Start from "what is AI already saying about my market". Test your queries in ChatGPT, Perplexity and AI Overviews before you start writing. Look at who it cites. Spot the blind spots. Those blind spots are your editorial opportunity. Not the universal recipes it's already recycling. Above all, add facts. Don't write something AI could write better than you.
What this study doesn't claim to be
I have to give you the limits of my study, because they matter.
It mostly covers B2B queries related to GEO and AI visibility — not every sector. If you're in consumer e-commerce or healthcare, your graph will look different. Run the test yourself — I found it fascinating. In fact, I'll keep running this kind of test, because they'll teach you far more than any GEO dictionary I could ever write.
It's a snapshot of July 2026. AI moves fast. A number that was true in February may no longer be true today. This kind of measurement should be redone at least once a year.
I don't have access to ChatGPT's internal data. I work in user mode, just like you. That's my limitation — and it's also what makes the study useful: it shows what someone asking a question actually sees, not what an OpenAI engineer sees.
And above all: even if you follow the numbers and the advice perfectly, nothing guarantees you'll be cited. This study documents leads, not a magic recipe. If someone sells you one, be wary.
The one line to remember
What I see from Australia — where Google's AI answers are already live, ahead of France — is that the GEO market is filling up with universal manuals sold as maps of the territory. They're not.
I work with clients across every industry, and what works best is running tests.
Read ChatGPT like a manual. Not like a map of your market.
The manual tells you what a beginner needs to learn. The map is yours to draw — with your data, your cases, your market, your language.
The full study
The complete study — methodology, numbers, limits, cases — is currently available in French only. An English translation is on its way; it will be published right here once it's ready.
Want the Excel workbook with the detail of all 458 citations? Email me at redaction@visiblesurlia.fr.
This newsletter lives in French
Visible sur l'IA is my French-language newsletter on LinkedIn — studies like this one, plus practical SEO and AI-visibility decoding, every week. If you read French, that's where the next study lands first.
Researched, tested and written by Roxane Pinault. — Roxane