My AI clone sounds just like me. That's the problem.

I've recently started writing again after many years of avoiding it. It wasn't a conscious decision, I just didn't feel the urge nor did I feel like I had anything to say. Funnily enough the thing that got me back to writing was AI and its freakishly good writing skills... and a bit of a disconnect.

Back in the day I wrote all sorts of things and felt like I'd developed a really clear voice and style. We're not talking great literature here but I was happy with what I was producing. I used to write a monthly article about websites for Desktop magazine, and a stack of course content for Shillington Education. Didn't matter what it was, I made sure it sounded like me. Why not take the piss, chuck in a dumb joke or odd reference to make a point land? It sounded like me and that was a good thing.

Don't get me wrong, I love having Claude construct a careful and considered email based on my angry ramblings (so does the kids school). Or get ChatGPT to simplify something technical when I've just got stuck in the weeds. Maybe a heartfelt message on a card for my Mum on her birthday... sure why not. Oh that's proper grammar and punctuation is it? I had no idea... bravo Claude! I'll hand it all over to you from now on.

The problem with that is most of the time it just didn't sound like me. What if my Mum found out? I wouldn't be the favourite anymore... not that she has favourites (well that's what she tells my brother and sister). So I decided to create a tone of voice profile for myself with the help of my trusty AI mate.

What are the benefits of a tone of voice profile?

There's a number of reasons why this might be a useful exercise.

• If you're using your favourite AI tool as a writing partner or editing assistant, then having the tone right from the start saves you time.
• Or you might be running a business so having that consistent tone of voice that aligns with your brand is super important.
• Perhaps you have a team of people you work with, then sharing that tone of voice profile maintains consistency across all your communications, no matter who is writing it.
• If you're automating simple tasks like email drafting for example, then having it sound like you from the start will save on tweaking it afterwards.
• Finally, it avoids that generic AI output that everyone else can achieve using the same tool.

What's the process?

1. First up I handed over some of my recent articles to Claude. Finally I had something recent that I felt really represented my writing style.
2. Next I gave it some recent emails and labelled them as either 'work' or 'personal'.
3. I also dropped in a few recent job application letters to give that mix of formal and informal.
4. I then asked Claude to review all this material and create a tone of voice profile for me.

The best part of this process was Claude's initial impression of my writing style: "conversational, self-deprecating, dad jokes, light swearing, practical, encouraging, Aussie-casual".

I'll take that. Who doesn't like a bit of light swearing and dad jokes? My kids obviously think I'm hilarious... although I'm still not allowed to use 'slay' for some reason.

So I saved the resulting 'tone-of-voice' markdown file into my Claude AI Assistant setup and made sure the agent was aware of it and how and when to use it. You could also save this into your settings in any other AI tool, just ask the AI where to add it if you're not sure. For a business focused profile you might create a global SKILL based on this profile that any team member can use. You could ask the AI to create this skill for you too. Boy these things are useful!

The results

Now to unleash my odd little clone on some real-life writing tasks.

I've been applying for a few jobs of late so what better way to test this out than on a job application letter, sure I love writing them, but I can't have all the fun, can I? So I picked three jobs, gave the AI the job ad content plus my CV and a few notes then asked it to draft a letter for each using my tone of voice profile.

In no time at all I had three very polished cover letters. They definitely sounded like me, used the sort of phrases and style of writing I'd use, and because it had my CV it said the right things about me and my experience. But it still didn't sound quite right. So I pointed it at some research I'd done on a particular topic and asked for a first draft of an article. Once again it said all the right things, but there was still something not quite right.

On re-reading the article and letters the thing that didn't sit right with me was that it was way too close to how I'd write it. Yes I might say that line or phrase something that way, but I didn't. I know I didn't, and that's what was getting me all worked up. It's like that whole uncanny valley effect where something is just way too close to being realistic but not quite, and you're brain has a double take..

So what am I to do? Was all that work for nothing?

Not at all, it's actually made me more determined to keep on writing, especially when it's something I care deeply about or an idea I'm trying to squeeze out of my brain. I actually like the process of writing to work through an idea. When I say 'like' I really mean that this is hard, my brain is working overtime, so this is probably good for me. It helps me find the gaps in my understanding and forces me to think more deeply about the topic. Handing that off entirely to the machine seems like a bit of a cop-out.

But I still like having the AI as my collaborator or sounding board. Critiquing my work, suggesting fixes to a paragraph or even just a good old spell checker. It's great for gathering additional research or finding gaps in my arguments or understanding. Oh and it's still much better at drafting those tricky emails or detailed instructions for a task I'm trying to explain to someone. In those instances having the voice profile will help immensely.

Maybe the most obvious prediction of what word(s) should come next isn't the most interesting one. Maybe the random, quirky, unexpected rambling ones are much more valuable because they're unexpected.

Sorry Mum all those cards from now on will be written by me and me alone! Would your favourite child lie to you?

I'd actually finished this article ready to publish before Anthropic announced they were going to watermark all AI generated content moving forward. I think its an interesting idea if they can make it work. I do think that if someone is paying you for the content you are producing then they should be aware of where that content is coming from. I'll definitely continue to pick my battles when it comes to what I write, what the AI helps with and what I hand over entirely. Funnily enough maybe this will lead to the rise of new jobs for all those expert touch typists out there. We'll give them those massive AI generated documents and get them to retype them without that pesky watermark.