The Authenticity Problem in Online Writing (And Why It Predates AI)

Online writing has felt progressively less authentic for longer than AI has been a concern. The complaints are consistent across platforms and years: everything sounds the same, everything is optimized, nobody is saying what they actually think.

The AI conversation sits on top of this. But the underlying problem started somewhere else.

The optimization layer

When you write for an audience you can measure in real time, you start unconsciously optimizing for what gets engagement. LinkedIn posts that start with a provocative single-line paragraph get more clicks. Twitter threads written in a specific punchy style get more shares. Blog posts written to rank for specific keywords look different from blog posts written to be useful.

This optimization is usually unconscious and gradual. Writers don't decide to become inauthentic. They just notice what works and do more of it — and over time, the distance between "what I'd say in person" and "what I write" grows.

The persona problem

Social media made it easy to have a public writing persona that's separate from how you actually communicate. The persona gets refined over time: softer, more optimistic, more strategic. The rough edges — uncertainty, contradiction, unresolved thoughts — get filed off because they don't perform as well.

The irony is that the rough edges are often the most interesting part. The thing that's not quite resolved is more engaging than the thing that's been polished into a clean take.

Why AI accelerated this but didn't create it

AI gives you a way to produce the persona-optimized version of your thoughts without doing the persona-optimizing work yourself. The output is the kind of writing that was already overrepresented online — confident, clear, slightly generic — but faster and cheaper.

So the internet got more of what it was already becoming. AI didn't introduce inauthenticity. It subsidized it.

What authentic online writing actually looks like

It has a position that isn't the default consensus. It's willing to be wrong. It contains something specific enough that only that person could have written it — a detail, a reference, a way of seeing something.

It doesn't read like it was written for metrics. And increasingly, readers can tell the difference — not because they're running it through detectors but because they've been reading enough optimized content to recognize when something isn't.

The demand for genuine writing hasn't gone away. The supply is just harder to find.