Manuscript screening for literary agencies: a second view, not a gatekeeper

Why automated rejection is the wrong shortcut

Literary potential is not the same as error-free prose. An unusual text may be interesting precisely because it breaks expectations. A smooth text may be technically competent and still lack a distinctive identity.

A machine used as the sole gatekeeper would risk disadvantaging the deviations that make new voices visible.

Where a structured second view helps

• Compare names, characters, and timelines across a manuscript
• Flag unclear motivation or missing cause and effect
• Collect recurring craft patterns
• Prepare questions for an internal manuscript discussion
• Create a Word file with tracked changes for further review

A responsible agency workflow

1. Separate formal criteria from list and representation criteria
2. Create a second view only for submissions that merit deeper reading
3. Read observations as questions, not verdicts
4. Evaluate originality, voice, and market position humanly
5. Only then decide on a request or representation

Data protection for submissions

Agencies process unpublished work that belongs to others. Legal basis, notices, contracts, and service providers must therefore be clarified before using a tool. EPOS-AI stores permanent project data encrypted on servers in Zug and Zurich. Individual AI requests are transmitted to a commercial model provider.

What a useful screening report should contain

An agency does not need a score that pretends to judge a manuscript objectively. A structured second view is useful when it prepares concrete questions for the reading process. A report should separate observation from evaluation.

A responsible submission workflow

1. Check formal submission requirements without AI first.
2. Process only manuscripts covered by clear consent and a defined purpose.
3. Do not replace the first human reading with a score.
4. For promising work, examine continuity, character logic, and open dramatic questions as a second view.
5. Trace every finding back to a concrete passage.
6. Keep rejection and representation decisions entirely human.

This keeps software in the role of a research tool. It does not filter people out and does not decide which voice deserves a chance.

Author voice in the agency process

Agencies are not looking for text that is merely clean and average. They are looking for a voice that carries something of its own. That is why smoothing during screening is risky. EPOS can build a voice profile from the manuscript and compare language suggestions with sentence rhythm, syntax, word choice, and deliberate punctuation.

In the documented measurement, the profile closed an average of 73 percent of the measurable distance to the author's own writing. The range was 58 to 88 percent. An independent blind study with multiple authors is still pending. These limits belong in the assessment. Read the full method and results.

Clarify consent and data protection in practice

A submitted manuscript is confidential material. Define who may upload files, how long projects remain, how deletion is documented, and what information authors receive at submission. Permanent project storage and the processing of individual AI requests should be assessed separately.

EPOS stores permanent project data in encrypted form on servers in Zug and Zurich. Content required for individual AI requests is transmitted to an external commercial AI service for processing. Include this distinction in your own privacy review and submission agreement.

More B2B guides

Conclusion

A literary agency should not ask an AI whether a book is good. The useful question is: which specific places deserve additional human attention?

Explore Pro Editing for literary agencies.