AutoReview Use Cases Guide

New to AutoReview? This guide assumes you know how to run a review. If you haven't used AutoReview before, start with the AutoReview overview article to learn how to set up and run your first review, then come back here for specific use cases.

Before you start: AutoReview lives in the Dossier view, not inside the document editor. Navigate to your program's dossier and select the AutoReview tab at the top of the page.

How AutoReview Works

AutoReview reads the sections you select, decides which ones are relevant to your instruction, and adds comments where it finds something to flag. It does not edit or delete any content, every change is yours to make.

Comments appear in the same panel you use for regular editing, labeled "Weave AI on behalf of [your name]."

Two things that affect every run:

• Instruction quality — the more specific your instruction, the better the results (more on this below)
• 100-comment cap — AutoReview adds up to 100 comments per run. For large dossiers, narrow your scope or break the check into multiple runs

Note: Runtime is 1–15 minutes depending on scope. You can leave the page while it runs — results are saved and available when you return to the AutoReview tab.

Write a Good Instruction

This is the most important thing to get right. Vague instructions produce low-quality results. Specific instructions produce high-quality results — consistently.

Instruction type	Typical precision
Vague ("check for inconsistencies")	~20%
Specific ("flag any pH range outside 6.5–7.5")	~75–93%

A strong instruction has three parts:

1. What to check — name the specific parameter, content type, or standard
2. What correct looks like — a range, rule, or criterion
3. What to flag — tell it exactly when to add a comment

Example: "Flag any location where the pH range for the drug substance is stated outside 6.5–7.5, or where the pH range differs across sections."

The use cases below include tested example instructions. Use them as starting points and adapt to your criteria.

Use Cases

1. Manufacturing Process Parameter Consistency (CMC)

When to use this: You have the same parameter — pH range, storage temperature, yield range — referenced across multiple Module 3 sections and want to confirm every reference is consistent and within specification.

Who this is for: Manufacturing (CMC) Scientist, Regulatory Operations, Quality / Compliance Reviewer

How to set it up:

1. Go to Dossier → AutoReview tab → New AutoReview
2. Select the relevant Module 3 sections (e.g., 3.2.S.2, 3.2.S.3, 3.2.S.4, 3.2.P.3, 3.2.P.5)
3. Write a custom instruction targeting the specific parameter

Example instructions:

Specific value check:

"Flag any reference to the pH range for the drug substance that is stated outside 6.5–7.0, or where the pH range differs from other sections in this module."

Critical process parameter sweep:

"Review all sections for references to critical process parameters. Flag any location where a parameter value differs from its first occurrence in this document, or is described differently in a manufacturing narrative versus a specification table."

Storage conditions:

"Flag any reference to storage conditions (temperature or humidity) that differs across sections of this module."

Tip: Run this immediately after updating a specification — before any other edits — to confirm the change propagated correctly.

2. Cross-Author and Cross-Document Consistency

When to use this: Multiple authors have drafted different sections and you want to confirm that numbers, narratives, patient population descriptions, and safety findings are consistent before the document leaves Weave.

Who this is for: Medical Writer / Regulatory Author, Program Manager, Quality / Compliance Reviewer, Regulatory Operations

How to set it up:

1. Scope to the full document, or to the related documents you want to compare (e.g., CSR + study protocol)
2. Select the Weave Consistency Reviewer preset — the fastest starting point, no additional instruction needed
3. Follow up with targeted custom instructions for specific areas of concern

Example custom instructions:

Narrative vs. table alignment:

"Flag any location where a number, percentage, or patient count in a narrative sentence differs from the value in the corresponding table in this document."

Dosing consistency:

"Flag any section that describes the dosing regimen with a dose, frequency, or route that differs from the primary dosing statement in Section 2."

Tip: Run the consistency reviewer as the final step before your document exits Weave. Resolve comments in Weave, then export a clean draft.

3. Simulated Health Authority Review

When to use this: Your document set is substantially complete and you want to anticipate what a health authority reviewer would flag, before you submit or send for formal internal review.

Who this is for: Regulatory Strategy Lead, Medical Writer / Regulatory Author, Non-Clinical / Clinical Scientist, Manufacturing (CMC) Scientist

How to set it up:

1. Scope to the module most relevant to the review type (e.g., Module 3 for CMC, Module 5 + Module 2 for clinical)
2. Write a custom instruction framing the review from the agency's perspective

Example instructions by reviewer type:

FDA CMC Reviewer:

"Review these documents as an FDA Office of Pharmaceutical Quality (OPQ) reviewer. Evaluate whether the applicant has provided sufficient assurance of identity, strength, purity, and quality. Flag critical deficiencies, areas needing revision, or sections where information appears insufficient for regulatory decision-making. Conclude each finding with: [ADEQUATE], [NEEDS REVISION], or [CRITICAL DEFICIENCY]."

