Slima 4.0 counts three things separately as you write: AI writes, pastes and imports, and AI consultations. Each channel keeps its own count and word total, recorded at the moment it happens rather than reconstructed afterwards. Around that sit Creation Playback (the whole project replayed along its timeline), a keystroke-level replay that walks through how a passage was built with AI-written parts marked in violet, a public journey page with an embeddable badge, and a provenance certificate carrying a hash-chain anchor (SHA-256) and a server timestamp. All of it is a record of process, and a basis for a declaration. None of it is a verdict on who wrote any particular sentence.
The rest of this post is why that distinction matters.
The disclosure box arrives long after the manuscript
Platforms have started asking authors to declare how AI was involved. Amazon KDP asks at upload. Contests and awards ask in the entry form. Publishers put a clause in the contract. In Japan, Shosetsuka ni Naro began requiring a declaration of AI usage category for submitted works on 1 September 2026, and Kakuyomu asks its own version of the same question.
The awkward part is that the box shows up about six months after the writing did.
You remember the night you were stuck and talked it through at length. Chapter three’s opening was maybe a draft you asked for and then rewrote end to end. The batch of interview notes went in sometime last spring. You can get that far from memory, and then you have nothing at all to decide which category any of it belongs in.
There are two ways to get it wrong. Understate it and you cannot explain yourself later. Overstate it and you have discounted a book you actually wrote. Both failures come from the same source: you are filling in the form from an impression.
The three channels do not add up to one number
Slima counts three things separately while you work.
One week on one book
There is a reason they are never summed.
A consultation is a trace of you thinking. You can spend ten evenings asking the AI where the structure sags and still have typed every word in the manuscript yourself. A paste or an import might be interview notes, or a draft you wrote in another editor last year. Only an AI write is the moment text that you did not type appeared in the editor.
Collapse all three into a single “how much of this book is yours” figure and every one of those distinctions disappears, while the number that survives looks impressively precise. So Slima does not produce that figure. It cannot be made honestly.
Watching it grow, and watching how it grew
The authorship record has two kinds of playback.
Creation Playback runs the whole project along its timeline. Files appear, chapters lengthen, reference material starts branching. Three months of work plays back looking like three months of work.
Keystroke-level replay is the close-up. Pick a passage and step through how it was assembled, with anything the AI wrote marked in violet. Which lines you typed and which you did not is something you can see rather than recall.
Replay currently supports plain-text manuscripts only. Screenplays, world maps and timelines are not verbatim text, so they are out of scope, and the interface says so plainly instead of showing you an empty player. A record that names its own limits is worth more than one that does not.
How much you built around the book
Version 3.8.2 added Reference Material. It counts what you built for this book: how many text files of setting, worldbuilding and notes, how many words in total, when you started, and how many relationship maps, world maps and timelines the book carries.
Only counts and types appear there. No file names, no contents. Whether it shows on your public journey page is a toggle, and it is off by default.
This matters for disclosure because most platforms are really asking where AI sat in your process. The six months you spent stacking up reference material live outside the manuscript, but they are unmistakably part of the work.
For readers, or for a submissions desk
The public journey page is the reader-facing version. Send the link directly, or put the embeddable badge on a profile or a blog; the badge links back to slima.ai.
The provenance certificate is for the more formal cases. It issues a publicly verifiable record of your creative history, carrying a hash-chain anchor (SHA-256) and a server timestamp. It is included for every work on Pro and Team, and can also be bought per book.
Certificates are permanent and cannot be revoked, because not being revocable is the condition for a certificate meaning anything. The trade-off is that all a certificate offers is a verification link.
Disclosure guidance is built into the report itself, mapping what Kakuyomu, Shosetsuka ni Naro and KDP each ask. Translating one record into three different phrasings happens there.
Everything you have made public, in one list
The Privacy Centre has a section called “What you have made public”: shared reading links, public journey pages and provenance certificates in one list, each openable or revocable on the spot. Certificates, as above, cannot be revoked, so that row shows a verification link only.
The name on your public page and your certificates comes from the public byline you set. Leave it empty and your account name is used. Sharing a work now shows the name that will actually be used rather than pre-filling your account name.
When you actually fill in the box
The sequence goes like this. Open the authorship record for the book and three stat cards sit at the top: AI writes, pastes and imports, and AI consultations, each with its own count and word total. Below that is “Words changed per save”, which separates the saves where a lot appeared or disappeared and marks whether it was an AI write or an import. Click one bar and that day opens up save by save, down to the time of day and the word delta. Where a save was never measured it says so rather than drawing a zero. Below that sits Reference Material.
Only after all of that does the disclosure guidance make sense. It will not decide which box to tick for you; that is the platform’s rules plus your judgement. What it does is smaller: it turns “what did I actually do here” from a question you answer from memory into one you answer by looking.
What it does and does not do
This part needs to be exact.
These are records of process, and a basis for a declaration. They are not a verdict on how any given passage came to exist, and we will not claim otherwise.
The timestamp is stamped by our server, not by a third-party authority. What the record shows is that this book took real time and real revision to reach the shape it is in. That is a strong piece of material, and it stops there.
Which means the answer you put in the disclosure box is still yours to decide. What Slima can offer is that when you decide it, you are looking at numbers rather than at your memory of last spring.
- What each file type is for: Writing Studio
- Why this became a real question in the first place: When “Did a Human Write This?” Stopped Being Rhetorical