AI for medical writing
Draft protocol synopses, CSR sections, patient narratives and lay summaries from the protocol, SAP, TFLs and safety files already in the study folder, and get files back with each figure cited to the table or listing it came from.
Study-735-221-death-narratives.docxExampleStudy 735-221 Clinical Study Report, Section 12.3.2: Narratives of Deaths
Subject 0418-007
A 71-year-old man with stage IV pancreatic adenocarcinoma was randomized to study drug and received his first dose on 12 November 2025 (Day 1). On Day 38 he presented with fever and hypotension and was hospitalized for sepsis from a biliary source, and his biliary stent was replaced the same day. He died on Day 41, and the investigator assessed the event as not related to study drug.
Subject 0503-012
A 64-year-old woman with pancreatic adenocarcinoma metastatic to the liver was randomized to placebo and received her first dose on 3 December 2025 (Day 1). CT on Day 56 documented disease progression, and she entered hospice care on Day 63. She died of disease progression on Day 77, which the investigator assessed as not related to study drug.
CheckedStudy days match the ADAE listing. The site SAE report for 0418-007 gives Day 39, so query the site.
How it works
- 1
Put the study folder together
Drop in the protocol and its amendments, the approved SAP, final TFLs, patient profiles, AE listings, site SAE reports and CIOMS forms, the investigator brochure, and the CSR template with the most recent approved CSR as an example.
- 2
Ask for the section
The request can be a synopsis, the section 12.3.2 narratives, a lay summary or the consent form rewrite. It writes in the template it finds in the folder, cites the TFL or listing behind each figure, and flags what it could not reconcile.
- 3
Trace, then decide
Numbers in the draft link back to their tables, so checking means tracing figures to the listings rather than re-deriving them, and interpretation belongs to the study team. Nothing enters the submission until the reviewer signs off.
Ask it for
“In the M11 structure, draft the synopsis for amendment 2 of protocol HZN-0447-203, and align the estimands with the revised SAP.”
HZN-0447-203-synopsis-am2.docx“Write the section 12.3.2 narratives for the 23 subjects who died in study 735-221, from the ADaM listings, patient profiles and the site SAE reports.”
Study-735-221-death-narratives.docx“Rewrite the key information section of the lupus trial’s consent form, and check the form against the required elements at 21 CFR 50.25.”
Consent-form-v4-key-information.docx“Check every number in CSR section 12.2 against the final TFLs and list each mismatch, rounding difference and value you could not verify.”
CSR-12-2-data-check.xlsx“Triage these 9 site SAE reports: check validity and seriousness, code verbatim terms to MedDRA preferred terms, and assess expectedness against IB edition 6.”
SAE-triage-log-Jul-2025.xlsx“Fill in the CONSORT checklist for the Phase 3 manuscript, point to the section that answers each item, and flag the items the paper leaves unanswered.”
CONSORT-checklist-ms-1429.xlsx
Inside the Medical Writing mode
Clinical trial protocol and synopsis
Drafts protocols and synopses in the M11 structure, keeps estimands in step with the SAP, and summarizes the SoA, visit windows and eligibility for sites.
Clinical study report
Builds CSR sections from the final TFLs and SAP, and carries the same findings into the EU lay summary and registry results.
Ctd clinical summary
For NDA and BLA work, writes Module 2.5 and 2.7 from the pooled CSRs and integrated TFLs, and checks that numbers agree across studies and the source reports.
Adverse event case processing
Triages site SAE reports, codes verbatims to MedDRA preferred terms, checks expectedness against the IB, and drafts each patient’s chronological narrative.
IRB submissions and informed consent
Prepares IRB applications, continuing review and reportable events, and rewrites consent forms so the required elements and key information read plainly.
Scientific manuscript
Writes manuscripts, abstracts and posters to CONSORT, STROBE or the journal’s checklist, and drafts cover letters and point-by-point reviewer responses.
Literature and evidence review
Runs the searches the protocol specifies, counts every record through the PRISMA flow, and grades the certainty of the evidence behind each conclusion.
- Office skills
- Excel, Word, PowerPoint, PDF
- Tools
- Read, edit and write files, Search files, Run code, Search and read the web, View images
- Connectors
- SharePoint, Google Drive, Gmail, and additional custom connectors
- House style
- Your fonts, layouts and rules, on every file
Who it’s for
Regulatory, clinical and publications medical writers at sponsors, CROs, agencies and in freelance practice.
Questions
Does it invent results or references?
No, the skills tell it to cite the source file for every number and statement it writes, and to note what it could not find or reconcile instead of guessing. So a fabricated citation or a result no table supports should turn up as a flagged gap rather than as confident prose. Treat any claim with no source as a question for the study team, and review every file it touched.
Can we put confidential study files or patient data into it?
Ask your sponsor and compliance team before you load the folder. Ultramarine Work changes files only in the project’s folder. When a request needs it, it can read files elsewhere on your computer, never passwords, keys or app data, and you can turn that off in settings. The model runs on Paper Instruments’ servers, so the content it works on is sent there to be processed. Paper Instruments is SOC 2 Type II, and our security and compliance details are on our compliance page.
Do journals require us to disclose AI use?
Yes, journals that follow the ICMJE recommendations ask authors to describe how they used AI-assisted tools in the cover letter and the submitted work, and to keep the tool off the author list. What you disclose is your own workflow, and you write it the way each publisher or congress asks for it.
Will regulators accept a submission drafted with AI?
That is a decision for your sponsor, QA and regulatory leads. Neither the FDA nor the EMA approves AI tools, so no vendor can promise acceptance. What you can put in front of them is a draft whose numbers tie to the TFLs and listings in the folder, plus the QC trail your SOPs require. FDA’s 2025 draft guidance sets credibility expectations for AI models that support regulatory decisions. Drafting documents is a different context.
Will it write a lay summary a participant can read?
Yes, it drafts EU lay summaries from the locked CSR under the EU plain-language rules: language a 12-year-old can follow, short sentences, one term per concept, results as counts like 3 in 10 rather than odds ratios or p-values, and PK findings given as high-level conclusions instead of raw parameters. Have a layperson or an editorial panel read the draft before it goes into CTIS. The sponsor has 12 months from the end of trial to post it, 6 months for a pediatric trial.