Editorial process
How the stories on this site are made, including the part most sites leave out.
The short version
The stories here are planned and drafted by AI: the premise, the cast, the plan and the prose. The rules each plan has to obey and the standard every story is measured against are set by the people who run Talehour. Each draft then has to clear a sequence of checks, and a draft that fails a serious one is not published — it is sent back and written again.
No editor reads every story before it goes up. We would rather say that plainly than claim a review that does not happen. Work that falls short is sent back before it reaches anyone, and the rest of this page is how that is decided.
1. The story bible comes first
Before any prose exists, a story gets a bible. It fixes the cast — ages, occupations, what each person wants, and crucially what each person knows and when they learned it — along with the setting and the rules the story may not break.
The bible also carries the limits a draft is held to: how long a story may run, and how much of it has to be dialogue. A story may tighten those limits but never loosen them, so the system that writes a story cannot quietly lower the bar it is judged against.
The plan is itself reviewed before a word of prose is written, by a different model reading it the way this site's readers would: whether the first lines put something at stake, whether each step costs the lead something she can name, whether the climax is earned and happens in front of people who matter to her, whether the ending lands the way the plan promises, and whether any of it could happen to a real woman. A plan that fails goes back to be revised with the reviewer's objection, and one that keeps failing is dropped.
The bible is then checked for contradictions by three independent passes, and a finding counts only when they agree. A bible that contradicts itself goes back once to be fixed. If it still does, the story stops there and nothing is written from it.
2. The story is planned before it is written
The plan sets out the opening line, what is at stake, each step that makes things worse and what it costs the lead, the climax and who is in the room, the proof that settles it, the turn, the resolution, the ending it promises and the last image. The draft fills in that plan rather than drifting wherever the previous sentence happened to point.
3. The draft
A story is written in one piece, from its plan and from the bible. Where a draft has been rejected before, the writer is told what was wrong with the last attempt, in the reviewer's own words. A draft that breaks one of the measurable style rules below gets one repair pass, and the repair is kept only if it leaves fewer faults than the draft had. A separate model then checks the ending: one that settles itself too conveniently is rewritten, and one that still does is held back.
4. Two kinds of checking, and a draft must clear both
The first kind is mechanical, and covers what a machine can be trusted to measure exactly: length against the story's own range, how much of the story is dialogue, sentence rhythm, phrases that read as machine-written, and a comparison against every other story on this site to catch repeated passages. These are not judgement calls, so they are enforced literally.
The second kind is a reading. A different model reads the whole story against the bible and against our published rulebook, and reports what it finds sorted by severity: contradictions of established fact, plot and structure faults, and matters of style.
One rule decides which model does that reading, and it is the rule the whole process rests on: a story is never judged by a model that could have written it. A reviewer marking its own work is not quality control at all. The reviewer runs on a different provider from the writer, and the same principle governs every other reviewing step below.
5. The gate
A story goes forward only on a clean pass — no contradictions, no plot faults, and at most a couple of minor style notes. Anything beyond that is parked and rewritten.
Three things make that gate hard to talk past:
- The verdict is computed from what the review actually found, never from the label the model puts on its own report. A review that says "pass" while listing a serious fault publishes nothing.
- A review that comes back incomplete or unreadable counts as no review at all, and the story waits rather than going out unchecked.
- Every quotation a reviewer gives as evidence is checked against the story. A complaint quoting words that are not there is discarded.
Few drafts fail this final reading, because most problems are caught earlier. One that does is written again.
6. When the story and the bible disagree
Usually the story is wrong. Sometimes the bible is — a date that cannot be true, an age that does not survive arithmetic. So a disagreement goes to a third model, on a third provider, which decides which side is at fault.
If the plan was sound, the story is rewritten and the plan is left alone. If the bible was wrong, it is corrected — but only one named sentence at a time, and only after the system has proved that sentence exists exactly once in the field being changed. Nothing here is ever handed a whole document to rewrite, because then nobody can tell what moved.
7. Before a story reaches the site
A finished story can be technically spotless and still be dull, and no rulebook catches that. So before a story is published at all, a stand-in reader reads it and answers one question: did it hold you to the last line, and would you send it to a friend?
If the answer is no, the story does not go up.
A story that passes gets its cover, a picture drawn by another model and checked for stray lettering before it is used. Only then does the story go up.
8. After publication
Once published, a story is not rewritten by the system. Corrections after that are made by hand, usually because a reader told us (below).
What the people here actually do
They decide what the rules are: the rulebook every story is measured against, the phrases that are not allowed, the limits on length and dialogue, the fixed list of genres a story may be filed under. None of those can be changed by the system that writes the stories.
They set how hard it tries, how many stories it writes a day, and how much is spent. They read the reports the pipeline sends out every day. When something goes wrong — a plan that contradicts itself, a fact that leaked from the outline into the prose, a stretch where the drafting quietly stopped — they go in and fix it by hand.
Corrections
Published stories can be changed, and are. Typos and continuity errors are fixed in place, by hand.
If you find something wrong, tell us: see contact. The story's title and a sentence are enough. Continuity reports are genuinely useful and we act on them.
What we do not do
We do not claim these stories are written by hand. We do not claim they are written entirely by a machine with nobody deciding anything, either — the rules, the standard and the checks are ours, and the machinery above exists because volume without a gate is worthless.
We do not publish a draft that fails the checks, and we do not edit a draft sentence by sentence until a complaint goes away. A story that cannot pass gets rewritten from its plan.
Why this page exists
Readers are entitled to know how what they are reading was made. A site that uses AI and says nothing is asking to be found out; a site that uses AI and overstates the human involvement is worse, because the claim collapses the first time anyone checks it.
So this page describes the process we actually run — including the part where no person reads every story — and the checks that stand in its place.