My LinkedIn feed is full of discussion about AI as a teaching tool, or students’ use of it in schools. I see far less about what it means for professionals in education. This is the second in a short series of articles on what I believe it actually means to use AI well in a professional context.
Disclosure 1: I used AI as a thinking partner in developing this piece. I will discuss what this means in practice, and why it matters, in the next instalment in this series.
Disclosure 2: I am still trying, with limited success, to use the em dash sparingly. It turns out the instinct runs deeper than I realised. More on that below.
Not long ago, I sat down to solve a practical problem. As I discussed in the first article in this series, I had come to realise the extent to which unedited AI output homogenises professional voice, with real costs to individuals and to the organisations that depend on their thinking. The next question was: what can you do about it? How do you get AI output close enough to your own style that it sounds like you and not a generic response that could have come from anyone?
What I thought I knew
My answer was to build a style guide. These were not new territory for me; I’ve always been a stickler for them. I’ve even written them for previous schools I’ve worked in, covering everything from using correct job titles and subject names to when to capitalise ‘university’ and how to pluralise ‘heads of department’. I even have a copy of The Economist Style Guide in my Kindle library as a source of reference and inspiration. But what I learned in the process of applying this to AI is that a style guide alone isn’t enough. You also need a corpus — a body of your own writing for AI to pattern-match against — and an instruction set that tells it how to use them together. What I didn’t anticipate was what building all three would teach me about the way I think.
What emerged first was this: thinking about how I express myself in general was the wrong frame entirely. I needed to acknowledge that I write across three distinct domains, each with their own set of rules.
My professional communications (mainly via email) have their own set of conventions around tone, relationship to the recipient, and a tendency to summarise and sign off before going into the detail in a kind of extended postscript. It turns out my structured documents (reports, proposals, positioning papers) are surprisingly consistent in, well, structure, as well as register. Then there are the opinion pieces: LinkedIn posts, articles, philosophy statements — the type of writing where my own voice is most visible.
Obvious in retrospect, perhaps. But it wasn’t at first. It turns out that establishing the taxonomy mattered more than I had anticipated. It meant I could build something precise for each text type rather than forcing all three into a single guide, which would, ironically, have produced exactly the kind of homogenisation I was trying to avoid.
As I iterated and tested versions of the system, I found that building it required a particular kind of precision. Impressions weren’t enough. Everything had to be evidenced, named, cross-referenced, and specific enough to be actionable. “Rob writes concisely” doesn’t tell an AI anything it can act on in a consistent way. “Sentences average 27 words in long-form pieces; subordinate clauses appear once per paragraph at most; short sentences under 15 words are used deliberately to land conclusions or mark pivots” does. The discipline of having to make every rule falsifiable forced a level of specificity I hadn’t realised I’d need.
That discipline nudged me toward using AI itself to build the system. I created a dedicated project and used it to analyse samples of my own writing, identify patterns I hadn’t consciously noticed, and draft the rules and instructions that would eventually form the guides. I won’t go further into the technical detail here, but if you’re curious about the specifics, get in touch.
Conscious choices and unconscious habits
There is something disconcerting about having your own writing analysed back at you. AI held up a mirror, and the reflection wasn’t quite as glamorous as I might have hoped. I had to decide whether to nudge it towards what I like to think is my style, or accept that I might have a somewhat romanticised view of it — and that leaning too hard into that ideal would sand away the very foibles and rhetorical habits that make my writing distinctively mine. Below are three things the mirror showed me: one about how I signal conviction when stating an opinion, one about identity and register, one about my relationship to my audience. Each raised questions you may wish to ask about your own writing.
The analysis of my email archive revealed that “I would suggest”, “I would lean towards”, and “my sense is that” are not stylistic variants of the same move. They sit at different points on a confidence spectrum and signal different things to the reader about how much pushback is being invited. I almost certainly deployed these instinctively, and I couldn’t have told you which phrase means what before seeing it laid out. That’s not just a writing observation. It’s a finding about how I structure authority and deference in professional relationships, and it matters particularly when I am advising rather than deciding. The right phrase signals how firmly I am pushing and how much room I am leaving for the other person to disagree.
When you state a view in a professional context, how clearly does your language signal the level of conviction behind it? Have you considered how the words you choose land differently depending on who is making the call?
The opinion pieces I had written told a different story about identity. Close reading across a large enough body of work showed that I use the first person in more distinct ways than I had appreciated, each carrying a different authority posture: the institutional ‘we’ of a school leader making commitments on behalf of an organisation; the professional community ‘we’ of a practitioner appealing to shared responsibility; and a more personal ‘I’ for when I am speaking for myself. Getting these wrong — using the institutional ‘we’ when the situation calls for the personal ‘I’, for instance — changes how authority lands with the reader in ways that are hard to unpick after the fact. I knew I shifted registers. I didn’t know the shift had that much internal structure, or how significantly that choice shapes the way a reader interprets my role, my authority, and my relationship to what I’m writing about.
Can you recall reading something where a personal ‘I’ would have cut through in a way that a cold institutional ‘we’ simply didn’t? When you write in a professional context, are you conscious of which version of yourself is doing the writing?
The most counterintuitive finding came from looking across all three types of writing. The initial analysis suggested I avoid loose superlatives — consistent with what I’d tell you if you asked me to describe my style. But the newsletter essays told a different story: I use them freely when writing to a school community. The surprise isn’t the finding; it’s what it implies. My register doesn’t only shift by format. It shifts by relationship to audience. When I address professional peers, I am economical and precise. When I address parents and students, something warmer and more declarative comes through. Getting that wrong in either direction carries a cost. Too informal in professional writing and you risk sounding unserious; too clinical with someone who needed a human connection and you risk sounding indifferent to something that is really important to them.
When you write to different audiences, do you shift register to reflect the relationship as well as the subject matter? Is there an audience you write to where you suspect the relationship deserves a different tone than the one you default to?
From sounding like me to thinking with me
The through-line across all three observations is the same: none of these things were hidden. They were all there in my writing. But I couldn’t have generated these findings by just thinking carefully about how I write, or by going back through my work manually. The volume and variety of material required systematic analysis at a scale I couldn’t do alone. That’s where AI came in. Working through a large enough body of work, it was able to surface patterns I hadn’t consciously noticed and couldn’t have found myself. Crucially, it could also structure them in a form it could actually apply consistently and effectively to imitate my style.
You cannot tell an AI how you write without first understanding, precisely and not just impressionistically, how you actually write. But getting it closer to your voice doesn’t mean the output is ready to send. The real value lies in interrogating what it produces — pushing back on its conclusions, finding where it has glossed over a judgement call you need to make, and identifying where your position is more pointed than what landed on the page. That is the subject of the next instalment: how to use AI as a genuine sparring partner rather than a sophisticated autocomplete.
Em dash count: 9. Not too bad.