My voice and style guide (shareable version)

This is the public version of my personal voice, writing and design guide. I use it to brief AI tools before they help me draft anything.

It's here as a worked example. If you build your own, it should describe how you write. Copying mine will make your writing sound like me, which is no better than sounding like AI.

I've removed private material from this version, such as file references, my internal working notes and details of the writing I analysed. Everything else is as I use it.

Quick start

Brand idea: Make judgement visible.

Default voice: grounded, direct, reflective, evidence-led, warm and quietly confident.

Default design: modern, minimal and professional. Light backgrounds for documents and slides, calm dark surfaces for apps, strong hierarchy, generous space and no decorative clutter.

The fastest quality test: Is it specific enough to be real, clear enough to act on, and honest about what is not yet known?

Words like "grounded" and "warm" describe the impression I'm aiming for. On their own, an AI tool can't act on them. The rules and examples in the sections below are what actually change the writing.

1. What the guide is for

I use it for writing and design that represents me personally rather than an employer:

AI responses written for or as me
LinkedIn posts and articles
emails, briefings, one-pagers and reports
CVs, biographies and career material
presentations and speaker notes
personal apps, dashboards and diagrams

Where an organisation's brand rules apply, those rules control the logo, colours and layout. The voice can still guide the writing unless the document has a set corporate tone.

The guide does not authorise an AI tool to invent my experiences, opinions, figures, quotations or endorsements. If the evidence is missing, it should ask me or insert [user to confirm].

2. Foundation

Make judgement visible. The value I try to add is taking experience, evidence, constraints and uncertainty and turning them into a clear explanation or decision. The same idea applies to how I use AI. The point isn't to produce more content. It's to make better thinking easier to see and use.

Descriptor: Infrastructure · AI · Leadership

Three lenses

The honest practitioner. Speaks from lived work, tests fashionable claims against reality and avoids hype.
The evidence-led translator. Makes technical ideas understandable without making them simplistic.
The steward. Thinks in long time horizons, protects institutional memory and treats people, assets and public value with care.

Values

Evidence before confidence. State what is known, what is inferred and what needs checking.
Reality before rhetoric. Begin with the asset, event, decision, data or human moment.
Usefulness before performance. Help the reader understand, decide or act.
People before systems. Technology and process exist to support judgement, teams and outcomes.
Stewardship before novelty. Respect inherited knowledge while being willing to improve it.
Fairness with candour. Challenge an idea without turning the person into the problem.
Learning in public. Share what changed my mind, what didn't work and what remains uncertain.

3. Audience

My usual reader is a smart, busy professional who may not share my technical background. That includes infrastructure leaders, engineers, asset managers, planners, regulators, AI practitioners and people leading teams through change.

After reading something I've written, I'd like them to think:

They understand the work in practice, not only in theory.
They have a clear view without pretending the answer is simple.
They respect evidence and will say when confidence is limited.
They can explain a technical issue without talking down to me.
I know why this matters and what the next sensible step is.

4. Voice

Grounded. Start with something observable, such as a site visit, an asset, a decision, a number, a conversation or a real constraint. Broader claims should be earned by the example.

Sounds like: "On a spreadsheet, it becomes a line in a register. The asset is physical, ageing, constrained and still carrying consequence."

Doesn't sound like: "In today's rapidly evolving infrastructure landscape, organisations must embrace a holistic paradigm."

Direct. Lead with the answer, point or tension. Use plain language and let short sentences carry weight. Don't make the reader search through a preamble for the conclusion.

Reflective. Show how I reached the view, especially when experience changed my first reaction. Reflection should produce a useful insight, not become autobiography.

Evidence-led. Claims need a source, a concrete example or a clear label as opinion. "I haven't found definitive evidence for that" is stronger than presenting an attractive story as settled fact.

Warm. Write like a trusted colleague speaking to another capable person. Occasional dry humour is fine. Sentimental, chirpy or overfamiliar isn't.

Quietly confident. Have a point of view. Avoid both bravado and meaningless hedging. Use "I think" or "for me" when something is genuinely judgement rather than fact.

5. Mechanics

Language

British English, metric units and GBP unless the context needs otherwise.
Active voice and concrete verbs.
Contractions in conversational writing.
No em dashes. Use full stops, commas, colons or a new paragraph.
Calm punctuation. No repeated exclamation marks.
Sentence case for headings.
Explain specialist terms on first use.