FDA Clinical Reviewer:

"Review as an FDA clinical reviewer. Flag sections where: evidence of effectiveness appears insufficient for the proposed indication; the safety profile is inadequately characterized; benefit-risk framing is inconsistent with the studied population; or the indication does not match the population studied."

FDA Biostatistics Reviewer:

"Review as an FDA biostatistician. Flag sections where: primary analysis validity is not clearly established; multiplicity control is not addressed; missing data handling is absent or inadequate; or statistical conclusions are not interpretable in clinical terms."

FDA Nonclinical Reviewer:

"Review as an FDA pharmacology/toxicology reviewer. Flag sections where required nonclinical studies appear absent without justification, exposure margins are not addressed, human relevance of findings is not discussed, or there are labeling implications not reflected in the proposed label."

Important: These are simulations, not actual agency reviews. Use them for gap analysis and preparation — not for predicting agency action. Always combine with expert regulatory review before submission.

4. Health Authority Feedback Incorporation Check

When to use this: You've received questions or comments from the FDA, EMA, or another health authority, made revisions in response, and want to systematically verify that each item has been addressed across your documents.

Who this is for: Regulatory Strategy Lead, Medical Writer / Regulatory Author, Program Manager

How to set it up:

1. List each specific agency question or concern you need to verify
2. Scope to all documents that could address those items
3. Write an instruction that lists each item and asks AutoReview to tag findings as [ADDRESSED] or [PENDING]

Example instruction:

"Review every section of every selected document. For each item below, add one comment per section where the item is relevant. Tag [ADDRESSED] if the content appears to resolve the item, or [PENDING] if it does not.

Items to check:

1. [Agency question or concern]

2. [Next item]

(Continue for all items)"

Tip: Limit each run to 5–10 agency items to stay within the 100-comment cap. For large feedback packages, run multiple targeted checks by functional area (CMC, clinical, safety separately).

5. Style Guide Adherence Check

When to use this: You want to check a document for style guide compliance — abbreviations, numbers, capitalization, terminology, units — before it goes to review.

Who this is for: Medical Writer / Regulatory Author, Regulatory Operations, Quality / Compliance Reviewer

When to use AutoReview vs. document-level instructions:

• Document-level instructions → enforce style during content generation
• AutoReview → audit style after generation, or on documents edited since generation

How to set it up:

1. Scope to the documents you want to check
2. Write your style guide rules as explicit, checkable criteria

Example instruction:

"Review this document for style guide compliance. Flag any of the following:

— An abbreviation used before it is defined at first occurrence

— A number below 10 written as a numeral in body text

— A percentage written as 'percent' instead of '%'

— 'Patient' used where 'participant' is the preferred term

— An adverse event term that is not capitalized

Add one comment per violation, noting the rule it violates."

Note: The ability to attach a style guide file directly is on the roadmap but not yet available. For now, write the rules explicitly into the instructions field.

Tip: Break style checks into 2–3 focused runs by category rather than encoding all rules in one instruction. This produces more precise, targeted comments.

6. Measurement and Formatting Precision

When to use this: You want to confirm that numerical values — significant figures, decimal places, units of measure — are reported consistently throughout a document or across related documents.

Who this is for: Manufacturing (CMC) Scientist, Medical Writer / Regulatory Author, Regulatory Operations

Example instructions:

Significant figures:

"Flag any numerical value in this document not reported to three significant figures. Note the value found and the section where it appears."

Unit consistency:

"Flag any location where the same measurement type (e.g., concentration, temperature, pressure) is reported in inconsistent units across sections."

Decimal places in tables:

"Review all data tables. Flag any table where the same measurement type is reported with a different number of decimal places in different rows or columns."

7. Pre-Submission Readiness Check

When to use this: You're approaching a submission deadline and want to confirm that no placeholder text, incomplete cross-references, or TBD markers remain in any section.

Who this is for: Medical Writer / Regulatory Author, Regulatory Operations, Program Manager

Example instructions:

Placeholder locator:

"Flag any section containing: placeholder text, bracket notations (e.g., [X], [TBD], [see Table X]), incomplete cross-references, or the phrases 'to be determined', 'to be confirmed', 'to be completed', 'insert here', or 'TBD'. Add one comment per occurrence with the specific text found."

