Research integrity

Synthesis Lab · last updated 14 October 2026

How Synthesis Lab uses AI in your research, what it checks, what it records, and what stays your responsibility. The short version: the assistant helps, you remain the author, and everything it does to your manuscript is on the record.

You are the author

The assistants in Synthesis Lab draft, edit, check and review at your request. They are tools, not authors: you decide what goes into the paper and you are responsible for its claims, data and citations. Nothing is submitted, published or sent to a journal by Synthesis Lab.

Every AI change is recorded

When an assistant changes your manuscript, whether in the app or through an AI client you connected, the change is logged: who asked for it, which model or client made it, when, the reason it gave, and the exact text before and after. You can read the record under History → Changes, undo any single change, and go back to any saved checkpoint. Co-authors on a shared project see the same record.

Citations are checked against the sources, not against the model's memory

Citation checks read the passage in the cited paper itself and say whether it supports the sentence that cites it, quoting the passage. The report states, in a line written by code rather than by the model, what was checked and which citations had no source to check against, so a gap is never presented as a pass.

When references are added from a reference list or a DOI, the details come from Crossref, arXiv, Europe PMC or your Zotero library. An entry that cannot be found is reported as not found; it is not filled in by guesswork.

AI peer review is a rehearsal, not peer review

Reviewer models read your manuscript (and, if you ask, the cited sources, supplements and analyses) and raise concerns. Every quotation a reviewer makes is checked by code against your text, and a quotation that is not there is flagged. Use these reviews to find weaknesses before you submit; they do not replace review by your field, and they should not be presented as if they did.

Numbers can be traced to the runs that made them

Analyses run in the app (scripts and notebooks) are saved with their code, environment and outputs, so a figure or number in the paper can be traced back to the run that produced it and run again.

Disclosing AI assistance

Many journals and publishers ask authors to disclose how AI tools were used, and do not accept an AI tool as an author. Check your target journal's policy. The change record above gives you what you need to write an accurate statement, for example:

"We used Synthesis Lab, with [model names], to [check citations against the cited sources / edit language / run a pre-submission review]. All AI-suggested changes were reviewed by the authors, who take full responsibility for the content."

Your unpublished work stays yours

Projects are private unless you share them, and only the people you add can open them. We do not use your manuscripts, sources or notes to train models. When you use an AI feature, the text needed for it is sent to the model provider you or the app chose; the privacy policy lists who those are. You can download a whole project as a zip at any time.

Problem reports

If something in the app breaks or gets in your way, the assistant can report it to us so you do not have to. Those reports describe what the app did; one that quotes your manuscript, sources or notes is refused before it is sent. You can turn this off in Settings.

Concerns

If you think Synthesis Lab produced something misleading, such as a citation check that passed when it should not have, tell us at support@synth-lab.com. See also the terms of service.