AI Ghostwriters on LinkedIn: 54% of Posts Are Machine-Written

The LinkedIn post opens with a confession: "Three years ago, I was one bad week away from walking away from everything." Five one-sentence paragraphs follow, then a bullet list of hard-won lessons, then the obligatory question to the community: "How do you all deal with this?" The post is credited to the CEO of a mid-sized industrial manufacturing company. It was actually written by a language model that has never met the man — fed a single keyword pulled from a two-minute voice memo.

For B2B marketing leaders, this isn't a fringe curiosity. It's a line item in the budget. They're the ones approving spend on ghostwriters and AI tools to position executives and leadership as authentic voices in the market. The problem: many no longer know whether the output still reads as credible — or whether their audience has already figured out that a prompt is talking, not a person. And whoever signs off on the vendor invoices is the one holding the bag when the CEO's "personal brand" gets exposed as fabricated.

This article pulls back the curtain on how the ghostwriting industry behind LinkedIn actually operates, why the platform itself makes the problem worse, how to spot machine-generated text, and how marketing leaders can adjust their content strategy before followers or competitors start asking whether any of it is real.

Scroll through LinkedIn for ten minutes in 2026, and you'll hit a strange sense of déjà vu: a sales director in Chicago, a SaaS founder in Austin, and a management consultant in Boston are telling completely different stories — in the exact same format. A hook line designed to spark curiosity. Three to five one-sentence paragraphs. A bullet list of "key takeaways." And, to close things out, the obligatory reflection question aimed at the community. The content changes. The template never does.

This isn't just a gut feeling. Detection firm Originality.ai analyzed nearly 9,000 English-language LinkedIn posts with more than 100 words back in 2024 — and the results were striking.

54% of the long-form posts examined were flagged as likely AI-generated — a 189% jump in the months following ChatGPT's launch compared to prior levels (Source: Originality.ai, 2024). Since then, the models have gotten dramatically better and the barrier to using them has dropped even further. There's no plausible reason to think that percentage has gone down. If anything, it's climbed higher.

Here's what stands out most: the viral posts — the ones racking up tens of thousands of reactions — are exactly where these machine-made signatures show up most often. The feed isn't rewarding what's distinctive. It's rewarding what's repeatable. And this isn't an English-language phenomenon confined to the U.S. market. The same templates now show up just as reliably in German-language B2B feeds, running just a few months behind the American market.

For the average reader, one critical line gets blurred in the process: was this post written by a human and then polished by AI? Or did the person whose name is on it see the text for the first time only after it went live? From the outside, there's no way to tell the difference — and that ambiguity is exactly what's fueling a growing distrust of anything on the platform marketed as "personal experience."

This homogenization isn't an accident, and it isn't organic. It's the output of a professional services industry operating behind the scenes — one that's worth taking a much closer look at.

Personal branding on LinkedIn stopped being a craft a while ago. It's now an industry with clearly defined tiers of value. At the top sit specialized ghostwriting agencies that sell executives all-inclusive packages: topic research, content calendars, copywriting, posting, comment management — the works. In extreme cases, the executive only has to show up for a monthly interview call. Some agencies skip even that and work exclusively from voice memos or keyword lists.

At the lower end of the price spectrum, AI SaaS tools automate the entire process. Taplio analyzes viral posts across the industry and generates draft copy in the desired tone. AuthoredUp offers formatting assistance, hook libraries, and performance analytics. Supergrow promises complete content pipelines, from topic suggestions to time-optimized posting. All three tools share one thing in common: they turn personality into a configurable product — tone dial included.

The pricing spread shows just how segmented this market has become:

Here's the part that should matter most to marketing leaders: in all three models, the published text is created without any active writing by the person whose name is attached to it. The "CEO's authentic voice" isn't a starting point in this system — it's an output parameter, dialed in somewhere between "relatable-vulnerable" and "visionary-provocative." In conversation after conversation with marketing leaders at mid-sized companies, the same blind spot keeps showing up: many don't even know which of the three models their own vendor is actually running. The invoice arrives every month; the production process stays a black box.

This model didn't take over just because costs dropped. There's a simple technical reason behind it: the LinkedIn algorithm systematically rewards exactly this kind of content.

LinkedIn's ranking system scores posts on measurable signals: How long do users stick around on a post (dwell time)? How many comments show up in the first hour? How many reactions pile up early enough for the algorithm to push the post into extended networks? Every one of these signals can be gamed with standardized post formats.

One-sentence paragraphs stacked with line breaks artificially stretch scroll time — the post looks longer, and dwell time climbs. The "see more" cutoff after a provocative hook line forces a click, which counts as an engagement signal. The reflection question at the end isn't a real invitation to dialogue — it's a comment-generating machine. The "controversial claim, then resolution" arc manufactures pushback in the first few minutes, and pushback is worth more to the algorithm than agreement.

