Post about AI on LinkedIn only if you have something specific to report from your own use of it, and be wary of using AI to write the posts themselves.
Two separate questions hide in the title. The first is whether AI is a good topic for a B2B firm's posts. The second is whether AI-written posts perform. The evidence on the second is clearer than on the first, and neither supports a blanket rule.
This guide covers what the available data shows, where it falls short, and an eight-post test you can run to see what works for your audience. It forms part of our guide to hiring a social media marketing agency in the UK.
What the data says about AI-written posts
The best study we found is from Originality.ai, a company that sells AI detection. It analysed 3,368 long-form LinkedIn posts from 99 influential profiles across 11 industries, published in 2025, and classed 53.7% as likely AI-written. PPC Land's report on the study sets out the figures.
Average interactions per post varied by category. In marketing and branding, posts classed as human-written averaged 771 interactions against 207 for those classed as AI-written. In healthcare and medicine the figures were 208 human against 116 AI. In tech and AI, AI-classed posts averaged 302 against 282, and in finance and business 127 against 118.
xychart-beta
title "Marketing and branding posts: average interactions (Originality.ai, 3,368 LinkedIn posts, 2025)"
x-axis ["Classed as AI-written", "Classed as human-written"]
y-axis "Average interactions per post" 0 --> 900
bar [207, 771]
In leadership and inspiration, AI-classed posts did better: 2,635 against 1,504.
How far to trust that study
Treat it as a pointer, not proof. Detection scores are probabilistic, and Originality.ai's claimed accuracy has not been independently verified. The comparisons are averages across profiles of very different sizes, not controlled tests.
The article reporting the study also contains inconsistencies. The percentage gaps it states do not match the raw averages it lists for several industries, and it gives two different minimum word counts. We have used the raw averages and left out the percentages.
What survives is a modest finding. In marketing and similar fields, posts that read as machine-written tended to get less response. In a few other categories, there was little difference.
Why AI-written posts underperform in B2B
We do not have proof of the cause, but the likely reason is simple. A post written by a model from a prompt contains nothing the reader could not get elsewhere. It has no first-hand detail.
B2B buyers read for signs that you know the work. Edelman and LinkedIn's 2024 research found that decision-makers trust thought leadership more than marketing materials, as Ragan's summary reports. Posts that sound like everyone else do not signal expertise.
More than half the long posts in the study were classed as AI-written, which means readers see a lot of it. A post that sounds like the rest of the feed gets scrolled past. Our guide to why most agency social content is invisible goes further into this.
If you hire an agency, ask how it uses AI. Our guide on how to tell if your SEO agency is using AI to write your content has the questions.
When posting about AI makes sense
Post about AI when you have used it on a real task and can say what happened. A short account of what you tried, what it saved, what went wrong and what you checked is specific, which is what most posts lack.
A law firm, an accountancy firm or a clinic can say which checks it puts around AI output. That is more useful to a prospective client than a general opinion on the future of the profession.
We found no study showing that posts about AI perform better or worse for firms outside the technology sector. The only category we can see in the Originality.ai data is tech and AI, where interactions were modest on both sides. So the case for AI as a topic rests on your having something to say, not on a proven boost.
When posting about AI does not make sense
Do not post about AI because the topic is in the news. A general take on "what AI means for our industry" is the most common post, and the easiest to ignore.
Do not post about AI if you have nothing to report. A reader who clicks through to your profile and finds no sign you have used it will draw their own conclusion.
Do not claim results you cannot show. If you say AI cut your admin time in half, someone will ask how you measured it.
Using AI safely as a drafting aid
There is a middle path: use AI to help with structure, wording or editing, but start from your own material. The post should contain your own facts, your own examples and your own opinion.
A practical rule is that every post needs one detail a model could not have supplied: a figure from your records, a conversation you had, a mistake you made. If it does not, rewrite it.
Also check client confidentiality and data protection before pasting anything into an AI tool. Do not paste client names, case details or personal data into a public tool. Check where your tool stores data and whether it trains on what you enter.
Accountants, consultancies and other professional services firms hold a lot of confidential information. Agree a written rule on what may go into an AI tool before anyone uses one.
Three shapes of AI post that carry information
If you decide to post about AI, three shapes tend to carry real information.
The first is the task report. Pick one job, such as summarising call notes or drafting a first reply to a routine query, and say how long it took before, how long it takes now, and what a person still checks. Use your own figures and say how many times you ran it.
The second is the failure report. Describe one time the tool got something wrong, how you caught it and what you changed. Few firms publish these, and they tell a reader that you check.
The third is the policy post. Say plainly what your firm will and will not put into AI tools, and why. Clients are asking these questions, and a clear answer is a sales asset.
Each shape has a number, an example and a decision. None of them depends on a news hook, so none goes out of date by Friday.
Should you tell readers when AI helped?
If AI did most of the drafting, say so when it matters to the reader. A client reading advice on a regulated matter will want to know a person checked it.
For a short post where you used a tool to tidy the wording of your own thoughts, there is little to disclose. The line is whether the substance came from you.
Whatever you decide, be consistent. A firm that says it never uses AI and then publishes obviously machine-written posts loses more trust than one that says plainly how it works.
Comments and messages are a separate matter
The advice above is about posts. For comments and direct messages, the stakes are higher because they are one-to-one. A comment generated by a tool and pasted under someone's post is easy to spot and can irritate the author.
LinkedIn also restricts automated activity, which we cover in the 15-minute comment routine. Write your own comments, even if they are short.
Sector notes
Healthcare firms face the sharpest version of this question. A dental practice or clinic posting health information needs a person with the right qualification to check every claim, whether a model or a person wrote the draft.
Originality.ai's figures show human-classed healthcare posts averaging 208 interactions against 116 for AI-classed ones. That is a modest gap on a small base, so we would not build a policy on it, but it does fit the instinct that readers want a person behind health content.
An eight-post test you can run
If you want evidence for your own audience, run a simple test over four weeks. Post eight times: four posts about a specific use of AI in your work, and four on other topics of similar quality and length.
Write all eight yourself, from your own material. Post at the same times of day, and do not boost any of them. Record impressions, comments, profile views and any conversation that mentions a post.
With so few posts, you will only see a large difference. If the AI posts get twice the comments, that is worth following up. If the numbers are close, you have no evidence either way and should choose on what you have to say.
What to measure
Look at comments from people in your target market. A post that attracts 12 comments from other marketers and consultants has not reached buyers.
Then look at enquiries. Ask each new enquiry whether they have seen any of your posts and which. Our 90-day LinkedIn plan shows how to set up that tracking.
Next steps
Growthcart is a digital marketing agency in Northampton. If you would like help working out what your firm has to say about AI, and how to say it without sounding like everyone else, get in touch.
