The short version: AI-written LinkedIn posts don’t get penalised by a secret algorithm, they get penalised by bored humans who scroll past them, and that scroll-past behaviour is exactly what trains the algorithm to bury them further. I tested this on my own account for a month and watched my reach drop from a normal 28,000-40,000 impressions down to under 4,500 on posts I’d let ChatGPT draft in full.
The post that made me stop trusting ChatGPT with my captions
Back in the spring I was slammed. Three client calls, a webinar to prep, and a LinkedIn content calendar that needed filling for the week. So I did what a lot of small business owners do now: I gave ChatGPT my bullet points and asked it to turn them into a post. It came back polished, on-topic, and grammatically flawless. I tidied the opening line, added an emoji, hit publish.
That post got 4,213 impressions and 11 comments. My usual range on a weekday post, written the way I talk, sits between 28,000 and 40,000 impressions with 60 to 150 comments. I assumed it was a fluke. Wrong time of day, wrong topic, algorithm having an off week. So I ran the same experiment nine more times over four weeks, alternating fully AI-drafted posts with my own rough, typo-ridden, opinion-heavy drafts.
The pattern held every single time. AI-drafted posts averaged 5,800 impressions. My own drafts averaged 31,000. Same posting times, same hashtags, same topics some weeks. The only variable was who wrote the sentences.
What’s happening isn’t a detector, it’s a mirror
Here’s the uncomfortable bit most people writing about “AI and the algorithm” skip over because it’s less satisfying than blaming a mysterious filter: LinkedIn (and Facebook, and Instagram) almost certainly doesn’t run your text through some AI-detection tool that flags and suppresses it on sight. What’s happening is much simpler and much harder to fix. People scroll past content that sounds like everyone else’s content, and the platform reads that scroll-past as “nobody wanted this,” so it shows the post to fewer people, which produces fewer comments, which produces less reach, and so on down the drain.
AI writing has a flavour. It’s not bad writing, it’s often better structured than what most of us produce off the cuff, but it’s smoothed out in a way that human readers clock in about half a second even if they couldn’t tell you why. No specific numbers. No named client. No sentence that sounds like it came out of an actual argument you had with someone at a networking event. It reads like it was written to be correct rather than written to be believed.
That’s the real problem, and it’s not one an algorithm update fixes. It’s a trust problem, and trust problems don’t get solved by prompting better.
The five tells that give an AI post away
- Rule-of-three lists inside a sentence: “faster, smarter, and more efficient” turns up constantly because it’s a pattern the model has seen a million times in training data.
- Zero named specifics: no company name, no number, no date, no place. Real experience always has a receipt attached.
- A tidy three-part arc: problem, insight, resolution, wrapped up with a bow in 150 words, with none of the mess a real story has.
- Emoji used as punctuation rather than emphasis, one after nearly every line.
- A closing question that isn’t really a question, like “What’s your take?” tacked on because the model was told to “drive engagement.”
I can spot my own AI drafts now within the first line, and so, it turns out, can the people scrolling past them.
Brands that never have this problem
The brands that keep winning on social don’t have a secret algorithm hack, they have a voice you’d recognise blind. Duolingo built an entire following on being unhinged and specific rather than polished, and if you want to see how far a distinct voice can carry a brand, the Duolingo marketing strategy is worth a proper read. GoPro does the same thing with visuals instead of copy, letting real customers’ footage carry the brand rather than a marketing department’s script, which I break down in the GoPro marketing strategy piece. Neither of those brands would let ChatGPT write their captions untouched, because the whole value is in sounding like nobody else.
How I fixed my own reach (the actual process, not a vague suggestion)
I didn’t stop using AI. I stopped using it as a writer and started using it as a research assistant. Here’s the exact process I use now, five steps, takes about 20 minutes per post:
- Step 1: I write the bones myself, badly, in under five minutes. Just the point I want to make, in my own words, typos and all.
- Step 2: I add one specific detail that couldn’t have come from anywhere else. A client’s name (with permission), an exact number from a call I had that week, a date, a place I was standing when the idea hit me.
- Step 3: I ask ChatGPT or Claude only to check structure and flag anything confusing, never to rewrite sentences. I explicitly tell it not to add adjectives.