None of these formats express personal style. They're responses to algorithmic incentives — reverse-engineered from whatever has historically performed.

This is where the real problem kicks in: the AI tools generating LinkedIn content today were trained on exactly the posts that drove the highest reach in the past. They reproduce the most successful patterns, those patterns drive reach again, that performance data feeds back in as training data for the next generation of tools — and the loop closes. It's a self-reinforcing homogenization cycle, where the feed gets more uniform with every iteration.

Modern language models like Claude Sonnet 5 or GPT-5.6 are more than capable of producing far more varied, individualized writing. But tool vendors aren't optimizing for individuality — they're optimizing for engagement metrics, because their customers are buying reach, not voice. For a deeper look at the strategic implications of AI-generated brand language, our comparison on AI brand voice in B2B breaks down the current generation of models in more detail.

Before we get into how to spot AI content, it's worth pausing on the counterargument. Because the common shorthand — "using AI equals deception" — doesn't hold up.

Here's an intentionally uncomfortable take: using AI isn't the ethical problem. Accusing a CEO of letting a language model fix typos, organize scattered thoughts, or translate a German draft into English is a distraction from the real issue. Nobody has ever criticized executives for having a PR team draft their speeches — as long as the core message is theirs and they stand behind it.

The line that actually matters sits somewhere else: between AI as a tool for shaping someone's own ideas, and AI as a full-blown impersonation of a stranger's identity. A post whose core idea, experience, and point of view genuinely came from the executive — and was simply polished by a machine — is fundamentally different from a fabricated "personal story" that never happened, published under the name of someone who never even read it.

"The real question isn't whether a machine helped write it. It's whether the person whose name is attached can actually stand behind the content — and whether they're being upfront with their readers about how it came together," says Dominik Waitzer, the author of this piece and a digital communications consultant.

What's notable is that some communities have zero issue with openly disclosed AI use. Developer networks and tech communities tend to respond to a note like "structured with AI, ideas are my own" with respect rather than skepticism — there, transparency reads as professionalism, not as a weakness to hide.

At the same time, plenty of executives still write every word themselves. Ironically, that's now what makes them stand out in the feed: their posts are messier — longer sentences, the occasional typo, tangents, no perfectly engineered story arc. What used to read as unpolished is increasingly becoming a signal of authenticity — a striking reversal of the old rules.

Meanwhile, the debate over disclosure requirements for AI-generated content is picking up steam. The EU AI Act introduces transparency obligations for certain categories of AI-generated content, and platforms like YouTube and Meta have rolled out labeling systems for synthetic media. LinkedIn, however, still has no enforced labeling requirement for AI-generated text posts — which means the responsibility, in practice, falls squarely on the companies themselves.

But drawing the line between legitimate assistance and outright deception in your own feed takes more than ethical categories. It takes concrete, observable signals — and that's exactly what comes next.

No single trait proves AI authorship. But if you spot three or more of the following patterns in a single post — or the same patterns recurring week after week in one person's feed — you're most likely looking at machine-generated or fully ghostwritten content. Across our own research, this combination of multiple signals has proven far more reliable than any single marker on its own.

1. Recycled opening lines. Hooks like "Three years ago, I never would have guessed this," "Unpopular opinion:" or "I made a mistake nobody talks about" come straight out of the hook libraries baked into popular tools. When the same formula shows up repeatedly from the same person, the template is showing.
2. Serial emoji bullet points. Lists where every point kicks off with a thematically matched emoji (rocket, lightbulb, target) are a classic signature of templating systems. People who write on the fly rarely format with that much consistency.
3. Wall-to-wall short sentences, no subordinate clauses. Language models tuned for LinkedIn reach churn out almost nothing but twelve-word-max main clauses. Real people write subordinate clauses, nest ideas, trail off mid-thought, drop in em-dashes — even when it's not exactly polished.
4. Vulnerability anecdotes with zero verifiable detail. "An employee once said something to me that changed everything." No name, no location, no date, no actual quote. Real memories carry specific, often irrelevant details. Generated anecdotes stay systematically vague, because the model is designed not to invent facts that could be checked.
5. Suspiciously symmetrical paragraph lengths. When every paragraph runs exactly two to three lines and the post looks like it was cut with a ruler, that points to machine production. Human writing has rhythm breaks.
6. The same three-beat arc, every time. Problem, insight, call-to-action — same order, same weighting, every single post. The structure is as predictable as a pop song's verse-chorus-bridge. Line up someone's last ten posts and find the identical arc seven times, and you're not looking at a personality — you're looking at a template.
7. A style mismatch between the post and the comments. Possibly the most reliable tell of all: the post itself is polished, rhythmic, error-free — but that same person's replies in the comments are choppy, terse, and inconsistently spelled. Two different authors, one profile.

