You finished the chapter. There's nobody to ask.
Get a few reader personas to read it through and tell you which part held them and which part lost them — with the line where each one stopped.
Get started nowThe hard part after finishing a chapter isn't revising, it's not knowing what to revise. Hand it to a friend and you usually get "I liked it"; post it to a forum and you may get nothing for three days. LitMemo's AI beta reader report has several reader personas read the chapter end to end, then tells you three things: how many of them want to keep reading, the one passage they all wanted more of, and the exact line where each of them stopped. Every finding carries your own text with it, so one click puts it back on the manuscript to revise against. It deliberately does not give you a 1-10 score — we measured that score ourselves and it doesn't hold up. That's explained below.
"It's done… but is this chapter any good? I've read it so many times I can't tell anymore."
You're too close to your own draft — by the seventh pass, the first-time experience is gone. And the people you could ask are either busy or will just say "it's good." What actually helps is "I stopped reading here," and nobody puts it that way.
The core features of AI Beta Readers — Get a few readers on the chapter you just finished
It opens with a verdict, not a scorecard
The first line of the report is "4 of 5 readers want to keep going" — a sentence you can act on immediately, not a set of numbers you have to interpret. Below it: the passage they all caught on, each reader's own words, and whether each one would keep reading. There is deliberately no 1-10 score and no 0-100 engagement number per passage. Numbers like that look precise and can't be checked, and a number you can't check just sends you revising the wrong part of the chapter.
- Opens with how many readers want to continue
- Quality as a band — high/medium/low, never a decimal
- Every conclusion carries your own text
Scoring and personas are separated (this is the foundation)
If one pass both plays the reader and assigns the score, the score follows the persona — write a reader who "never gives low marks" and you will get high marks. So manuscript quality is judged by a stage that **cannot see the persona at all**, computed once per report and shared by every reader; the readers only describe what they read and where they stopped. Tested against four adversarial personas (superfan, in-fandom, prompt injection, over-specific), the quality fields came back identical to the system readers. You can shape the reader. You cannot shape the verdict.
- The quality pass never sees the persona
- Custom readers change reactions, not quality
- Honesty rules sit after the persona and outrank it
Then take it straight back to the manuscript
The least useful shape for a report is a PDF you read once and close. The main action here is "go see where they caught" — each finding anchors back into the editor by **quotation**, so you revise with the reader's reaction beside the line. Anchors follow the quote rather than a paragraph number, so inserting a passage earlier and shifting everything after it doesn't break them. And once you've rewritten that passage, the finding says so and retires itself instead of pointing an old reaction at new text.
- Findings anchor into the editor, revise in place
- Revised passages retire themselves — no stale advice
- Hand any single finding to Sumi to dig further
Have you run into this too?
「You just finished a chapter and want to know if the opening holds, with nobody around to ask」
Pick the chapter, pick three to five readers, and a few minutes later you have "how many want to keep going" plus the line each of them stopped at. No waiting on a friend's schedule, no posting an unfinished draft in public.
「You've done a revision pass and want to know whether it actually landed」
Run the same chapter again and set the two reports side by side. Last time three of five wanted more at the same passage — how many this time? Reports are kept with their timestamps, and before-and-after comparison is exactly why this is designed to be re-run.
「The chapter is long and you suspect the AI only read the opening before commenting」
When a chapter runs past what one report can read, it tells you up front: "this will run as 2 reports, about 20 credits." You get 2 reports back and the whole chapter genuinely gets read — instead of a quiet skim of the first stretch written up as if it were complete.
The usual way vs LitMemo
| The usual way | LitMemo | |
|---|---|---|
| What you get | A number out of 10 | How many keep reading + where each stopped |
| Score reliability | Re-run the same draft, the score moves | No score — a band you can verify |
| Custom readers | Write a lenient one, get high marks | The quality pass can't see the persona |
| Long chapters | Quietly reads only the opening | Says how many parts up front, reads all of it |
| After you read it | A report you close and forget | Findings return to the draft; revised ones retire |
Get started in four steps
- 1
Pick a chapter
Tick one from the list — you can see exactly what will be read
- 2
Pick readers
Three readers with different tastes are set up for you; swap if you want
- 3
Run it
Cost and part count sit next to the button; you can leave the page
- 4
Take it to the draft
You'll be notified; open the report and put findings back in the editor
Frequently asked questions
Two things: it doesn't give you a score, and its quality judgement can't see the reader persona. Ask a general AI "how good is this chapter, out of 10" and you get a number — one that moves when you re-run it and moves again when you rephrase the question. In our own testing, a single persona gave five completely different manuscripts the identical score. The beta reader report answers checkable questions instead: how many readers want to keep going, where each of them stopped, and which line it was. You can go back to the draft and verify all of it yourself.
You can build readers on the MUSE plan — duplicating an existing reader and editing it is the fastest route. But you shape reactions, not the verdict: manuscript quality is judged by a stage that cannot see the persona, computed once per report and shared across all readers. Tested against four adversarial personas — including one that stated outright it never gives low marks, and one attempting prompt injection — the quality fields matched the system readers exactly. The report also marks which readers are your own.
It never quietly reads only the beginning. When a chapter runs past what a single report can cover, the setup page tells you first how many parts it will run as and what the total costs. You get that many reports back, and each one states which stretch it read and how much of the chapter that is. Splitting exists because a report is one continuous read — too long and the middle gets skimmed. That's a quality ceiling, and no plan raises it.
It scales with the number of readers — roughly one to three minutes for three. You can close the tab: the run continues in the background and sends you an in-app notification when it finishes or fails. Reports stay in the project's beta reader list with their timestamps, so you can come back whenever. Run the same chapter several times and you keep all of them — comparing before and after is a supported use, not a side effect.
Wherever the pen rests, there is home
Don't let cracks in the details
ruin your good story
Start now, and scale up as your work grows