No. 031
LinkedIn's VP quotes 93%. The report he cites prints 92%.
A LinkedIn marketing post cites 93% of the 24 most-cited articles. The Meltwater page it links prints 92%, which is 22 of 24. The studies under it disagree too.
93%vs92%
What exactly does the post claim?
The post is “Why Your Brand Needs a Point of View AI Can’t Replicate”, published 29 September 2026 by Davang Shah, VP, Marketing at LinkedIn. Its central lines are:
Content structure is important, but it’s also becoming table stakes.
Structure is necessary. It just isn’t sufficient.
To get there it cites six figures. Here is each one against the page it points at.
| Claim in the post | Source it links | What that page says | Status |
|---|---|---|---|
| ”Every single one uses bullet lists or numbered items.” | Meltwater blog | bullets 100% | Matches |
| 93% “use clear H2 or H3 section headings” | Meltwater blog | clear headings 92% | Differs by one point |
| 75% “name specific companies or tools” | Meltwater blog | named entities 75% | Matches |
| 67% “include hard numbers and data” | Meltwater blog | quantitative data 67% | Matches |
| ”Meltwater’s research (which analyzed 9.5M citations)“ | Meltwater blog | 9.5 million AI citations across six AI platforms | Matches |
| ”Semrush’s research (which analyzed 300K citations)“ | Semrush study | 325,000 unique prompts and 89,000 unique LinkedIn URLs. No total citation count anywhere on the page | Not found |
| ”Forrester reports that 94% of buyers use AI during their buying process.” | Forrester press release | The release’s only 94% is a different statistic | See below |
Four of the seven match exactly. That is worth saying plainly, because the useful version of this exercise is not a gotcha count. It is that two of the three that do not match are the two doing the most argumentative work.
The “300K citations” figure is the clearer of the two. The Semrush study is titled “We Analyzed 89K LinkedIn URLs Cited in AI Search”. It reports 325,000 unique prompts run across ChatGPT Search, Google AI Mode and Perplexity in January and February 2026, and 89,000 unique LinkedIn URLs cited in the results. It does not report a citation total. “300K” appears on neither the study nor the Semrush-authored playbook hosted on LinkedIn’s own blog.
The cause of either difference is not established. A typo and an older draft look identical from outside, and this page notes it rather than explaining it.
What can the 24-article audit actually show?
Less than it looks, and the limit is structural rather than sloppy.
The Meltwater page describes the 24 as “the top 24 most-cited LinkedIn articles”. It states no selection method: not how the ranking was produced, not over what window, not out of how many candidates. The phrase itself appears in the page’s image alt text rather than its prose.
More importantly there is no comparison group. We know 100% of the top 24 use bullets. We do not know what share of uncited LinkedIn articles use bullets, and without that the figure is a base rate rather than an effect. If 100% of cited articles and 97% of uncited articles use bullets, the number is true and tells you nothing about why the cited ones were cited.
The same page also undercuts the simple reading of its own format figures in two ways. First, its content-type prevalence numbers are 54% for “Best X” listicles, 50% for side-by-side comparisons, 33% for “how to choose” guides, 17% for educational explainers and 8% for thought leadership with data. Those sum to 162%, so they are prevalence rates across overlapping categories, not shares of a whole. Second, the page says in its own prose that “pure opinion is less likely to be cited on its own” and ranks thought leadership with data last at 8%. A post arguing that point of view beats structure is citing an audit whose opinion category came last.
The page’s methodology sentences do not agree with each other either. It says “millions of prompts” in one place and “thousands of test prompts” in another, unreconciled. The full report is behind a form, so the selection method cannot be checked from outside.
How much of AI’s LinkedIn citations are articles?
This is where the numbers stop being a rounding story.
| Measurement | Platforms covered | Window | Articles as a share of LinkedIn citations |
|---|---|---|---|
| Profound | ChatGPT only | 15 Nov 2025 to 15 Feb 2026 | 6.0% to 8.9% |
| Meltwater | six AI platforms | four-week project | 12% |
| Semrush | ChatGPT Search, Google AI Mode, Perplexity | Jan to Feb 2026 | 50% to 66% |
| LinkedIn’s own guide, 11 Aug 2026 | not stated | not stated | ”roughly 60%”, no source given |
Taking the top of Semrush’s range against Profound’s is a factor of 7.4. Taking the bottom of it against Meltwater’s is 4.2.
These can all be honestly produced, because they are not measuring the same thing. Profound’s and Meltwater’s shares are of LinkedIn citations; the platform sets differ; the windows differ; and the unit differs, since Profound separates profiles as a third category running from 33.9% down to 14.5% over its window while Meltwater’s split is text posts 72%, articles 12%, video 11%. What none of them publishes is enough detail to reconcile the others.
