Spiral

The standard way to make an AI tool sound like you is to describe yourself in a prompt. 'Write in a conversational but authoritative tone, avoid jargon, use short sentences.' Everybody has typed some version of this, and everybody has watched the output come back sounding like a generic writer following instructions — which is exactly what it is.

The problem is that self-description is unreliable. Writers are poor judges of their own habits, and the features that make prose recognisable are mostly ones the writer never consciously chose: how long a sentence runs before it turns, whether a dash or a semicolon does the work, which words appear far more often than they should.

Spiral, built by Every, takes the measurement route instead. It analyses writing you have already produced, extracts those patterns, and applies them. This review covers what stylometry actually does, why the approach is genuinely different, where it breaks down, and who should care.

Why Spiral Is Not Another Copy Generator

• Stylometry, not prompting. It measures sentence length, punctuation habits, word choice and syntax in your existing writing and applies those patterns.
• You supply samples, not instructions, which sidesteps the fact that almost nobody can accurately describe their own writing voice.
• Styles are shareable, so a team can write in one voice rather than five approximations of it.
• Format-aware outputs for tweets, blog posts and LinkedIn, since the same idea genuinely needs different shapes.
• It is a product from Every, and the site claims its own copy came out of Spiral — a demonstration you can judge by reading the page.

Measurement Beats Self-Description

Stylometry is an established field with a long history in authorship attribution — the discipline used to argue about who wrote a disputed text. It works because writing style is measurable in ways writers do not control and often cannot see.

Sentence length distribution, punctuation frequency, vocabulary richness, syntactic patterns, how often you open with a subordinate clause: these are stable, individual, and largely invisible to the person producing them.

Spiral's Style Engine applies this to generation rather than attribution. You give it your writing, it builds a voice profile from the measurements, and it drafts against that profile.

The reason this matters is not that it is more sophisticated. It is that it removes the step everyone gets wrong. You are no longer being asked to describe your voice; you are being asked to supply examples of it, which is something you can actually do.

Nobody can describe their own prose accurately. Almost everybody can produce ten examples of it. Spiral asks for the second thing.

Where The Approach Runs Out

Stylometry captures how you write. It does not capture what you know, what you think, or why you would make a particular argument — and those are usually what makes writing worth reading.

The realistic consequence is that Spiral produces drafts that sound like you saying something generic, rather than drafts that say what you would have said. That is still valuable, because voice is a real editing burden, but it is a narrower claim than the marketing implies.

There is a second limit worth naming: a voice profile learned from your past writing reproduces your past writing. Habits you were trying to break get preserved and reinforced, and a writer whose style is still developing risks fixing it in place. If you already dislike something about how you write, the tool will faithfully give it back to you.

And samples matter enormously. Ten pieces of your best work produce a useful profile. A mixed pile including things you wrote badly, wrote for a different audience, or did not really write yourself produces a profile of nobody in particular.

The Team Case Is Stronger Than The Solo One

Shareable Writing Styles turn out to be the more commercially interesting feature, for a reason that has nothing to do with AI.

Brand voice guidelines fail in a predictable way. They are written once, distributed as a document, read by everybody, and then interpreted five different ways by five writers who all believe they are following them. The result is copy that is inconsistent in ways nobody can quite point at.

A measured style profile is a different kind of artefact: it is derived from writing the organisation actually approved, and it applies the same way for everyone. That is closer to a specification than to a guideline.

Around this sit a knowledge base for reference material and brand documents, real-time collaboration with editorial feedback, and MCP plus API access — meaning Spiral can be driven from Claude Code, Cursor or ChatGPT rather than only from its own interface. The MCP support in particular signals a tool designed to sit inside an existing workflow rather than demanding a new one.

Formats Are Not Interchangeable

Spiral produces format-specific output for tweets, blog posts and LinkedIn, and treating that as a minor convenience misunderstands it.

A thought that works as a tweet has a compressed structure with the tension at the front. The same thought as a blog post needs setup, development and a reason to keep reading. On LinkedIn it needs the particular rhythm that platform rewards, which is neither of the other two.

Most tools handle this by generating once and asking you to trim, which produces the recognisable artefact of a blog paragraph squeezed into a tweet. Generating for the format from the start is a better design, and the stated use cases — social posting, newsletters, meeting recaps, presentations — are all formats where shape matters as much as content.

Building A Profile That Actually Sounds Like You

The quality of what comes out is decided almost entirely at setup, and most people rush it.

1. Collect ten to twenty pieces you are genuinely happy with, rather than everything you have published.
2. Exclude anything ghostwritten, heavily edited by someone else, or written for an audience you no longer address — those dilute the profile with someone else's habits.
3. Keep formats separate where your voice genuinely differs. Your newsletter voice and your LinkedIn voice may not be the same person.
4. Generate a first draft on a topic you know well, then compare it against something you actually wrote on a similar subject. That comparison is the only real test.
5. Read for the tells — sentence rhythm and punctuation habits — rather than for whether the content is right. Content correctness is a separate problem.
6. Add reference material to the knowledge base so drafts carry your facts rather than generic ones.
7. Refresh the profile occasionally, because a voice frozen at your 2024 writing will slowly stop being yours.

The Vocabulary Behind This Product

Three terms carry most of the weight in understanding what Spiral does differently.

Stylometry

Statistical analysis of writing style — sentence length, punctuation, vocabulary and syntax. Long used to attribute disputed texts to authors; here used to reproduce a voice rather than identify one.

Voice profile

The measured pattern extracted from your samples. It captures how you write, not what you know — which is the single most important limit to understand before buying.

MCP

Model Context Protocol. It lets Spiral be driven from Claude Code, Cursor or ChatGPT, so drafting happens where you already work instead of in another tab.

Knowledge base

Stored reference material and brand documents the drafts can draw on. This is where facts come from, since the style profile supplies only the manner of saying them.

How Spiral Is Put Together