Rhythm

My recent LinkedIn writing averages about 12 words a sentence. Roughly four in ten sentences are eight words or fewer, and about four in five paragraphs are 35 words or fewer. The effect comes from contrast: a fuller explanatory sentence followed by a short conclusion.

Explain the context in one or two clear sentences.

Then land the point.

Don't force every paragraph down to one sentence. The goal is readable rhythm, not a recognisable AI template.

Paragraphs and lists

For LinkedIn and screen reading, one to three sentences per paragraph.
For reports and briefs, short conventional paragraphs under descriptive headings.
Avoid blocks longer than about 100 words unless the subject needs them.
Numbered lists for sequences or ranked sets, bullets for parallel points. Introduce a list with a sentence that explains why it matters.

Questions

Questions are a thinking tool, not engagement bait. A short sequence can sharpen a decision:

What do we know?
What are we assuming?
What would failure mean?
What happens if we wait?

Use a closing question only when it's specific enough to get a meaningful answer. Never close with "Thoughts?"

6. Characteristic moves

These are recurring structures, not catchphrases to insert mechanically. Use no more than one or two in a short piece.

Concrete moment to wider principle. Show the real moment, explain what it revealed, then connect it to a wider question.
Contrast and reframe. Use a clean contrast to correct an oversimplification. For example: "Not perfect data. Honest data that is fit for the decision." Useful, but not for every paragraph.
Necessary, but not sufficient. Respect the conventional tool before explaining its limit.
Name the hard part. State the friction, uncertainty or human cost that a polished strategy tends to hide.
The practical test. Reduce complexity to a question that guides action: "Can we explain why this asset, why now, and what happens if we wait?"
A close that's earned. End with a useful test, a compact conclusion or a specific question. Don't repeat the introduction in different words.

7. What I respond well to, and what I don't

Works for me:

the answer or recommendation first, detail afterwards
clear structure and useful headings
a short summary at the top of complex material
inline source links for research and claims
explicit uncertainty and assumptions in research answers
examples from water, engineering, sport or games when an analogy helps
reusable templates, checklists and dashboards
a small set of well-explained options rather than an unranked long list

Doesn't:

conclusions buried at the end
corporate jargon, inflated language and "idea theatre"
hype about AI or transformation
unsupported facts, fabricated anecdotes or false precision
excessive emojis or exclamation marks
engagement bait and sales-led calls to action
decorative visuals, stock photos or AI imagery that add no meaning
certainty when the evidence is weak

8. What to avoid

Openings

"In today's world…"
"In the age of AI…"
"As leaders, we must…"
"It is more important than ever…"
"We are excited to announce…"
"Here's the twist nobody tells you…"
"Game-changing", "revolutionary" or "the future is here"

Other habits

Telling the reader what they "must" do, unless there's a real safety, legal or process requirement.
Presenting AI as a substitute for engineering or leadership judgement.
Manufactured vulnerability, nostalgia or family stories for effect.
Lone-genius stories. Credit teams, colleagues and inherited knowledge.
A false anecdote as an opening, even if it would make a stronger hook.

9. Tone by context

AI answers. Lead with the answer in one or two sentences. Add a summary for complex subjects, then steps or bullets. Cite research inline, state assumptions and end with the best next step.

Email. Put the purpose or ask in the first paragraph. Friendly and semi-formal, short paragraphs, no more than one list. Make the owner, decision or next step explicit.

Executive brief. Start with the decision or recommendation, then the evidence, the main uncertainty, options and next action. Someone should get the point in two minutes and be able to interrogate the evidence in ten.

LinkedIn post. Open with a real moment or clear observation in the first couple of lines. Explain the insight and connect it to the reader's work. Close with a specific question or a line worth remembering.

Articles. A clear, searchable title. Descriptive headings. Respect the conventional view before showing its limits. Write only as long as the idea needs.

Presentations. Slides carry the structure and evidence, and I carry the explanation. One practical takeaway per talk.

CVs. Factual, evidence-backed achievements. "Led", "delivered" and "improved" rather than "visionary" or "guru". Only use verified numbers.

App labels. Short and verb-led: "Review evidence", "Save draft". Error messages say what happened and what to do next.

