Why We Built an AI-Powered Writing Style Analyser (And How It's Changing Client Communications)

When you're managing communications for multiple executives, voice consistency becomes a scalability problem. You're not writing for yourself — you're writing for people whose communication patterns you need to understand deeply enough to replicate accurately.

The traditional approach involves extensive briefing, reading previous work, and a lot of iteration. It works, but it doesn't scale. The more executives you're writing for, the more cognitive overhead, and the more variation creeps in as context switches happen faster than your mental model updates.

We've been thinking about this problem for a while. Our answer was to build a tool that does what a very experienced communications professional does — but systematically, at speed, without the context-switching penalty.

Our Approach: Reverse-Engineering Communication Patterns

The writing style analyser takes text samples from a person — LinkedIn posts, articles, email excerpts, whatever's available — and builds a structured profile of how they communicate.

This isn't about surface features. Anyone can tell you whether someone writes in short sentences or long ones. What we're after is the underlying logic: how they structure arguments, what kind of proof they reach for first, how they signal expertise without stating it directly, where they use hedging language and where they don't, what topics they treat as settled and what they treat as open questions.

These patterns are stable over time. People's communication style evolves, but the fundamental architecture of how they think on paper is remarkably consistent. Once you've mapped it accurately, you have something genuinely useful: a profile that tells you not just what this person sounds like, but how they think.

The style dimensions that actually matter:

Sentence Structure and Information Flow

Does the person front-load their key claims or build to them? How do they handle complex information — do they simplify aggressively or trust the reader to follow nuance? How long are their sentences on average, and how much does that vary?

Epistemic Positioning

How confident are they in their assertions? Do they qualify heavily or state things flatly? When they're uncertain, do they say so explicitly or hedge through word choice? This dimension affects trust and credibility signals in ways that are hard to fake if you get it wrong.

Evidence Patterns

What counts as proof for this person? Do they reach for data first, or examples, or analogies? How do they handle counterarguments — do they address them directly or implicitly? Understanding this lets you anticipate what kind of content they'll find credible and what they'll push back on.

Vocabulary and Register

Not just whether someone uses jargon, but which jargon, and in what contexts. Whether they code-switch between technical and accessible language, and how they signal the transitions. Whether they use humour and how — dry observation, self-deprecation, absurdist tangents.

Structural Markers

How do they signal transitions? What phrases do they use to introduce examples, caveats, conclusions? These tics are highly individual and often subconscious. They're also some of the most effective voice markers when reproduced accurately.

What This Enables in Practice

The immediate application is ghostwriting support. When we're producing content for executives, we run their samples through the analyser before we start. The profile gives our writers a concrete reference point rather than a vague brief to "sound like them."

This matters because the failure mode in ghostwriting isn't usually getting the facts wrong. It's getting the register wrong — the level of formality, the confidence markers, the structural choices. Content that's technically accurate but stylistically off tends to get heavily edited, which creates friction and slows down the production cycle.

With accurate style profiles, first drafts come back closer to final. The revision cycle shortens. More importantly, the executive doesn't have to do as much of the work of correcting for voice, which is often the most frustrating part of the process for them.

The secondary application is consistency checking. When you're producing high volumes of content for someone, it's easy for variation to creep in — especially across different writers or over time. The style profile gives you a benchmark to check against.

The Limits of Style Analysis

It's worth being clear about what this doesn't do.

Style analysis captures patterns, not judgment. It can tell you how someone typically structures arguments; it can't tell you whether a particular argument is one they'd actually make, or whether a specific claim is consistent with their actual views. That contextual knowledge still has to come from human understanding of the person and their domain.

It also has limits with sparse data. The more text samples you have, the more accurate the profile. With minimal data, you're extrapolating, and extrapolation introduces noise.

And there are writers whose style is genuinely inconsistent — who write differently in different contexts in ways that aren't just code-switching but reflect actual incoherence in their communication patterns. The tool will capture whatever patterns exist, but if the underlying communication is inconsistent, the profile will be too.

Why This Matters for B2B Communications

Most B2B thought leadership is generic. It hits the expected points in the expected order with the expected level of confidence. It's indistinguishable from every other piece on the same topic.

The executives who cut through aren't just saying smarter things. They're saying things in a way that's recognisably theirs. The style is part of the signal. It tells readers that this is a real perspective from a real person, not a committee-produced press release dressed up as opinion.

Getting voice right is what makes the difference between content that builds genuine credibility and content that ticks a marketing box. The analyser is a tool for getting there faster, with less iteration, and with more consistency over time.

We're continuing to develop it. Current focus is on improving accuracy with shorter text samples and adding comparison functionality — so you can see how two different executives' styles relate to each other, which matters when you're producing content for a leadership team that needs to have distinctive voices while maintaining a coherent company narrative.