- Step 4: I read it out loud. If it doesn’t sound like something I’d say to a friend over coffee, I cut it.
- Step 5: I post it and answer every single comment within the first hour myself, because that early engagement window is what tells the platform this post is worth showing to more people.
My reach came back within about two weeks of switching to this process. Not overnight, because the algorithm needed a few posts to relearn that my account produces content worth surfacing.
Check what people engage with, not what you assume they do
Most small business owners guess at what’s working instead of checking. If you run any content on your own website alongside your social posts, watching where people pause, scroll, and click matters more than any gut feeling about tone. That’s the same logic behind heatmap tools, and if you’ve never looked at how a brand like Hotjar built its whole product around exactly this behaviour data, the Hotjar marketing strategy breakdown is a good primer on why watching real behaviour beats guessing every time.
It’s also worth remembering that text posts aren’t the only format fighting for attention. Interactive content consistently pulls people in for longer than a paragraph ever will, which is part of why I’ve written before about using calculators to engage and convert more visitors. If your LinkedIn reach is struggling, sometimes the answer isn’t fixing the writing, it’s changing the format entirely for a week and seeing what happens.
Where AI earns its place
None of this means throw the tools away. AI is brilliant at the unglamorous parts of content that nobody enjoys and nobody’s judging you for outsourcing: repurposing a long post into three shorter ones, drafting subject lines you’ll rewrite anyway, checking a post for grammar before it goes out, or summarising a client call into bullet points you can turn into a story later. I automate a fair chunk of my repurposing with simple trigger-based tools, and if you want to see how far basic if-this-then-that automation can stretch a small team’s output, the IFTTT marketing strategy writeup covers the same principle for a much bigger audience. Automate the plumbing. Never automate the personality.
There’s also a calmer way to think about all this that most “AI will replace your marketing” articles skip: the goal was never to sound impressive, it was always to sound like a person your ideal client would trust with money. Some of the steadiest, most successful brand voices out there aren’t loud at all, they’re quiet and consistent, which is worth studying if you assume louder always wins. The Calm marketing strategy is a decent example of a brand that built trust through restraint rather than noise, and restraint is exactly what most AI-drafted posts are missing, ironically, because they’re trying too hard to sound helpful.
If you’ve looked at your own reach numbers over the last six months and suspect AI has quietly crept into more of your content than you realised, that’s usually a fifteen-minute audit rather than a full rebrand, and it’s the sort of thing an outside pair of eyes catches faster than you will on your own account. That’s the kind of practical fix an AI consultant for small business gets asked to sort out more than almost anything else right now.
The uncomfortable part nobody wants to hear
Here’s the bit that stings: if your AI-drafted posts are underperforming, it’s not really a technology problem, it’s proof that your own writing was carrying the account. The tool didn’t fail you, it just couldn’t fake the thing your audience was there for, which was you. That’s a harder pill to swallow than “the algorithm is broken,” but it’s the true one, and it’s also the good news, because it means the fix is entirely in your control. You don’t need a better prompt. You need to write the first draft yourself again.
Frequently asked questions
Does LinkedIn detect and penalise AI-written posts?
There’s no confirmed detector that flags and buries AI text on sight. What happens instead is behavioural: readers scroll past content that feels generic, and that lower engagement signal is what causes the platform to show the post to fewer people. The effect looks like a penalty, but it’s caused by human boredom, not a hidden filter.
Can I still use ChatGPT for my LinkedIn content at all?
Yes, use it for structure checks, grammar, and repurposing a long post into shorter formats. Avoid letting it write your first draft or invent your opening line, since that’s where the generic tone creeps in and where readers notice fastest.
How long does it take to recover reach after posting too much AI content?
In my own case, switching back to a human-first draft process brought reach back within about two weeks and roughly six to eight posts. The recovery isn’t instant because the platform needs fresh engagement signals to relearn that your account produces content worth surfacing.
What’s the single biggest giveaway that a LinkedIn post was written by AI?
Missing specifics. Real posts have a name, a number, a date, or a place attached to them. AI drafts tend to stay general because the model has no actual experience to pull a detail from, and readers feel that absence even when they can’t name what’s missing.
Related reading: Why Your Analytics Are Hiding How Many Customers AI Is Sending You and Mailerlite Marketing Strategy: How They Built a Brand That Wins.
