A regulatory affairs lead for a first-in-human rare-disease program has forty-five days until a Type B pre-IND meeting with the FDA, a sponsor team that wants AI to draft half the briefing book without anyone quite saying so out loud, and a deadline that is not the meeting date. The deadline that actually governs her life is the day the briefing package is due to the agency, which lands no later than thirty days before the meeting, which means she has roughly fifteen days, not forty-five, to produce a complete, defensible Type B briefing book filed into Module 1.6 of the eCTD with its appendices. She opens the enterprise large language model and asks it to draft a briefing package from the program's nonclinical data, the proposed clinical protocol synopsis, the prior FDA interactions, and the three questions the sponsor wants to ask. A structured draft appears in under a minute: background, product description, nonclinical summary, clinical development plan, the specific questions, and the sponsor's positions. It is a real accelerant. It is also operating inside a dense lattice of FDA meeting-management time clocks that the model does not know and will cheerfully get wrong, and confusing the meeting clock with the Written Response Only clock, or missing the package-submission deadline, is the kind of error that costs a meeting slot the program cannot afford to lose. This lesson is about drafting the briefing book well and getting the clocks exactly right.

What a Briefing Book Is For, and What Is Actually in It

An FDA formal meeting briefing book exists to give the agency everything it needs to give the sponsor useful, specific answers to specific questions, and its quality is judged by whether the review team can prepare efficiently and respond substantively. A well-built package opens with administrative information and a product-development background, states the proposed indication and the regulatory history, summarizes the relevant nonclinical pharmacology and toxicology, lays out the clinical development plan and the design of the proposed study, and then, crucially, presents the sponsor's specific questions each paired with the sponsor's own position and the data supporting it. The questions are the heart of the document; everything else exists to let the reviewer evaluate the questions in context. A briefing book that buries vague questions in a wall of background, or asks questions the data in the book cannot support, wastes the single most valuable interaction a sponsor gets with the agency before filing.

The package is filed as a named artifact, the meeting briefing document submitted in eCTD Module 1.6, the section reserved for meeting-related correspondence and background materials, with the supporting data carried in appendices and cross-referenced into the relevant CTD modules. This matters for AI-assisted drafting because the model is good at producing the prose of the background and summary sections, where it is recombining well-represented patterns of regulatory writing, and far less reliable at the two things that actually determine the meeting's value: the precision of the questions and the soundness of the sponsor's positions. The lead who understands this lets the model accelerate the scaffolding and reserves her own judgment for the questions and positions, because those are where a generated draft can be fluent and wrong in ways that a reviewer will notice immediately and that can shape the entire meeting outcome.

The Meeting Types and Why They Differ

FDA formal meetings under the PDUFA meeting-management framework come in named types, and the type determines the clock, so the lead must classify the meeting correctly before anything else. Type A meetings are for stalled or urgent matters, such as a clinical-hold resolution or a dispute, and they carry the fastest clock. Type B meetings are the milestone meetings, including pre-IND meetings, End-of-Phase-2 meetings, and pre-NDA or pre-BLA meetings; they are the workhorses of program planning, and the rare-disease pre-IND in this scenario is a Type B. Type B (End-of-Phase-2) is sometimes tracked distinctly because it carries a slightly longer scheduling window. Type C meetings are the catch-all for any other meeting that does not fit A or B, often used for specific scientific or statistical questions. Type D, the newest addition under recent PDUFA commitments, is a narrowly scoped meeting on a small number of focused issues, intended to be lighter-weight and faster than a Type C.

The reason the types matter to the AI-assisted writer is that the model will happily produce a briefing book without ever pinning down which meeting type it is, and the meeting type silently determines every downstream deadline. A draft that is excellent in content but built on the wrong meeting-type assumption can lead a sponsor to plan against the wrong submission date, request the wrong kind of meeting, or set internal milestones that miss the agency's window. The classification is a human regulatory judgment that depends on the program's stage and the questions being asked, and it is exactly the kind of judgment the model cannot make reliably because it cannot know the program's true status. The lead classifies the meeting; the model drafts within that classification.

The Meeting Clock Versus the Written Response Only Clock

The single most important distinction in this lesson is that a sponsor's meeting request can be granted in two fundamentally different forms, and they run on different clocks. A live meeting, whether face-to-face, by teleconference, or by videoconference, is scheduled within a target number of days of FDA receiving the request. A Written Response Only, or WRO, outcome means the agency declines a live meeting and instead provides written answers to the sponsor's questions by a target date, and the WRO clock is measured to the date the written responses are due rather than to a meeting date. These are not the same event and not the same deadline, and a briefing-book author who conflates them will misjudge when the package is due and when answers will arrive. The WRO is increasingly common for certain meeting types and is sometimes the FDA's preferred format, so treating every granted request as a live meeting is a planning error.

