Meghan Oates-Zalesky 
Dossier Preparation With AI: What Changes and What Does Not
Dossier preparation with AI changes three things: who drafts the first version, how reviewers verify each claim, and what happens when a source document updates. The structure of an AMCP dossier, a global value dossier (GVD), or an HTA submission stays the same. The expertise behind it stays the same. Instead, what changes is the mechanics of assembling, checking, and maintaining the document.
This article is for HEOR analysts, medical writers, and market access leads who own those documents, as well as department heads accountable for their timelines. It explains how AI-enabled dossier authoring works, the benefits it offers, the key features to consider when evaluating a platform, and the practical constraints to keep in mind.
Why Does Dossier Preparation Still Take Months?
Dossier preparation takes months because the evidence is scattered, the authorship is split, and every update restarts the cycle. However, none of these is primarily a writing problem. Rather, each is a coordination problem.
- Evidence lives in many places. Clinical study reports, FDA labeling, literature reviews, economic models, and real-world data each sit with a different owner.
- Four functions own one document. HEOR writes the economic sections, clinical or medical writers own the clinical evidence, market access owns payer positioning, and medical writing owns the narrative. Drafts move by email and version names multiply.
- Reviewers verify citations by hand. For every claim, someone opens the source, finds the page, and confirms the passage.
- A single source change ripples everywhere. A new data cut or a revised budget-impact model forces someone to find every dependent section and rewrite it manually, then restart review.
Some organizations outsource the whole exercise to specialist agencies. That moves the cost but not the mechanics. The evidence generation services pharma teams rely on can shorten the literature work that feeds a dossier, but the assembly and maintenance of the document itself has stayed stubbornly manual.
What Do Payers and HTA Bodies Expect From a Dossier Today?
Payers and HTA bodies now expect dossiers that stay current, cite transparently, and arrive on fixed timelines. However, a manual process struggles with all three.
The AMCP Format for Formulary Submissions, Version 5.0, published in 2024, frames the dossier as a living document that evolves across the product lifecycle, with real-world evidence and health-disparity information receiving explicit attention. In Europe, the EU Joint Clinical Assessment under the HTA regulation began in January 2025 for new cancer medicines and advanced therapies, with manufacturers submitting clinical evidence within 100 days of the first request.
For a deeper look at the evidence side of these requirements, see how AI-powered SLR supports AMCP dossier development and how HEOR teams build HTA-ready evidence dossiers.
How Does AI-Enabled Dossier Authoring Work?
AI-Enabled Dossier Authoring: From evidence to a traceable, structured report
AI-enabled dossier authoring works by turning the dossier into a structured object rather than a file. Each section carries its own instruction and its own evidence; the platform generates content per section from that evidence, and every generated statement keeps a link back to the passage that supports it.
What Is a Living Dossier?
A living dossier is a report whose sections know which evidence they depend on. MadeAi’s Dossier Expert, a report starts from a saved template, for example, an AMCP dossier or a GVD, or from a blank structure, and the author defines sections and subsections in a tree.
Each section has two inputs. The instruction describes what the section should cover and doubles as the generation prompt. The evidence is the set of documents attached to that section, either pulled from an indexed Evidence Hub or uploaded directly as PDF, Word, Excel, or slide files. Generation runs section by section or across the whole report, and the output can include paragraphs and tables.
A typical workflow has six steps:
- Create the report from a template or a blank structure and assign a category.
- Define or adjust the section tree.
- Write an instruction and attach evidence for each section.
- Generate a section, or generate the full report.
- Review the draft, follow citations to sources, comment, and edit with collaborators.
- Approve and export the final report.
How Does Dossier Preparation With AI Keep Claims Traceable?
Dossier preparation with AI keeps claims traceable by attaching a citation to every statement that draws on evidence. A section generated with no evidence attached produces no citations, which is itself a useful signal during review.
Selecting a citation marker opens a source panel that shows the supporting quotation, the document name, and the location. For PDF sources, the document opens beside the report at the cited page with the passage highlighted. When a statement draws on several passages, each quotation appears separately. Citations stay available even if the source document is later removed from the section.
This turns citation checking from a page-by-page search into a one-click confirmation. The reviewer stops asking where a statement came from and starts asking whether the source supports the wording, which is the question that requires expertise.
One caution: if a reviewer edits a generated statement, the citation still points to the original passage, which may no longer match the edited text. The platform shows an Edited indicator, but the human must re-verify the claim.
What Happens When a Source Changes?
Similarly, when attached evidence changes after generation, the affected section is marked Outdated, and its existing text and citations stay unchanged until someone regenerates it. Nothing is silently rewritten.
