AI on LinkedIn Without Losing Reach: How to Create Content B2B Audiences Trust
LinkedIn has not declared war on artificial intelligence. It has declared war on content that lacks a recognisable author, experience, or point of view.
That is an important distinction. The problem is not using AI for research, organising material, or proofreading. It begins when the tool takes over the entire process—from idea to publication—and the expert merely clicks “generate.”
The risk is no longer theoretical. LinkedIn reported that it detects hundreds of thousands of attempts to automatically post comments every day, and that it blocked billions of automated actions within a few months. More than one million users have also used the option to report content as “AI slop.” According to the platform, the visibility of content classified this way has fallen by 40%. (news.linkedin.com)
For B2B teams, this changes the rules of the game. Publishing more competent posts is no longer an advantage. The advantage is content that a competitor could not publish under their own name.
LinkedIn Does Not Judge Style Alone. It Judges the Value of the Perspective
The riskiest interpretation of the current changes is: “We need to hide the fact that we use AI more effectively.”
That leads nowhere.
You can remove telltale phrases, limit the number of emojis, and avoid formulaic structures. But if the text still contains no original observation, project-specific detail, or decision the author can defend, it will remain generic.
LinkedIn defines “AI slop” more broadly than content generated by a model. It includes content that is:
low-effort,
repetitive or derivative,
devoid of a clear point of view,
designed primarily to capture attention,
professional-sounding but adding nothing to the conversation.
The platform also states that content appearing to be AI-generated and lacking an original perspective is less likely to be distributed beyond the author’s direct network. In initial tests, its systems reportedly identified generic content correctly in 94% of cases. (news.linkedin.com)
This changes the role of company content. A post should not merely “cover a topic.” It should show that the author:
has worked on a similar problem,
made a specific decision,
noticed a non-obvious pattern,
can identify an exception to a popular rule,
has a position that cannot be generated from the first few search results.
Expertise must be visible in the content itself, not merely appended to the author’s job title.
Audience behaviour data confirms this as well. In a study covering more than 134,000 posts, content that sounded AI-written generated, on average, 21% less engagement than other posts from the same author in the same year. The study’s authors do not claim this is a direct algorithmic penalty. Rather, they suggest that readers respond less favourably to text that sounds generated. (ligosocial.com)
The Line Is Between Assistance and Autopilot
LinkedIn itself is developing AI-powered features. AI can help users refine their profiles, create message drafts, analyse content, or process video materials. The platform is not rejecting the technology. It wants to limit its use where automation replaces people and produces mass, superficial engagement.
In practice, there are three models for using AI.
AI as an Assistant
This is the safest model. The tool supports the process but does not take over authorship.
It can:
organise meeting notes,
suggest questions for an expert,
identify gaps in an argument,
shorten overly long sentences,
correct grammar,
prepare several structural options,
flag passages that sound too general.
The human remains responsible for the thesis, examples, assessment of the situation, and final version of the text.
AI as a Ghostwriter
In this model, the tool receives a topic, a few sentences of context, and an instruction to write an “engaging LinkedIn post.”
The result often looks polished, but it follows a predictable pattern:
a broad, dramatic hook,
a general observation,
a list of tips,
an artificial contrast,
a banal conclusion,
a question designed to trigger comments.
The text may contain no errors, but it contains no author either. This is precisely the model that increasingly threatens reach and reputation.
AI as a Publishing Bot
The furthest-reaching model is one in which tools independently generate, publish, and comment on content on a user’s behalf.
LinkedIn explicitly considers automated comments published through extensions, scripts, and external tools to be prohibited. The platform may limit their visibility, remove them from the most relevant comments section, and restrict account functionality. The same risk applies to engagement pods and coordinated, artificial amplification of posts. (news.linkedin.com)
LinkedIn’s policy for solutions using Marketing API data also requires users to remain involved. If content is to be published on their behalf, they should be able to edit it beforehand. The policy also highlights the need to limit the automated publication of AI-generated content without user involvement. (learn.microsoft.com)
The simplest rule is this: automate material preparation, but do not automate accountability for publication.
A Workflow to Implement: AI at the Beginning and End, the Expert in the Middle
An effective process does not mean abandoning AI. It means assigning it the right role.
Research Stage: Do Not Ask AI What to Publish
Instead of the prompt:
Write a post about the most common mistakes in digital transformation.
ask the tool to prepare the groundwork for a conversation with an expert:
What questions will help uncover non-obvious lessons from the project?
Which parts of this thesis require data or examples?
What counterarguments might a COO raise?
Where might the audience consider the topic too general?
What information is needed from the author to make the text credible?
AI does not create the position at this stage. It helps identify where the author’s perspective needs to appear.
Knowledge-Gathering Stage: Collect Material the Model Does Not Know
An expert’s most valuable contribution is usually created not while writing, but during the conversation.
Twenty or thirty minutes of a well-run interview are enough to uncover:
a decision made under pressure,
a fragment of a customer conversation,
a false assumption from the beginning of the project,
a metric that changed the direction of action,
an outcome that contradicted expectations,
an opinion that part of the market will disagree with,
an example of a situation in which a popular practice did not work.
