Author interview
PerfectPaper asks targeted questions about design decisions and fixed constraints before review, then carries your answers into the critique.
SOLUTIONS
Check three things before uploading an unpublished manuscript to an AI tool: your plan's training default, its retention window, and your institution's policy.
Affiliations
Current platform
PerfectPaper now carries context from setup through research, revision, and export—without turning the paper into a generic writing prompt.
Prepare
PerfectPaper asks targeted questions about design decisions and fixed constraints before review, then carries your answers into the critique.
Search the journal catalogue, choose up to three targets, and compare compatible open-access journals before the review starts.
Brief up to three custom reviewers, declare ground truths, attach instructions, and choose standard or deep-research depth with specific tools.
Investigate
Specialist reviewers inspect the full paper in context, including figures and tables—not isolated paragraphs.
Deep-research reviewers can search the web and scholarly literature, inspect sources, and attach vetted citations to research-backed findings.
The reading room shows which review areas are working, which findings have arrived, and when a research step could not complete.
Revise
Move between each comment and its passage, read your paper as you wrote it in Word, filter feedback, and discuss any finding.
Preview suggested revisions, apply accepted changes, keep an edit history, and reverse a change without losing the review trail.
Export the revised paper and saved feedback as DOCX, annotated PDF, or print view, and prepare an anonymous copy for blinded review.
It depends on which product and which plan you are signed in to. Consumer chat products, paid consumer tiers, and API or enterprise plans differ from one another on data retention and on whether your content may be used to improve future models, and their defaults differ too. Before uploading an unpublished manuscript to any general-purpose AI product, check three things: the training default for your specific plan, how long conversations are retained, and whether your institution permits that service for unpublished research.
Those three checks are the whole answer. The rest of this page explains what each one means and where to find it.
Whether your content can be used to improve a model is usually a per-plan setting rather than a per-company fact. The same vendor often applies one default to a consumer account and a different one to an API or enterprise account, and the control may be a toggle you have to find rather than a default you inherit.
Read the terms attached to the plan you are actually signed in to, and check the setting rather than assuming it. This is the item researchers most often get wrong, because a colleague’s accurate description of their own account may not describe yours.
Retention is a separate question from training, and the two are routinely confused. A service may commit to not training on your content while still storing that content for a period for abuse monitoring, support, or legal reasons.
For an unpublished manuscript the question worth asking is how long a copy exists after you delete the conversation, and whether deletion happens immediately or on a schedule.
Most universities and research institutes now have written policies on third-party AI services, and a growing number name specific products as approved or not approved for unpublished research data. Research offices, IT security teams, and grant compliance teams may each hold a view, and a grant’s data management plan may constrain the answer further.
Checking takes one email and settles the question more definitively than any vendor page can, including this one. If you are the person being asked, the procurement checklist lists the questions worth putting to any vendor.
Whether it is safe to upload and whether you must disclose are separate questions with separate answers.
Journal policies for authors typically permit AI assistance and ask for a disclosure statement. The confidentiality rules that prohibit uploading a manuscript outright are usually aimed at peer reviewers handling another author’s work, not at authors handling their own. Confusing the two leads researchers to avoid tools they are permitted to use — journal AI policies for authors and referees covers the distinction in full.
PerfectPaper never uses your content to train a model — not the paper, not the model’s reasoning, not its reply, not the derived metadata. Manuscripts are stored in EU object storage and can be deleted from your account, with a self-service path for erasure requests. Every processor that receives your text is named publicly along with what it receives, so an institutional reviewer can evaluate the chain without having to ask.
The wider picture is on secure AI review for unpublished research, including what PerfectPaper does not claim.
The product permits it. Whether it is advisable depends on your plan’s training default, its retention window, and your institution’s policy on third-party AI services for unpublished research data. Check all three before uploading.
No. Sending a manuscript to an AI service is not publication and does not create a public record of the work. Prior publication concerns public dissemination, not processing.
A manuscript under review is still confidential. If you are the author you may generally use AI assistance on your own work subject to disclosure. If you are the assigned referee, most publishers prohibit uploading the manuscript to any external service.
Staff access is normally limited and governed by the provider’s terms, though some providers reserve limited human review for safety or abuse investigation. The specific commitment appears in the terms attached to your plan.
A service whose terms state that customer content is not used for training, that names its subprocessors, and that gives you a deletion path. Those three properties can be checked in advance. General assurances cannot.
Last updated September 9, 2026
Upload your paper and receive structured, sourced feedback before you submit.
Connection lost
Reconnecting…
Something went wrong on our end
Attempting to reconnect