IT SOUNDS LIKE
The author's writing style as something you can work with
In IT SOUNDS LIKE a person builds a style card of their own writing and works with AI so that the text reads as if they wrote it. The service works with written text.
We check whether a writing style can be described precisely enough for AI to follow it, and whether the person stays the author while it does.
- Status
- Prototype
- Direction
- Perceive
- For
- Authors, editors, studios, creative and communication teams
- Material
- Written text
- Core object
- Style card: a compact structured description of an author's writing style
- Success measure
- Blind tests: can the user tell generated text from their own
- Access
- By request
Status
Prototype. A limited demonstration is available.
For Whom
Authors and creative teams who want help from AI that keeps their writing style.
Specialists who build a persona of a professional (an editor, a lawyer, a strategist) and want the persona's writing to be predictable and checkable.
How it can work
Situation. Generated text is usually recognizable as generated: smooth, averaged, with someone else's rhythm. The longer the prompt that describes a style, the less reliably the model follows it.
System. A chat with three steps (identity, education, skills) builds a compact style card. After each turn the system writes a sample text in that style. The user rates it, the rating shifts the style vector, and the vector rebuilds the prompt.
Human role. The person supplies the material, judges every sample, and decides what counts as their style.
What we study
What stays the author's when a writing style becomes an object that can be edited. We look at the gap between intention and result. Rating samples may sharpen a person's sense of their own style. The opposite risk also exists: the person starts to aim at what the model can reproduce.
How it works
Style card. A small typed record with who the author is and for whom they write, the territory they have the right to write about, their skills, a style vector on eight axes (warmth, formality, abstraction, density, certainty, rhythm, irony, conversational tone), a few anchors, that is, passages from the person's own writing with a note on what matters in each, short rules, and lists of phrases to avoid and to prefer.
The prompt is compiled from the card by a deterministic function. The budget is under 1,800 characters, against more than 22,000 in the earlier layered version.
Measure. The share of blind tests in which the user can tell generated text from real text should fall as the card improves. We report it as a research indicator.
Teams of personas. On top of the style card, the studio lets a person appoint personas as specialists to tasks within a project. The person stays the copilot: appoints, introduces the task, and may close the chat so that the persona continues without them. The result waits for their review.
Limitations
A style card is built from the user's own texts. The project is not designed to imitate another person's writing without their consent.
The tool may sharpen an author's style or blur it. That is the thing we measure, so we claim no result in advance.
A persona is not a person and not a licensed professional. Responsibility for any text that leaves the studio stays with its author.
Access
Access is by request. Terms of use are agreed before you start working with the project.
HandS Lab research tools and prototypes are not a catalog of commercial products. If a specific applied solution involves commercial deployment, it is carried out by an external operator or partner.