AI-Generated vs. Authentic: How to Automate Social Without Losing Your Voice
People can't reliably tell AI-written text from human-written text. In a 2023 study published in the Proceedings of the National Academy of Sciences, participants identified AI-written short self-descriptions with only 50 to 52% accuracy, about the same as a coin flip (Jakesch, Hancock, and Naaman, 2023). And yet when a social post carries an AI disclosure, it gets roughly 7 to 8% fewer likes, according to an analysis of more than 1 million TikTok posts (Carney, Riveros, and Tully, 2026).
Read those two findings together and something interesting shows up. People mostly can't tell. But once a post is labeled as AI-made, they engage with it less, and the researchers traced that drop in part to an assumption that the creator put in less effort. The fix is less about hiding AI and more about showing real care.
What Actually Gives It Away
So what are people reacting to? A September 2026 analysis of 90,000 AI-written articles by the growth agency Graphite found that every model's word choices were closer to at least one other model's than to human writing, and that AI sentence lengths varied much less than human ones (Graphite, 2026). A peer-reviewed experiment found the same sameness in creative work: writers who used AI story ideas produced stories rated as more creative, but those stories were more similar to each other than stories written by humans alone (Doshi and Hauser, 2024).
Sameness is the common thread, and it's about effort. A post that follows the same shape every time, reaches for the same safe words, and never names a specific detail reads that way whether a person or a machine wrote it. AI just makes it easy to produce a lot of it very quickly.
The Fix Is More Specific Than “Add a Human Touch”
Two things the research points to, and neither requires abandoning automation.
Add what only you know. Graphite found human writers speak in the first person, tell small anecdotes, and draw on personal experience, which its AI articles did not do in the same way (Graphite, 2026). Those are the details a tool can't invent about your business. One real, specific detail, like a customer's situation or a real number from a real job, does the work.
Change the rhythm. In Graphite's analysis, AI sentence lengths varied much less than human ones (Graphite, 2026). Adding one very short sentence and one longer, looser one is a quick way to break the pattern.
Put together, that's a small, sustainable habit: let AI produce the first draft, then make it specific and uneven instead of smooth. These are my own recommendations, built from the research above, not a tested formula.
There's a third way to get ahead of this, and it changes where the fix happens. Instead of catching genericness after a draft is written, a real setup process can train the system on how you actually talk before it ever drafts a word, your own phrasing, your own way of explaining things, not a generic, one-size-fits-all tone. Done well, that means less editing needed after the draft, because the draft already sounds like you instead of like anyone.
This Connects to Something Already True About Good Content
The fix here isn't really an AI-specific rule. In the research above, the reaction tracks sameness and perceived effort more than the tool itself. The researchers found the drop in engagement for AI-labeled posts was driven in part by people assuming the creator put in less effort, and it did not come from a general dislike of AI (Carney, Riveros, and Tully, 2026). AI content gets noticed not because it's AI, but because it so often defaults to generic when left alone. The fix is the same one that makes any content better, applied on purpose.
One Thing to Check Right Now
Pull up the last five posts your business published. For each one, could a competitor in a different city have posted the exact same thing without changing a word? If yes, that post had no specific detail anchoring it to your actual business, the exact gap that makes content read as generic, automated or not.
Frequently Asked Questions
Does this mean I should stop using AI to help with posts?
No. The research doesn't say to stop using AI. It shows that people engage less when they assume a creator put in less effort, and that AI writing tends toward sameness. A quick review and one real detail address both, and you keep the time AI saves you.
Do I need to disclose when a post was AI-assisted?
Follow each platform's rules on labeling. The research is mixed on what labels do: one large study found posts with an AI disclosure got roughly 7 to 8% fewer likes (Carney, Riveros, and Tully, 2026), while Ipsos reports that 79% of people say companies using AI should have to disclose it (Ipsos, 2026). There's no safe wording to copy. The more dependable protection is a post with a real, specific detail in it.
What's the single highest-leverage change I can make starting today?
Add one real, specific detail to whatever you post next: a customer's actual situation, a real number from a real job, something that couldn't have been generated without you. It speaks to both problems the research raises, sameness and perceived effort.
Here's the Plan
You don't need to choose between automating your social presence and sounding like a real business. You need a system that produces a draft quickly and still leaves room for the one real detail that makes it unmistakably yours.
If you'd rather have a system that's actually trained on your voice from the start, so there's less generic to catch in the first place: most business owners don't have time to spare on every post, which is exactly the gap this is built to close. Book a free, zero-pressure strategy session with me today. If you want us to build and manage it for you, great. If not, you still walk away with a free visibility roadmap.
This piece is part of our Social Media Automation series. Start with the overview: Consistent Without the Chaos: How AI Keeps Your Business Visible on Social Media
Sources & Further Reading
AI Tells, Graphite (2026), the source for the sameness findings and the human writing traits