Instead of asking an expert, “What do you think about AI in marketing?”, ask:
When did you last reject an AI-generated recommendation?
What did the tool miss?
What data or experience allowed you to notice it?
What would have happened if the team had applied that recommendation uncritically?
What rule do you follow after that experience?
The answers to these questions are what create content with ownership.
Drafting Stage: The Human Builds the Argument
The first version does not have to be polished. It should, however, contain three elements:
A thesis — what the author genuinely believes.
Evidence — the experience, example, or observation supporting that opinion.
A consequence — what the audience should do differently.
An example of a generic version:
Companies should approach marketing automation with caution. Technology can increase efficiency, but it is important to maintain a human approach.
Any company could publish a passage like this.
A version with expertise added:
In one project, we automated lead qualification before the sales team had aligned on the definition of a good opportunity. The system worked according to its configuration, but it directed sales reps’ attention to the wrong accounts. The problem was not the technology. We automated a decision we had not previously been able to describe clearly.
The second version works because it contains a situation, tension, and a conclusion rooted in practice. AI can help shorten it. It should not, however, invent experience the author has never had.
Editing Stage: AI Improves the Text, Not the Personality
Once a substantive draft is ready, AI can be used to:
reduce repetition,
identify unclear wording,
improve the order of arguments,
break up long paragraphs,
proofread the language,
prepare several shorter hook variations.
It is worth using prompts that limit the scope of intervention:
Improve the readability of the text, but do not add new arguments, examples, metaphors, or conclusions. Preserve my vocabulary and position. Flag every change that could alter the meaning.
This approach aligns with the direction LinkedIn has taken. The platform is replacing its broad post “enhancement” feature with a tool focused on proofreading and clarity, without rewriting the author’s voice. (news.linkedin.com)
Publishing Stage: The Human Approves and Responds
Before publishing, the author should read the entire piece and answer three questions:
Would I genuinely put my name to every sentence?
Could I expand on this argument in a conversation with a customer?
Does the text contain anything AI could not have known without my involvement?
If the answer to the final question is “no,” the content still needs work.
The same principle applies to comments. AI can help structure a response, but the comment should come from actually reading the post. It should also add a new perspective rather than repeat the author’s main point in different words.
The Anti-Slop Test Before Every Publication
B2B editorial teams need a more demanding standard than linguistic correctness. Before approving content, it is worth running five tests.
The Logo-Swap Test
Remove the company, product, and author names.
Could a competitor publish the same text without changing any of the remaining sentences?
If so, the content is too general.
The Evidence Test
Highlight every sentence containing a recommendation.
Then check whether the recommendation is backed by:
an example,
an observation,
a number,
a causal mechanism,
clearly described experience.
If the text offers five tips and no evidence, it resembles an AI-generated summary of the topic rather than an expert’s contribution.
The Conversation Test
Read the post aloud.
Would the author say the same thing in a meeting with a customer? Would they use the exact same phrasing? Would they really end the conversation by asking, “What has your experience been?”
If the text sounds like a speech that no one would ever deliver aloud, it needs simplifying.
The Cost-of-an-Opinion Test
Every recognisable point of view costs something. It may require challenging an industry practice, admitting a mistake, or adding a caveat that makes the message less simple.
Ask:
What does the author disagree with?
When does their recommendation not work?
What would they advise a customer against, even if it meant a smaller project?
What trade-off must be accepted?
If the opinion carries no cost, it is probably a safe cliché.
The Expert Reaction Test
Send the completed content to the author and ask them not to approve it, but to challenge it:
Which sentence is oversimplified?
Where is context missing?
What would a sceptical customer say?
Which passage cannot be defended?
What sounds like marketing language rather than your opinion?
This is a far more effective process than asking, “Can the post go live?”.
How to Scale Content Without Scaling Mediocrity
Marketing teams use AI because they need to create more content with limited resources. The answer should not be a return to an entirely manual process. The key is to separate tasks that can be scaled from those that require human judgement.
You can automate:
conversation transcription,
note categorisation,
identifying recurring themes,
preparing question lists,
creating structural variations,
proofreading,
adapting approved content into other formats.
You should not automate:
choosing a position,
interpreting experiences,
creating fictional customer stories,
publishing without approval,
commenting on behalf of an expert,
artificially amplifying engagement,
generating fake discussions between accounts.
It is also worth changing how effectiveness is measured. The number of published posts alone does not reveal whether the system is building expert authority.
A better set of metrics includes:
the percentage of posts containing first-hand experience,
the number of substantive comments rather than all reactions,
the number of sales conversations that reference a specific publication,
saves and shares of content,
expert participation in the content creation process,
the number of recurring themes associated with a specific author.
What should be scaled is the system for extracting knowledge, not the production of more versions of the same post.
Over recent years, AI has reduced the cost of producing text. As a result, text itself has ceased to be a scarce resource. What remains scarce is experience, the courage to interpret it, and accountability for the words spoken.
That is good news for companies that genuinely have expertise. They do not need to compete with bots on publishing volume. They need to build a process that turns experts’ everyday decisions into content that cannot be copied.
AI can accelerate that process. But it cannot take the place of the person the audience needs to trust.