The same two studies disagree on originality by a similar margin. Meltwater reports 72% of cited content original and 28% reshared. Semrush reports approximately 95% original and 5% reshares.
And the denominators behind “LinkedIn is big in AI search” are not comparable at all. Semrush has LinkedIn appearing in 11% of AI responses on average, ranked second, with 14.3% on ChatGPT Search, 13.5% on Google AI Mode and 5.3% on Perplexity. Meltwater has LinkedIn at 0.53% of all citations, also ranked second, behind YouTube at 1.52%. Meltwater defines its term: “Citation share is the percentage of total citations that reference LinkedIn.” One is a share of responses, the other a share of citations. They differ by a factor of about 21 and both are correct.
Which 94%?
The post attributes “94% of buyers use AI during their buying process” to Forrester and links the Forrester press release. On that release the only 94% is a different statistic, about buyers in groups of six or more reporting clear benefits. The figure as stated does appear on Forrester’s own blog of 21 January 2026: “Nearly all business buyers (94%) report using AI during their buying process”, from a survey of nearly 18,000 global business buyers.
The same 94% is also credited elsewhere to a different company entirely. LinkedIn’s own August guide says “According to 6sense, 94% of B2B buyers now use generative AI for research.” Meltwater’s press release of 12 May 2026 likewise attributes “94% of B2B buyers use LLMs during their buying process” to 6sense. We could not open 6sense’s own page from here, so this piece does not quote it; what is checkable is that two pages credit the figure to 6sense while the live LinkedIn post credits it to Forrester and links a Forrester page where 94% means something else.
So the same headline number is attributed to two different research houses on three different pages, one of which is LinkedIn’s own. The attribution is recoverable in every direction. That is the problem: a figure that can be sourced to anything is not evidence of much.
One more link in the chain is worth recording. LinkedIn’s August guide contains the sentence “Meltwater’s 2026 analysis found that 75% of LinkedIn citations came from individual member profiles”, and the anchor text “Meltwater’s 2026 analysis” links to the Semrush study.
Who measured, and who sells what
Every figure above comes from a company with something to sell, and most of them say so on the page. That is not an accusation, it is a disclosure list.
Semrush states that it “collaborated with LinkedIn to obtain deeper insights” and notes that internal LinkedIn data informs one of its findings. Meltwater’s release quotes the same LinkedIn VP who wrote the post citing Meltwater, and its full report sits behind a lead form. Profound and Scrunch both sell AI-visibility monitoring. LinkedIn sells advertising, and as PPC Land observes, LinkedIn is itself among the most frequently cited domains in AI search, so its platform benefits from marketers treating LinkedIn as the place to publish.
None of that makes a number false. It does mean no independent measurement of this question exists in public.
Is the thesis right, then?
Unproven, in both directions, and the honest answer is more interesting than a verdict.
The one study here that tries to isolate an effect rather than describe a sample is Scrunch’s. It looks at 12,000 LinkedIn post observations from ChatGPT, narrowing to about 4,000 posts with controls for author audience size and post age, over 15 January to 15 April 2026. Posts with specific technical details were 77% more likely to be cited; named entities, 33% more likely. Unicode formatting was 58% less likely and link-in-comments 31% less likely. Its summary line is that “Several of the most effective LinkedIn engagement tactics have no measurable effect on AI citations.”
That supports the spirit of the LinkedIn post: substance beats decoration. It does not support the specific claim, because neither “heading” nor “bullet” appears in the body of that article. Headings and bullets were not among the tested variables. So “structure is necessary” has no test of necessity behind it in any of the four studies cited, and the one study with controls covers one platform and posts rather than articles.
How to read any citation-share number
Six questions, in the order they usually break:
- What is the denominator: responses, citations, URLs, prompts, or posts?
- Which platforms, and is one of them doing all the work?
- What window, and how long ago did it close?
- What is the unit: an article, a feed post, a profile page?
- Is there a comparison group of uncited items, or only a description of winners?
- Who paid for it, and what do they sell?
Questions one and five are the ones that matter most here. A share of responses and a share of citations differ by a factor of twenty on the same platform in the same year. A description of 24 winners with no losers in the sample cannot tell you what caused the winning.
Method and corrections
Every page named above was opened on 12 October 2026 and the quoted strings were matched against the text served. Meltwater’s full report is gated, so its selection method could not be checked. The “top 24” phrasing appears in that page’s image alt text. The Scrunch article prints its date as 05.06.2026 without stating whether the day or the month comes first, so no date is given for it here.
The arithmetic in this piece is ours: 22 of 24 is 91.67%, 23 of 24 is 95.83%, 93% of 24 is 22.32, and the five Meltwater content-type figures sum to 162%.
Credit for spotting the impossible 93% goes to PPC Land. Live pages can be edited after publication. If any figure quoted here changes, or if we have one wrong, we will correct it and say what changed.