AI and LinkedIn ·
Corporate prompt bias: why every brand has sounded the same since 2024
If you open LinkedIn and read the official posts of ten consultancies, five banks and a dozen scale-ups back to back, you will notice something strange: they all look written by the same hand. The same kind of headline, the same optimistic opening paragraph, the same antithesis in the second, the same list of three values in the close. That impression is not prejudice, it is a measurable phenomenon with a concrete cause.
The cause is called corporate prompt bias. It is what happens when hundreds of communications departments, with very similar briefs, hand those briefs to the same two or three generative models expecting "professional" output. The inevitable result is a collapse towards the average register of the corpus those models were trained on: pre-AI corporate LinkedIn itself. The snake eats its tail, and every brand has sounded the same since 2024.
By Sheena de PunkVoice · Edited by Mario Pérez
What corporate prompt bias exactly is
A generative model returns the statistically most probable answer to a prompt, given its training corpus. When a community manager asks for "a LinkedIn post in a professional, warm and aspirational tone to announce the launch of a new service", the model does not invent: it retrieves the average pattern it saw across thousands of corporate LinkedIn posts written between 2015 and 2023.
That average pattern exists and is very stable: open with a question or a context data point, a mission paragraph, three benefit bullets, an uplifting call to action to close, the occasional emoji. Any brief that lands near that description gets output close to that pattern, regardless of the industry, the country or the nominal tone requested.
The bias is not "corporate tone". It is the statistical convergence of every corporate tone towards a single centroid, which happens to be the one expressing the least individual judgement because it averages the judgement of thousands of earlier communicators. It is literally the sector's lowest common denominator.
Why the phenomenon explodes in 2024 and not before
The combination that sets it off is twofold. On one side, in 2023 generalist models reach a level of fluency sufficient to produce publishable corporate text with no touch-ups. On the other, office suites (Microsoft Copilot, Google Duet, Notion AI) integrate those models directly into the editor, with no copy-paste friction. Writing a post goes from a thirty minute task to a three minute one.
In 2024, that lower friction means the vast majority of medium and small communications departments adopt the workflow. There is no strategic decision: it is operational common sense. And because every department starts from similar briefs and uses the same two or three models, the output converges within weeks.
The result can be measured with a simple classifier trained on corporate posts from 2020 and applied to those from 2025: stylistic variance between brands falls by more than 40% in two years. In human terms: anyone reading attentively finds it increasingly hard to tell one brand from another beyond the logo.
The five symptoms of the bias in the corporate feed
These are the traits that appear combined in a very high percentage of affected corporate posts. When one appears alone it means nothing. When four or five appear at once, the post is signed by the centroid.
- Opening with a rhetorical question expecting no answer ("Did you know that...?", "What if we told you that...?") followed by a generic, unsourced data point.
- A prefabricated antithesis in the second paragraph ("It is not about X, but about Y", "Beyond X, we are talking about Y").
- A list of three values or pillars with perfect syntactic parallelism and no concrete example anchoring them.
- An aspirational close with a vague call to action ("Let us keep building the future together", "This is only the beginning").
- A total absence of nameable details: no specific client, no figure that is not round, no internal project, no person.
What LinkedIn is doing about it
The platform has a serious problem if its corporate feed becomes indistinguishable. Engagement on company posts has been falling since 2023 and in 2026 sits below 40% of its 2021 peak. The internal response combines three measures.
The first is a stylistic classifier penalising generic texture in any post, personal or company. A corporate post triggering all five symptoms above is shown to a fraction of the page's followers, even when technically there is nothing to object to. The second is prioritising employee posts above company page posts in the feed when both cover the same topic, on the reasoning that the individual voice adds texture the corporate one lost.
The third, quieter one, is an extra push for page posts that include concrete examples: a named client, a specific figure, a project with a name. That push is not documented in the official help pages, but it shows up consistently in large companies' dashboards.
How a brand leaves the centroid without abandoning AI
The tempting reaction is "let us go back to writing by hand", but that is unfeasible at the scale medium sized departments operate at. The realistic reaction is changing the starting prompt and the input material, not the tool.
A brief that produces non-centroid text has three traits an average brief lacks. It explicitly names the specific person or client the post is about (not "our clients" but "Marta, operations director at Distribuciones López"). It provides at least one specific, non-round figure only you know (not "we improved efficiency" but "we cut average response time from 42 minutes to 17"). And it states which opinion of your own the post defends, not which neutral message it communicates.
When those three traits enter the prompt, even a generalist model returns something noticeably more human, because the starting material already breaks the sector average. Adding a thirty second final review on top, to put a turn of phrase of your own into the hook and the close, shuts the door on the centroid.
The centroid is a ceiling, not a safe zone
For decades, the default corporate communications strategy was "do not stand out for the wrong reasons". With generative AI that strategy inadvertently produces "do not stand out for anything". The centroid stopped being a safe zone and became a ceiling of invisibility.
The brands recovering corporate reach on LinkedIn in 2026 are the ones accepting that a useful post looks more like a personal note than a press release. They name, they quantify, they take positions and they sign. And for that it makes no difference whether a person or a model wrote the draft: what matters is that the brief was not the same brief another four hundred companies are sending out this same week.