I used to ask AI to use real-life examples without actually giving it any.
Yep, sounds ridiculous, right?
I can laugh about this now, but at the time I thought I was being pretty clever with my baked-in tone-of-voice system instructions.
I would write things like:
My tone should be encouraging. Clear and practical. Warm but focused. Use real-life examples. Invite the reader into the process. Focus on helpfulness and personal connection over hard selling.
On the surface, that sounds sensible. I had told AI how I wanted the writing to feel, so surely that was the bit that would stop the content sounding generic.
I was so naive.
Because I was asking it to do things I had not actually given it the substance to do.
If I had not supplied the real-life example, AI had to fill the gap. If I had not gone far enough into what I meant, AI had to fill the gap. And if I had not made an important decision about the point or what I actually wanted to say, guess what? AI made that decision for me too.
Tone guidance can shape the expression. It cannot supply the thinking underneath it.
Fortunately, after a few different systems, I did get better at this.
Later, I put my learning into a framework around getting more clarity before using AI. It was human-led, hands-on and gave great awareness (so I was told) of what it took to stay in control of the tool.
You would work out who the piece was for, what it needed to do, the message it needed to communicate and the outcome for the reader.
Then there was a final step: what must not change?
That was where I tried to protect things like tone, meaning, intent, important phrasing and professional boundaries. One of my examples was along the lines of:
AI must not change my plain-spoken, measured tone. No dramatization, heightened persuasion or adaptation of my lived experience.
That felt like a much better way of working than simply saying, “write this in my tone of voice”.
I taught that framework in workshops too, and in one workshop someone told me it had not worked. The output did not sound like her.
New to presenting, my confidence took a hit and my first internal reaction was: I’d failed.
But my intuition kicked in and we went through what she had done. Had she followed all the steps?
No.
She had missed the final one. The exact step I had put there to stop AI changing the things that were supposed to make the writing feel like yours.
At the time, that made perfect sense to me. There was the problem. She had skipped the step… phew.
Looking back now, I know that was only part of the answer.
Even with that step in place, it could only protect what she had actually supplied. If the lived experience or the deeper meaning was missing, there was nothing for that instruction to preserve.
You can give AI excellent instructions around the writing and still leave the most important parts for it to guess.
That is the bit I underestimated for a long time.
I was putting a lot of pressure on the instructions around the writing, when the bigger issue was what the process had actually captured from me before the writing started.
And this is where the way I work now is very different.
There was a stage where, if AI content came back generic, I would look at the interaction itself.
Had I explained myself properly? Had I given it enough context or enough of my own experience? Had I handed too much over?
Those questions helped me understand what was going wrong.
But I do not work piece by piece like that anymore, and I definitely would not try to solve all of this with one giant prompt.
Clarity Content OS is being built specifically so I do not have to remember halfway through a draft that I should have gone deeper earlier.
The methodology makes me establish what is actually there to work with, get clear on what I really mean and what the reader needs, strengthen something if it is still too surface-level, and bring back relevant thinking I have already established rather than making me recreate it from scratch.
I no longer treat the AI interaction as the centre of the content process.
That matters because otherwise it is very easy to diagnose the wrong problem.
A draft comes back generic, so you add more tone guidance. It still feels flat, so you add more context. It still does not quite feel like yours, so you add another list of words you use and words you would never use.
Some of those instructions are useful. I still care about words, phrasing, how I naturally explain something and the boundaries around how an experience is represented.
But they are not carrying the whole job anymore.
What matters more is whether enough of your actual thinking, experience and meaning made it into the process before AI was asked to write.
If that part is thin, AI can still give you something polished. It can still give you something readable. It can even give you something that sounds broadly like you.
And that is exactly why “does this sound like me?” has become too low a bar for me.
Something can sound like me and still not be mine.
If AI has filled in the real point for me, invented an example, taken the thought further than I would, or quietly made a decision I should have made myself, I am still left with something I would not genuinely stand behind.
A piece can pass the voice test and still fail the authorship test.
That is the standard I care about now.
I want the expression to feel natural, of course. But I also want to know that the piece represents what I actually mean.
That is why I am much less interested now in asking, “How do I make AI sound more like me?”
I am more interested in the process around it. AI is involved throughout Clarity Content OS, but its role is very different from the one I used to give it. It can help me capture, organise, question, connect and strengthen my thinking, without being left to decide what I mean or fill in the parts that should come from me.
That distinction has made AI more useful to me, not less. I am much clearer now about what I want the tool to help with, what still needs me, and what standard the finished work has to meet.
Because “that sounds like me” is not enough.
I want to read something and think:
Yes. That is mine.
And ultimately, I would genuinely put my name to it.
