FREE DOWNLOAD
Generic AI writing costs you reach on LinkedIn
An observational study of 287,120 posts, including a cohort of 78,281 published since January 2026. Every post is compared with the other posts of its own author.
- The four turns of phrase that cost the most reach, and the three that gain it
- Why French pays two to three times more than English
- Full tables by language and period, with confidence intervals
Get the study
The four findings
No email needed for these. The study holds the method, the tables by phrasing and the annexes.
- 1
A penalty that appeared in 2026
Before this year the link between AI phrasing and reach was close to zero in English and weak in French. On the cohort published since January 2026 it is clear in both languages. A filter appears to be at work, and it matches the policy LinkedIn made public in May 2026.
- 2
The effect concentrates at the top
On average it stays modest. But on the 5% of posts with the most AI-marked phrasing, the loss reaches -3.4% in English and -13.4% in French, measured against the author’s own usual reach.
- 3
Not every AI marker costs you
Four turns of phrase account for most of the loss. Three others, usually lumped in with them, actually gain reach: they are genuine engagement practices, not AI formatting.
- 4
French pays two to three times more
French-speaking audiences punish ready-made formulas harder, and those formulas are more common there: “It’s not X, it’s Y” sits on roughly one French post in six.
and the full tables in the study
How it was measured
Rolling corpus of 287,120 posts across 6,000 authors, English and French, text and image only. Reach is compared WITHIN each author, so audience size drops out of the equation: an author is only ever compared with themselves. The corpus split at 1 January 2026 lets each period be estimated on its own. This is observational: it establishes a correlation, not strict causation.
Questions
- What does this study actually measure?
- LinkedIn reduces the reach of generic phrasing, not of AI use as such. We cannot observe who used a model: we measure ten families of turns of phrase, and a human who writes “Here’s how” is counted too. On the cohort published since January 2026, the posts carrying the most of them lose reach in both languages measured, down to -13.4% on the most marked French posts.
- How much reach does it actually cost?
- Less than the headlines suggest, and more than nothing. On average the effect is small, because most posts are barely marked. On the 5% with the most AI-marked phrasing it reaches -13.4% in French. Removing those phrasings recovers a few points; it does not double your impressions.
- Which phrasings should I stop using?
- Four: the “Here's how” announcement, the “It's not X, it's Y” pivot, the dramatic “The result?” bridge and the “Stop X, start Y” instruction. Three others gain reach and should be kept: displayed candor, the closing question and the P.S.
More resources
Write in your own voice
MagicPost flags the phrasings this study measures, in your posts, before you publish them.
Try MagicPost free