Internal author notes:

"Flag any section that appears to contain internal authoring notes or draft-stage comments not intended for the final document — these may appear in brackets, parentheses, or sentences beginning with 'Note:', 'Author note:', 'Check:', or 'Confirm:'"

8. Document Redundancy Review

When to use this: Your document has grown through multiple editing rounds and you suspect there may be duplicated content: verbatim repetition, tables duplicating narrative text, or the same term defined in multiple sections.

Who this is for: Medical Writer / Regulatory Author, Quality / Compliance Reviewer, Regulatory Operations

Example instruction:

"Review this document for content redundancy. Flag any of the following:

1. Verbatim or near-verbatim repetition of the same sentence or paragraph in more than one section

2. A data value stated in a narrative that duplicates a value in a table in the same section, without adding context

3. A term or abbreviation formally defined in more than one section

For each finding, note the two locations where the duplication occurs and suggest whether to consolidate or remove one instance.

Do not flag intentional summaries or abbreviation reminders at the start of a new section."

Reviewing and Acting on Comments

Once a run completes, comments appear in the document's Comments panel.

1. Navigate to the section and open the draft
2. Work through comments — each one contains a finding and a suggested action
3. Make edits manually using the Refinement tool or by editing directly in the text block
4. Resolve comments you've addressed; leave open those that need further discussion

Note: A one-click "convert comment to suggestion" feature is in development. When available, AutoReview findings will be convertible to tracked-change suggestions for acceptance or rejection. (Coming soon)

FAQ

Q: Where do I find AutoReview?

A: In the Dossier view, at the top of the page. Select the AutoReview tab. It is not inside the document editor.

Q: What's the difference between AutoReview and Data Verification?

A: Data Verification checks whether content in your document can be traced back to a source file in your data room ("Is this supported by the source data?"). AutoReview checks whether content meets criteria you define ("Is this consistent and correct?"). Use both — they answer different questions.

Q: Can AutoReview edit or delete content?

A: No. It only adds comments. Every edit is yours to make.

Q: Can I save a custom reviewer for reuse?

A: Yes. Saved custom reviewers appear in your AutoReview tab alongside the Weave preset and can be reused across programs.

Q: Can AutoReview work on documents not generated in Weave?

A: Not currently. AutoReview works on documents in the Weave dossier. The ability to bring in externally authored documents is on the roadmap. (Coming soon)

Q: Will two runs with the same instruction produce the same results?

A: For rules-based instructions (flag any value outside range X), results will be highly consistent. For interpretive instructions (benefit-risk assessments, style checks), some variation in phrasing is expected across runs.

Quick Reference: Instruction Templates

Use Case	Starting Instruction
CMC parameter consistency	"Flag any reference to [parameter] stated outside [X to Y], or where the value differs from other sections in this module."
Cross-author consistency	Use the Weave Consistency Reviewer preset.
Narrative vs. table alignment	"Flag any narrative sentence where a number differs from the corresponding value in the most recent table in the same section."
FDA CMC simulation	"Review as an FDA OPQ reviewer. Tag each finding [ADEQUATE], [NEEDS REVISION], or [CRITICAL DEFICIENCY]."
FDA Clinical simulation	"Review as an FDA clinical reviewer. Flag sections where evidence of effectiveness is insufficient or benefit-risk framing is inconsistent with the studied population."
Agency feedback check	"For each item listed below, tag [ADDRESSED] or [PENDING]: [list items]"
Style guide — abbreviations	"Flag any abbreviation used before it is defined at first occurrence."
Style guide — numbers	"Flag any numeral below 10 in body text that is not spelled out."
Significant figures	"Flag any numerical value not reported to [X] significant figures."
Unit consistency	"Flag any location where the same measurement type is expressed in different units across sections."
Placeholder locator	"Flag any section containing placeholders, brackets, TBD markers, or incomplete cross-references."
Redundancy review	"Flag verbatim or near-verbatim duplication across sections, and narrative text that duplicates a table value without adding context."

Getting the Most from AutoReview

Run AutoReview before you send a document to a reviewer — not after. Catching issues early compresses your review cycles and means reviewers spend time on judgment calls, not basic consistency.

If your first run produces a low signal-to-noise ratio, the fix is almost always to sharpen the instruction. Narrow the criteria, name the specific parameter, and give the AI a concrete test to apply.

The teams that get the most out of AutoReview typically run it in two passes: the Weave Consistency Reviewer preset for a broad sweep, followed by one or two targeted custom instructions for the highest-risk areas in their program.