Which AI engine cites LinkedIn the most?

By absolute count, Grok (122 total rows, 27 actually cited). By citation rate once retrieved, ChatGPT (59.8% conversion vs. Grok's 15.5%). Gemini never cited or retrieved a single LinkedIn source in this dataset.

What's the difference between a source and a citation?

A source is anything the model's raw response shows it retrieved while assembling an answer, including pages it didn't end up using. A citation is a source that actually appears in the visible answer. ChatGPT, Google AI Mode, and Grok expose both; Perplexity and Copilot only ever expose the already-cited subset.

AI Search · Hotels

Is LinkedIn useful forAI visibility in hotel search?

Name: LinkedIn as a Hotel AI Citation Source 2026
Creator: Nicolas Sitter
Published: 2026-08-24
License: https://creativecommons.org/licenses/by/4.0/

TL;DR: Across 105,377 hotel-search captures with citation data, LinkedIn shows up in just 197 of them (0.19%) — but that number hides three very different stories. ChatGPT cites LinkedIn roughly 60% of the times it even sees it, almost always a hotel's own company page. Grok sees LinkedIn far more often (122 captures) but mostly as social posts it retrieves and then ignores. Perplexity, Copilot, and Gemini barely touch it — Gemini never once in this dataset. LinkedIn's occasional usefulness depends entirely on which model, which content type, and whether you're counting what the model saw or what it actually cited.

Nicolas Sitter

Published August 2026

0.19%

of all hotel captures have any LinkedIn citation

60%

of the time ChatGPT retrieves LinkedIn, it cites it

13%

of the time AI Mode retrieves LinkedIn, it cites it

LinkedIn citations from Gemini, ever, in this corpus

Read the Report

Summary 1. Sources vs. citations 2. It depends which engine 3. What kind of query Methodology FAQ

Executive Summary

Rare overall. Not rare everywhere. Not the same thing once it shows up.

We added a dedicated linkedin_count column to our AI hotel-search capture pipeline and backfilled it across every stored capture: 105,377 hotel-search answers with citation data, going back to December 2025. LinkedIn appears in 197 of them — a real signal, not zero, but thin enough that a single small sample would have missed most of the story (an earlier 400-row-per-engine check found essentially nothing; the full backfill found six times that in absolute count once every engine and every historical week was included).

The headline 0.19% number is the least interesting part. What matters is that it is not one phenomenon: it is a ChatGPT phenomenon (company pages, cited more often than not), a Grok phenomenon (social posts, retrieved constantly and cited rarely), and a non-event on three other engines. Anyone deciding whether a hotel's LinkedIn presence is worth maintaining for AI visibility needs the breakdown, not the average.

Section 1

“LinkedIn shows up” and “LinkedIn gets cited” are different claims

Two engines in this corpus (ChatGPT, Google AI Mode) expose a broader retrieved pool separately from the sources that actually make it into the visible answer. Grok exposes the same split, structurally. Perplexity and Copilot don't — everything their API surfaces to us is already labeled cited, so there's no way to tell what they saw and rejected. Treat their 100% conversion rate below as a measurement ceiling, not a behavior.

LinkedIn citation rows in citations_flat by platform, cited vs. retrieved-only. 332 rows total, 183 distinct URLs, all-time.

Engine	Retrieved, not cited	Cited	Conversion rate
ChatGPT	49	73	59.8%
Grok	147	27	15.5%
Google AI Mode	21	3	12.5%
Copilot	not exposed	9	100% (structural ceiling)
Perplexity	not exposed	3	100% (structural ceiling)
Gemini	0	0	—

ChatGPT is the outlier in the useful direction: once LinkedIn clears whatever retrieval threshold gets it into the pool, ChatGPT cites it roughly 3 times out of 5 — a much higher hit rate than any other source category we've measured on this corpus. Grok is the inverse: it pulls LinkedIn into its retrieved pool constantly (147 misses vs. 27 hits) but treats most of it as background noise, not answer material.

Section 2

It depends which engine — and which LinkedIn

LinkedIn is not one kind of page. Company pages, long-form “Pulse” articles, and ordinary social posts get pulled in at completely different rates depending on the engine.

LinkedIn citation rows by content type and engine. Company pages = a business's own LinkedIn profile; Pulse = long-form articles (often third-party listicles, not written by the hotel); social posts = ordinary feed posts, from hotel brand accounts or random users.