Generate everything reprocesses every section. Generate what is pending option touches only sections that have never been generated or that carry the Outdated flag; current sections stay as they are. A summary then reports completed, skipped, and failed sections.
Regenerating a section replaces all of its content, including manual edits. In practice, that means teams should capture substantive edits as comments or as instruction changes rather than as untracked text changes, so the next regeneration carries them forward.
Who Benefits, and How?
The four functions that co-author a dossier each gain something different, and the buyer gains a shorter, more predictable cycle.
| Role | Task | What changes |
|---|---|---|
| HEOR analyst | Economic value sections, model inputs | Attach the model and data cut to the section; regenerate when the cut changes |
| Clinical or medical writer | Clinical evidence sections | Generate from CSRs and labeling with citations already in place; spend review time on interpretation |
| Market access strategist | Payer-facing positioning | Work in the same document as evidence owners; comment and resolve rather than email |
| Medical writing lead | Executive summary and narrative | Draft from approved sections; keep the narrative consistent across variants |
| Market access or HEOR head (buyer) | Timeline, quality, budget | Visible progress states, exportable activity logs, fewer restarts |
Beyond the AMCP dossier, the same mechanics apply to GVD, HEOR reports, and HTA submissions, which is why the market access solutions a team chooses should treat dossier authoring and evidence generation as one connected workflow rather than two.
What Are the Risks and Limitations?
The approach still requires human oversight. Generated text must be reviewed for accuracy and clinical appropriateness. Hand-edited content that later diverges from its original citation needs re-verification. Templates and evidence attachments must be maintained carefully; poor instructions produce lower-quality drafts. Organisations must also ensure that the platform meets internal security and data-governance requirements before ingesting proprietary clinical and commercial documents.
Ultimately, no system eliminates the need for expert judgement on interpretation and positioning. The value lies in removing the mechanical burden so that experts can apply judgement more efficiently.
What Comes Next for AI in Dossier Preparation?
The next step is a dossier maintained continuously rather than refreshed in cycles. The pieces already exist: living SLRs, section-level dependency tracking, and selective regeneration. Three developments look likely:
- Shared evidence bases across variants. One indexed evidence set feeding an AMCP dossier, a GVD, and a JCA submission, with each variant citing the same passages differently.
- Tighter coupling between literature review and authoring. A new included study flows from the review into the Evidence Hub and flags the sections that depend on it.
- Standardized disclosure of AI use. As HTA bodies formalize expectations, the activity log and citation trail become the submission’s methods record.
Life science AI will not remove the judgment calls that make a dossier persuasive: which comparator, which framing, which caveat. It will remove the time spent finding, formatting, and re-finding the evidence those judgments rest on.
Conclusion
Dossier preparation with AI, implemented as a living, section-structured and fully traceable process, directly addresses the fragmentation, version chaos and update burden that characterise current practice. The combination of click-to-source citations, Outdated flags, hand-edit protection and in-document collaboration converts a multi-month manual cycle into a controlled, auditable workflow measured in days. Teams retain full responsibility for content accuracy while removing the mechanical overhead that previously consumed most of their effort.
For market access and HEOR leaders evaluating process improvements, the practical next step is to map one existing dossier type against the living-document workflow and measure the reduction in hand-offs, citation verification time, and update effort. Further detail on related evidence practices appears in evidence generation for AMCP dossiers and HTA-ready evidence dossiers.
FAQs
How does AI change dossier preparation?
AI automates initial drafting and coordinates evidence across cross-functional teams while streamlining citation verification by linking claims directly to source materials in real time. Authors efficiently update living documents, leaving subject-matter experts with full strategic control.
What is a living dossier?
A living dossier directly links individual sections to underlying evidence files and prompts while tracking data dependencies. Teams instantly identify content requiring updates after new data cuts and selectively regenerate target sections without overwriting manual edits.
How does AI maintain claim traceability?
The system generates precise citation markers for every statement supported by attached evidence. Reviewers click citations to view text quotations and highlighted source pages side-by-side, converting manual verification into a rapid, one-click process.
What happens when a source document updates?
The platform flags affected sections with an Outdated indicator whenever source evidence changes while preserving existing text, citations, and manual edits. Authors then execute targeted updates to regenerate only the pending sections.
Does AI replace human reviewers?
No, AI handles mechanical compilation tasks rather than executive strategy or clinical interpretation. Human experts actively evaluate narrative positioning, verify claim accuracy, and approve final submissions while the platform frees them to focus on scientific judgment.
Which dossier types benefit most from AI authoring?
AMCP dossiers, GVDs, and HTA submissions benefit significantly from the platform. HEOR reports and value-evidence materials also leverage the shared evidence architecture, helping any cross-functional document gain speed and traceability advantages.
