Does using AI make your work look bad?

No. Using AI is not what readers object to. They object to work that shows no evidence of a human decision anywhere in it. A piece that took a thousand rounds of judgement does not read as AI even when AI helped at every step, and a piece that took ten seconds reads as hasty even when a person typed every word. The tell is effort, not authorship.

How do people tell if something was written by AI?

Not reliably by surface features. Em dashes, the word 'delve', tidy three-item lists and AI-detector scores all produce false accusations against careful human writers. What readers actually catch is the absence of specificity: no named example, no number that could only come from your own experience, no opinion that risks anything, and no evidence that anything was cut.

Are AI content detectors accurate?

Not accurate enough to act on. They flag human writing as machine-written often enough that schools and publishers have walked back their use, and they penalize plain, well-organized prose in particular, which means non-native English speakers and disciplined writers get flagged disproportionately. Treat a detector score as noise, not evidence.

Why does being over 50 help with AI work?

Because the scarce skill is no longer producing a draft, it is knowing which draft is wrong and why. That is taste, and taste is accumulated from decades of seeing work land and fail in a specific domain. AI has collapsed the cost of the first draft to nothing, which makes judgement about the fifteenth draft the thing that separates output, and judgement is the part experience actually gives you.

How do you make AI-assisted work not look like AI?

Reject the first draft on principle. Put in at least one detail only you could know: a real number, a named failure, a date. Cut the part you are proudest of if it is not earning its place. Take a position that could be wrong. And keep going past the point where it is good enough, because good enough is exactly what ten seconds produces.

— For people who still think a thing should be good —

You don't lose for using AI. You lose for looking like you spent ten seconds.

There's a panic going around that using AI will get you found out. Wrong panic. Nobody can reliably tell, and the people claiming they can are accusing innocent writers at an embarrassing rate.

What people can tell, instantly and without a tool, is that you didn't care very much. That's the thing costing you work. And it happens to be the thing your last thirty years fixed.

I'm three months into a video game. It's about half done.

I'm building it with AI's help. Every last detail. More than a thousand rounds of feedback so far — me looking at what it gave me, saying no, saying why, saying try it this way instead, throwing out things I'd spent a whole afternoon on.

In roughly the same window, several thousand people have posted videos about a new model's benchmark scores. Many of those were made in an afternoon. Some in ten minutes.

We both "used AI." One of us is making a thing.

The tools got equally good for everybody on the same Tuesday. What didn't get distributed equally is the willingness to sit there and go again.

The tell is not what you've been told it is

You've seen the lists. Em dashes. The word "delve." Suspiciously tidy groups of three. "It's not just X, it's Y."

Those lists are mostly nonsense, and they do real damage. Careful writers who've used em dashes since 1987 are now getting accused of being robots. Detector tools flag human writing as machine-written often enough that schools and publishers have quietly backed away from them — and they're hardest on plain, well-organized prose, which is to say they punish exactly the people who learned to write properly.

Here's a grading scale for it. If a detector tells you a piece is 87% AI, treat that the way you'd treat a stranger at a bus stop telling you your blood pressure by looking at you. He might be right. You would not take a pill over it.

So what are readers catching? Something much simpler, and much harder to fake.

Reads as hasty "There are several key benefits to this approach."

Reads as made "I tried this three times. Twice it fell over on the same step, and here's the step."

Neither sentence proves who typed it. The second one proves somebody was there. That's the whole game.

What "looking like AI" actually means

Strip away the folklore and it comes down to an absence. Work that reads as machine-made is missing the fingerprints of a decision. No number that could only come from your own week. No name. No date. No opinion that could get you argued with. Nothing cut, so nothing chosen.

It's the prose equivalent of a hotel room. Perfectly fine. Clean. Nobody lives there.

The five things missing from work that reads as machine-made

A detail only you could know. A real number, a named failure, a specific Tuesday.
Evidence something was cut. If nothing was thrown away, nothing was chosen.
A position that could be wrong. Safe on every side means nobody is home.
An unexpected shape. Five even sections with five even paragraphs is a template, not a thought.
Any sign of the second attempt. First drafts have a smell, and everyone can smell it.

Notice that not one of these is about whether you used AI.

Why this is very good news if you're over 50

Here's the part that should cheer you up considerably.

For most of your career, producing the draft was the expensive bit. Writing the report, building the model, making the thing exist at all — that took skill, and it took hours, and being able to do it was worth money.

That's over. The first draft now costs nothing and takes eleven seconds. Which sounds like bad news right up until you ask the obvious follow-up: if everyone can produce a draft, what's scarce?

Knowing the draft is wrong. Knowing why it's wrong. Knowing what good looks like in your particular corner of the world, because you've watched a few hundred attempts land and fail there.

That's taste, and taste does not download. It accrues. It is the one asset in this entire situation that gets more valuable the longer you've been alive, and it is precisely what someone making their fourth benchmark video this week does not have.

The twenty-six-year-old and you now have the same tools. They can produce a draft as fast as you can. What they cannot do is look at it and know, in a second and a half, that the third paragraph is a lie. You've sat in the meetings where that paragraph blew up.

The fix, which is boring

I'd love to give you a clever trick. There isn't one, and any page offering you a list of words to avoid is selling you the same ten-second thinking that caused the problem.

What works is going again.

Reject the first draft on principle. Not because it's bad — it's often fine. Because "fine" is the exact texture of work nobody remembers.
Put in one thing only you could know. One real number, one named mistake, one date. It takes a sentence and it changes the whole piece.
Cut the bit you're proudest of if it isn't earning its place. This hurts, which is how you know it's the job.
Say something that could be wrong. Then label it as your opinion, so you're honest as well as brave.
Go one round past "good enough." Good enough is where the ten-second crowd stops. That gap is the entire competitive advantage on offer.

A thousand rounds is not a brag. It's slightly embarrassing, if I'm honest, and my game is still only half finished. But I'll tell you what a thousand rounds buys: nobody who plays it is going to wonder whether anyone cared.

The short version

Using AI is not a confession. Publishing something you didn't think about is. The market is flooding with work that took ten seconds, and the flood is doing something useful — it's making obvious effort rare, and therefore worth something.

You've been doing obvious effort your whole working life. You just didn't know it was about to become a differentiator.

If you want the practical version

This page was the argument. These are the pages that show you the moves — how to actually work with AI so what comes out has your fingerprints on it, not the machine's.

Pick one AI tool and stop shopping Tool-hopping is its own form of ten-second thinking. Choose one, go deep.
The "just ask AI" trick The habit underneath all of this: stop learning software, start asking for outcomes.
Turn decades of experience into income with AI The taste you already have, pointed at something that pays.

No pitch on this one. I just wanted to say the thing out loud, because the panic about being caught using AI is pointed at completely the wrong target. — Scott