These markers aren't some academic curiosity. Buyers, journalists, and competitors are increasingly running this exact checklist — consciously or by instinct. And that has direct consequences for how much trust B2B brands can bank on.

The moment an audience discovers that an executive's "personal voice" has been fully outsourced, the fallout doesn't stop at the individual. It hits the brand. The audience's logic here is brutally simple: if the company's supposedly most authentic layer of communication — the CEO's personal experience — turns out to be fabricated, why would anyone trust the product claims, the case studies, or the client references?

In B2B, this carries particular weight, because thought leadership isn't a vanity exercise here. It's a trust mechanism. The research from Edelman and LinkedIn makes the stakes clear:

According to the Edelman-LinkedIn B2B Thought Leadership Impact Study, 73% of B2B decision-makers say thought leadership content is more trustworthy for evaluating an organization than its traditional marketing materials. That's exactly the trust foundation at risk when a supposedly personal voice gets exposed as a prompt output.

In sales cycles that run six to eighteen months, a trust breach like this doesn't stay abstract — it shows up in the numbers. Leads that entered the pipeline through executive content go cold. Conversations built on the management team's perceived expertise lose their footing. And unlike a bad campaign, lost personal credibility can't be bought back with next quarter's budget.

At the same time, a countertrend is emerging: in crowded niches, competitors are increasingly positioning themselves through visibly authentic communication — unpolished videos, self-written posts with rough edges, documented moments instead of staged ones. What the homogenized feed devalues becomes a point of differentiation. Standing out against the prompt-generated sameness is currently the cheapest positioning strategy available — the only cost is the willingness to be imperfect.

Taking this risk seriously doesn't mean scrapping your content strategy. It means adjusting it with intention.

The answer to the ghostwriting trap isn't "turn off the AI." It's clear rules, hybrid workflows, and deliberate differentiation. Four steps have proven effective in practice.

1. Set an internal AI policy. Put it in writing: which levels of AI assistance are acceptable in executive content, and which aren't. Editing and translation: no restrictions. Structuring and drafting: fine, but the executive must sign off on the substance. Fully generated "personal" anecdotes: off the table, full stop. Also define the threshold at which disclosure kicks in — a short note like "structured with AI assistance" costs nothing and protects a lot. What plays out in practice again and again: policies that only get communicated verbally, never written down, dissolve within a few months.
2. Build a hybrid workflow. Let AI handle research, outlining, and the first draft — the executive delivers the final wording of the core message and adds the personal details no model can invent: specific situations, names (where appropriate), numbers from the actual business, real doubts. Rule of thumb: at minimum, the first and last three sentences of every post should come verbatim from the person posting. For a closer look at how to set up these workflows cleanly, without adding extra work for leadership, check out our AI & Automation practice.
3. Stress-test your own posts before publishing. Run your company's executive posts through detection tools like Originality.ai on a regular basis to catch unintentionally generic output — before followers, journalists, or competitors do it for you. A post that reads like prompt output does damage even if a human typed every word. Treat this as a quality audit, not a loyalty test for your executives.
4. Write against the feed on purpose. Favor imperfect, specific details over polished story arcs: a real customer conversation instead of a generic "learning journey," an open question with no tidy answer, an actual number from your own operation instead of an industry platitude. What can't be replicated in a homogenized feed is the only thing left that stands out there. For how these principles fit into a broader, channel-specific content strategy, see our approach to Social Media Marketing — and if you're working on brand positioning at the same time, our Brand Strategy & Design practice offers the right framework for that.

These four steps aren't a one-time project — they're an ongoing calibration. Tools change, the algorithm changes, and audience expectations shift faster than any guideline can keep up with.

For marketing leaders, this reframes the real question. It's no longer about whether executive content involves AI — in most cases, it already does. It's about how traceable and how genuinely their own that content stays. That's exactly where the line gets drawn between who gets recognized as a credible voice in the years ahead and who disappears into the prompt-generated sameness. Companies that calibrate their policies and workflows now are securing an advantage that will only grow as trust in generic executive posts keeps eroding. Wait too long to make this shift, and you'll end up buying back lost credibility at a much steeper price — through longer sales cycles and followers who got burned once and never came back.

Here's the concrete next step: this week, pull your company's last ten executive posts and check them against the seven text patterns outlined above. If three or more show up, it's time to act — not eventually, but before someone else notices first. At the same time, lock in your internal policy on AI use and disclosure. Because in a feed full of prompt echoes, a demonstrably authentic voice isn't a nice-to-have anymore. It's the last differentiator left standing.