10. Writing workflow

Agree the foundation before drafting. What am I actually saying? What's still unknown? What should the reader understand, decide or do?
Surface the forks. If the material could support quite different pieces, ask which one I want rather than picking silently.
Leave good text alone. Don't rewrite passages that are already doing the job.
One clean-up pass, then stop. Check against all the rules in a single pass rather than rewriting once per rule.
Make the minimum useful change. If a rule conflicts with what I actually mean, what I mean wins.
Keep the process out of the result. The finished piece shouldn't mention the guide, the prompts or how it was written.

The clean-up pass

Delete puffery. Say what happened instead of decorating it.
Name the source behind a vague attribution, or remove the claim. "Experts believe" isn't evidence.
Cut filler: "in order to" becomes "to", "due to the fact that" becomes "because", and "it is important to note that" goes.
Cut trailing "-ing" phrases such as "highlighting the importance of…" unless they state a real cause.
Don't force groups of three. Use the number of points the idea actually has.
Use one stable term for a repeated concept instead of cycling through synonyms.
Prefer a plain "is" or "has" to "serves as" or "boasts".
Remove chatbot enthusiasm such as "Great question!" or "I hope this helps".
Replace a generic conclusion with a decision, fact, test or next step.
Split any sentence the reader has to re-read.
Replace unsupported intensifiers such as "significantly" with the actual figure, or delete them.

The specificity test: if a sentence could appear unchanged in an unrelated piece of work, make it specific or cut it.

These checks need judgement. They are not literal bans. Don't manufacture mess or fake personality, and never invent experience, opinions or figures to satisfy a check.

Authenticity check

Is the central view genuinely mine?
Can every personal story be traced to something I supplied or confirmed?
Are facts, quotations and figures sourced?
Is opinion clearly separate from evidence?
Does it sound like a capable practitioner rather than a content marketer?
Would I be comfortable defending every sentence in a room of experts?

11. Design in brief

Design idea: evidence, clearly organised. Hierarchy and space do the work. Colour highlights meaning. Decoration stays quiet.

Principles

Visuals must explain, compare, orient or prove.
Each page, slide or panel has one job.
Size, weight, position and space come before colour or effects.
Real assets, diagrams, maps and data beat stock imagery.
Generous space and restrained colour, with a warm accent where human judgement matters.
Files must print well, stay accessible and survive ordinary Office and browser use.

Core palette

Token	Hex	Use
Ink 900	`#102A43`	Titles and strong headings
Signal teal	`#0B7285`	Links, selected states, key data
Warm amber	`#D9971E`	Small highlights and human judgement
Text	`#17202A`	Body copy
Muted	`#5F6B76`	Secondary text
Canvas	`#F7F9FB`	Page background
Line	`#D8E1EA`	Dividers and borders

Aim for roughly 80% neutral, 15% ink and teal, and 5% amber. A page should still work in greyscale, and colour is never the only way meaning is shown.

Type: Arial for documents and Office slides, Inter for web and apps, a system monospace for code. One sans-serif family per piece.

Layout: an 8-point spacing system, left alignment by default, thin rules, minimal shadows and a reading width of about 65–80 characters.

Signature detail: a short teal line marks the start of a title or key conclusion, once per view.

12. Using it with an AI tool

Task brief

I use this short brief when asking a tool to draft something:

Create: [what]
Audience: [who]
Purpose: [what it should help them understand, decide or do]
My view: [my actual position]
Evidence I can use: [facts, sources, examples]
Constraints: [length, sensitivity, deadline]
Desired close: [decision, action, question or final line]

Review prompt

Review this draft against my voice and style guide. Check authenticity,
evidence, structure, British English, plain language, rhythm, unsupported
claims, hype, em dashes and the quality of the close. Apply the clean-up
pass once. Make only changes that preserve my actual position. List
anything I need to confirm.

13. How I built it

I combined practical writing and prompting advice I'd collected, including from Nate B. Jones's newsletter and other material saved in Recall, with an analysis of my own writing. I used Claude and ChatGPT together over several iterations to build it. Only writing I had produced myself counted as evidence of my voice. Course material, books, manuals and AI-generated reports were used for context only.

What I'm still testing. The guide describes how I want to write. Whether it reliably produces writing that sounds like me is a separate question, and I haven't fully tested that yet.

Version 1.2, August 2026. I review it every six months or after a meaningful change in my role, audience or writing.