AI can copy your voice. It was never the thing worth protecting.

Everyone in my inbox is guarding the wrong door.

My inbox this fortnight has been one long anxiety attack with a Substack logo on it.

Strip out the recipe newsletters and the horror serials and the bloke who writes entirely in ALL CAPS about geopolitics, and what is left is a single nervous question asked forty different ways. They can copy your voice now. The machines have read enough of you to do you. So what is left that is yours?

The advice underneath is always the same, and it is always a little frightened. Lean into your taste. Your judgment. Your lived experience, your receipts, the trust you have built with readers. Be so particularly you that no model can counterfeit it. I read a good version of this argument last week, and a dozen worse ones, and they all rest on a quiet assumption I want to drag into the light, because I think it is wrong.

The assumption is that your voice is the precious thing. The fingerprint. The bit worth bolting to the floor.

It isn’t. And I can tell you why with rather more authority than I would like, because I spent the better part of sixty-six years building the most convincing voice you ever heard, and it nearly killed me.

Here is the part the panic keeps walking straight past. A voice is a performance of fluency. That is all it is. A reliable, repeatable way of sounding like yourself. And a performance of fluency is precisely the thing that a great many late-diagnosed neurodivergent people got praised for, right up until the performance ate them alive.

There is a clinical word for it: camouflaging. Researchers have spent the last decade documenting it properly, the conscious and effortful business of masking how your mind actually works so you can pass as the standard model. Hull and colleagues (2017) titled one of the foundational papers “Putting on My Best Normal,” which is about the most accurate five words anyone has managed on the subject. You learn the script. You sand off the odd edges. You perform the version of competence the room expects, the room claps, and nobody, least of all you, clocks the cost until the whole bill arrives at once.

I got my AuDHD diagnosis at sixty-six. I had been doing my best normal since before half of you were born, and doing it so well that the misdiagnoses stacked up over the years like unpaid parking fines. The voice worked. The voice was the problem.

So when a writer tells me they are terrified a machine can now reproduce their voice, my first, uncharitable, very Australian thought is this: mate, so could you, and look where it got you.

Because the model can have the voice. Take it. What it cannot have is the decade I spent as one of Australia’s loudest social media evangelists before the whole thing curdled and I walked away with my nervous system in pieces. It cannot have the weight of sitting with a veteran at three in the afternoon while he decides, in real time, whether to keep going. It cannot have the morning the fog came down the valley here in Đà Lạt and I understood, in my body rather than my head, why I had to leave Australia to stop being ill. The model can produce a sentence about any of that. It cannot have paid for it.

That is the thing under the voice. Not style. Cost. The scar tissue, and the judgment that grows over scar tissue the way bark grows over a wound: knowing what to leave out, knowing which sentence is a lie even when it scans beautifully, knowing when a paragraph is true and when it is merely impressive. A model can be impressive all day long. Impressive is cheap now. It was always cheaper than we pretended.

Now the bit that will get me snottograms.

I co-write with a machine. Have done for two years. This essay was drafted in collaboration with Claude, the way most of my work now is: I bring the argument and the scars and the final, non-negotiable no, and it brings the speed and a tireless willingness to be told the last paragraph was rubbish and to try it again. I do not hide it. It is the subject of my next degree. I have come to think the hiding is the new mask, and that a good many writers currently performing lonely-genius solo authorship are about to learn what I learned at sixty-six. The performance is exhausting. It fools fewer people than you hope.

The writers clutching their voice to their chests are guarding a door the burglars already strolled through. Meanwhile the real valuables, the lived receipts and the judgment and the willingness to tell the truth at your own expense, are sitting in an unlocked room with a sign on it that reads too much effort to fake.

So I am not going to tell you to protect your voice. Protect the other thing. The expensive thing. Go and live a life costly enough that no model can afford it, report back as honestly as you can manage, and let the machine help you type it up if it likes.

The voice was never yours to lose. It was only ever the bit on top.

-----

Reference

Hull, L., Petrides, K. V., Allison, C., Smith, P., Baron-Cohen, S., Lai, M.-C., & Mandy, W. (2017). “Putting on my best normal”: Social camouflaging in adults with autism spectrum conditions. Journal of Autism and Developmental Disorders, 47(8), 2519–2534.