The concrete timelines under the current framework are worth stating precisely, because precision is the whole point. Type A meetings are scheduled within roughly thirty days of the request. Type B meetings are scheduled within roughly sixty days, with the End-of-Phase-2 variant at roughly seventy days. A pre-IND request handled as a Written Response Only carries a target of roughly thirty days for the written response. Type C meetings are scheduled within roughly seventy-five days. Type D meetings are scheduled within roughly fifty days. Cutting across all of these is the rule the lead in our scenario is living under: the meeting package is due to the agency no later than thirty days before the scheduled meeting date, or before the WRO response date for a written-response outcome. That thirty-day package rule is why a forty-five-day runway to a meeting is really a fifteen-day runway to a deadline, and it is the calculation the model will not do for her unless she makes it do it, and then checks it.

Why the Model Gets the Clocks Wrong

A large language model is a pattern completer, and regulatory time clocks are exactly the kind of precise, numeric, frequently-revised facts that pattern completion handles badly. The PDUFA meeting-management timelines have been adjusted across successive reauthorizations, Type D meetings are relatively new, and the exact day counts have shifted over the years, which means the training corpus contains multiple inconsistent versions of these numbers from different eras. When the model writes that a Type C meeting is scheduled within a certain number of days, it is producing a statistically plausible number drawn from that mixture, and a plausible day count is indistinguishable on the page from the currently correct one. The model can also confidently merge the meeting clock and the WRO clock, because in the corpus those concepts appear near each other constantly, and proximity in the training data is exactly what produces a fluent conflation.

This is why the timelines in this lesson are stated as targets under the current framework rather than as eternal constants, and why the operational rule is that the briefing-book author verifies every date against the current FDA meeting-management guidance and the PDUFA commitment letter in force, never against the model's recollection. The model is useful for drafting the language that explains the sponsor's understanding of the timeline; it is not a source for the timeline itself. The defensible practice treats any date or day-count the model produces as a claim to be reconciled against the authoritative guidance, the same discipline applied to a fabricated TLF citation in a Clinical Overview, because a wrong meeting deadline and a wrong table citation are the same category of error: a confident, plausible, unverified fact that the model had no way to actually check.

Drafting the Questions and Positions, the High-Value Core

The questions and the sponsor positions are where the meeting is won or lost, and they are where AI assistance is both most tempting and most dangerous. A good meeting question is specific, answerable in the time the agency has, and tied to a decision the sponsor needs to make: not "does FDA agree our program is acceptable" but "does the Agency agree that the proposed 28-day repeat-dose toxicology package in two species is sufficient to support the proposed first-in-human single-ascending-dose study." The model, asked to draft questions, tends toward the vague and the over-broad, because vague regulatory questions are abundant in its training data and because it cannot know the precise decision the sponsor is trying to unblock. The lead must sharpen every question to the specific agreement she needs, and she must ensure each question is paired with a sponsor position and the supporting data, because a question without a position invites a non-committal answer and wastes the slot.

The deeper risk is that the model will generate a sponsor position that the program's data does not actually support, stating it with the same fluent confidence it brings to a supported one. A position that the nonclinical package justifies a particular starting dose is a scientific claim that must trace to the actual toxicology data in the appendices; if the model asserts it and the data do not carry it, the sponsor walks into the meeting with a position the reviewer will dismantle, or worse, files a written position the agency rebuts on the record. The discipline is the same claim-to-source reconciliation taught throughout this program: every position in the briefing book must trace to data in the package, and the named author owns that trace. The model can structure the argument and draft the prose; it cannot decide whether the argument holds, and a Type B pre-IND meeting is too scarce a resource to spend on a position the data cannot defend.

Assembling the Package Into Module 1.6 With Appendices

The finished artifact is a complete briefing package filed into eCTD Module 1.6, and assembly is where AI assistance pays off in hours saved if it is governed correctly. The model can draft the administrative section, the background, the nonclinical and clinical summaries, and a clean first pass of the development plan, and it can help ensure internal consistency across sections so that the dose stated in the summary matches the dose in the protocol synopsis and the appendices. It can build the cross-reference map from each summary claim to its supporting appendix, which is genuinely useful because a briefing book with a tight claim-to-appendix trace is one a reviewer can navigate quickly. What it cannot be trusted to do is confirm that those cross-references point to real, correctly numbered appendices, the same fabricated-citation risk that haunts every AI-assisted regulatory document, so the reference QC on the package must be human.