Engine	Company pages	Pulse articles	Social posts	Total rows
ChatGPT	96	25	1	122
Grok	60	33	79	174
Google AI Mode	5	13	5 (+1 personal profile)	24
Copilot	1	0	8	9
Perplexity	0	0	3	3

ChatGPT: company-page citer

96 of ChatGPT's 122 LinkedIn rows (79%) are company pages — a specific hotel's own LinkedIn profile. Example: a “luxury hotels in Zermatt” query cited linkedin.com/company/cervo-mountain-resort directly. ChatGPT treats LinkedIn mostly as a business directory, not a content source, and almost never cites a bare social post (1 out of 122).

Grok: social-feed miner

79 of Grok's 174 LinkedIn rows (45%) are ordinary posts — hotel brand announcements, guest reviews, even a stranger's opinion post on “Michelin vs. Forbes Travel Guide” that surfaced twice for unrelated hotel queries. Grok retrieves LinkedIn like a social platform (which tracks with how it treats Reddit and X/Twitter elsewhere in this corpus), but converts it to an actual citation only 1 time in 6.

Content quality is inconsistent even when a citation happens. One captured answer for “luxury hotels in Amsterdam with spa” cited a LinkedIn Pulse article literally titled “Where to Stay in New York City: 45 Top Boutique Hotels.” The same NYC listicle also showed up, unrejected, across several other unrelated destination queries. When LinkedIn does get cited, it is not always the LinkedIn page you'd want cited.

Section 3

What kind of query surfaces LinkedIn

Every LinkedIn citation we found traces back to a broad, destination-level discovery prompt — never a query naming a specific hotel. Representative examples pulled directly from the data:

Prompts that surfaced a LinkedIn citation or retrieval, sampled across engines.

Query	Engine	Cited?
luxury hotels in Zermatt	ChatGPT	Yes
best hotels in Amalfi Coast	ChatGPT	Yes
hotels near Matterhorn, Zermatt	Copilot	Yes
luxury hotels in Beverly Hills	Grok	Yes
design hotels in SoHo New York	ChatGPT	No (retrieved only)
best hotels in Maldives for business travelers	Google AI Mode	No (retrieved only)

An honest limit of this dataset: every prompt in the AI Hotel Landscape corpus is a destination-level discovery query — “best/luxury/affordable hotels in [city], for [traveler type] / with [amenity]”. There is no “is [specific hotel name] any good” query type in this corpus to compare against. So the honest finding is narrower than “LinkedIn favors discovery queries” — it's that within a corpus made entirely of discovery queries, LinkedIn shows up on exactly the query type this corpus tests, at the rates above, and we have no data on whether named-hotel lookups behave differently.

Methodology

Corpus. 105,377 AI hotel-search captures with non-null citation data, all-time (December 2025 onward), across six engines: ChatGPT, Perplexity, Copilot, Gemini, Google AI Mode, and Grok. Every prompt is a destination-level discovery query — “best/luxury/affordable hotels in [city]”, optionally modified by traveler type or amenity.

Column. A new linkedin_count column was added to the capture table (mirroring existing youtube_count/reddit_count/facebook_count columns), matching citations against the linkedin.com domain, and backfilled across every stored capture with citation data. 197 captures have a non-zero count.

Sources vs. citations detail. The per-engine retrieved-vs-cited split in Section 1 comes from the underlying flattened citation table, which additionally records a cited boolean per source row for engines whose raw payload distinguishes a retrieval pool from the final answer (ChatGPT, Google AI Mode, Grok). Perplexity and Copilot's APIs only ever expose the already-cited subset, so their conversion rate is reported as a structural ceiling, not a measured behavior.

Content-type classification. Each LinkedIn URL was classified by path pattern: /company/ (business page), /pulse/ (long-form article), /posts/ (social post), /in/ (personal profile), /jobs/ (job listing). 183 distinct URLs across 332 citation rows; a handful of URLs recur across multiple captures or multiple positions within the same capture.

Access. The column addition and backfill were reviewed before being applied; the backfill itself was a straightforward re-read of already-stored citation data (no re-scraping, no new captures).

FAQ

Marginally, and only through a company page, not a content strategy. LinkedIn citations are rare overall (0.19% of captures) but skew heavily toward a hotel's own company page rather than posts or articles — and ChatGPT specifically cites LinkedIn company pages at a high rate once they're retrieved (60% conversion). If a hotel already has an active LinkedIn company page, there's a small real chance of it surfacing in a ChatGPT answer. Posting content to LinkedIn in the hope of AI citation is a much weaker bet — most engines retrieve social posts far more than they cite them.

Every number here is from our own AI hotel-search captures

105,377 captures, all-time, all engines — CC-BY-4.0. Using this data? A citation or link back to nicolassitter.com is always appreciated.