The package-assembly checklist the lead runs is therefore a blend of what the model accelerated and what only she can certify: the meeting type is correctly classified, the meeting and WRO clocks are correctly applied against current guidance, the package-submission date is calculated as no later than thirty days before the meeting or WRO date, every question is specific and decision-linked, every sponsor position traces to data in the appendices, every cross-reference points to a real appendix, and the whole package is filed into Module 1.6 in the correct eCTD structure. The model gave her a draft in under a minute and saved her days on the scaffolding. The fifteen-day runway is spent on the judgment the model cannot supply: classifying the meeting, fixing the clocks, sharpening the questions, defending the positions, and verifying the references. That is the work that turns a fast draft into a briefing book that earns the meeting it is asking for.

The Meeting Request Versus the Briefing Package, Two Documents on Two Timelines

It is easy to collapse two distinct documents into one mental object, and the model encourages this because it will draft either on request without flagging the difference. The meeting request is the document the sponsor submits to ask for the meeting; it names the meeting type, states the purpose, lists the proposed questions in draft form, and identifies the attendees and the preferred format. The agency responds to the request by granting or denying the meeting and, if granting, scheduling it within the type-specific window or designating a Written Response Only outcome. Only after the meeting is granted and dated does the second document, the full briefing package, come due, and it comes due on the thirty-day-before clock that runs from the scheduled meeting or WRO date. The request opens the process and starts the scheduling clock; the package is the substantive document the review team reads to prepare.

The practical hazard for the AI-assisted writer is sequencing. The questions in the request and the questions in the package must be consistent, but the package questions are the refined, position-backed versions, and a lead who lets the model generate fresh questions for the package that drift from the ones in the request creates a discontinuity the review team will notice. The disciplined workflow treats the request's questions as the spine and the package as their fully supported expansion, with the model helping to elaborate the supporting data and positions rather than reinventing the questions. Getting this sequence right also clarifies the schedule: the request goes in early to start the clock, the agency's response fixes the meeting or WRO date, and the package deadline is then calculated backward thirty days from that fixed date, which is the moment the apparent runway becomes the real one.

What This Means for the RA Lead on Monday

Treat the model as a drafting engine for the parts of the briefing book that are recombination, the background, the summaries, the development-plan prose, and treat your own judgment as non-delegable for the parts that decide the meeting, the classification, the clocks, the questions, and the positions. Calculate the package deadline first, before drafting anything, because the thirty-day-before rule converts your apparent runway into your real one and everything else is scheduled backward from it. Verify every meeting-management date against the current FDA guidance and the PDUFA commitment letter in force, never against what the model wrote, and treat the meeting clock and the WRO clock as distinct deadlines that you confirm separately. Sharpen every question to the specific decision you need, pair it with a position, and reconcile every position to the data in the appendices. The model can hand you a complete-looking briefing book in under a minute; whether it is a briefing book that earns a substantive answer from the FDA depends entirely on the judgment you bring to the half the model cannot do, and the named author of the package owns the meeting it produces.

Key Takeaways

• The package deadline, not the meeting date, governs the schedule: the briefing package is due no later than 30 days before the meeting or the WRO response date. A forty-five-day runway to a Type B pre-IND meeting is really a fifteen-day runway to a deadline, and the model will not do this calculation unless the lead makes it and then verifies it.
• Classify the meeting type before drafting, because the type silently determines every clock. Type A is scheduled within roughly 30 days, Type B within roughly 60 (70 for End-of-Phase-2), a pre-IND WRO within roughly 30, Type C within roughly 75, and Type D within roughly 50; the classification is a human regulatory judgment the model cannot make reliably.
• The meeting clock and the Written Response Only (WRO) clock are different deadlines, and the model will fluently conflate them. A live meeting is scheduled to a meeting date; a WRO is measured to the date written answers are due, and treating every granted request as a live meeting is a planning error, since WRO is increasingly common.
• Verify every meeting-management date against current FDA guidance and the PDUFA commitment letter, never the model's recollection. Day counts have shifted across reauthorizations and Type D is new, so the training corpus mixes eras; a plausible-but-wrong day count is the same category of error as a fabricated TLF citation.
• The questions and sponsor positions decide the meeting, and they are where AI is most tempting and most dangerous. The model drifts toward vague, over-broad questions and can state positions the data does not support with full confidence; sharpen every question to a specific decision, pair it with a position, and reconcile every position to the data in the Module 1.6 appendices.