PDF Auditor
Upload a PDF and get an automated brand-compliance audit against your own guidelines, with an AI semantic review layered on top.
Overview
PDF Auditor checks PDF documents (reports, cover decks, contracts, brand collateral) against a set of brand guidelines you supply — either pasted as text or uploaded as a Markdown (.md) file. It runs a deterministic rule engine first (fast, exact pattern checks) and then layers an AI semantic review on top to catch issues the rules can't express (tone, clarity, brand voice). The result is a 3-panel review workstation where every finding can be accepted, dismissed, or annotated with a free-text note before you export the final report.
PDF Auditor shares its analysis engine (AuditEngine.ts) with PPTX Auditor — the same deterministic rule set, the same Claude-based AI review, and the same job-queue infrastructure power both screens. Only the file parser and the file extension differ: PDF Auditor extracts text with pdf-parse, while PPTX Auditor reads PowerPoint XML directly.
Key Features
Custom Guidelines
Paste brand guideline text directly, or upload a .md file. A built-in ACME-style default set is provided as a starting template.
Deterministic Rule Engine
Fast, exact-match checks run first — draft markers, banned abbreviations, missing legal text, title formatting, and more.
AI Semantic Review
A Claude-based pass reviews tone, clarity, and brand voice — the things a regex can't catch.
3-Panel Review Workstation
Page list on the left, page content in the middle, and findings on the right — everything you need on one screen.
Accept / Dismiss / Annotate
Triage every finding individually. Add a free-text note to any finding to record context for reviewers or track a follow-up.
Export Options
Download a PDF summary (all findings or accepted-only) or export the full findings set as structured JSON.
Running an Audit
- Go to Audit → PDF Auditor.
- Upload a
.pdffile using the drop zone or file picker. - Provide brand guidelines: paste text directly, upload a
.mdfile, or use the built-in default guideline set as a starting point. - Click Run Audit. The file uploads to
/api/pdf-upload, and the client polls/api/jobs/:iduntil the job completes. - Review findings in the 3-panel workstation: select a page on the left, read its extracted text in the middle, and triage issues on the right.
- For each finding, click Accept or Dismiss, and optionally add a note using the note icon.
- Export the report as PDF (all findings or accepted-only) or JSON, or save it via
/api/pdf-auditsfor later reference (tenant-scoped).
Deterministic Rules Checked
These checks run instantly against the extracted PDF text, before the AI review begins:
| Rule | Severity | What it catches |
|---|---|---|
| Draft marker | High | The word "DRAFT" appearing anywhere on a page |
| "TGT" abbreviation | High | Use of the "TGT" abbreviation instead of spelling out "Target" |
| Title punctuation | Medium | Page titles ending with unwanted punctuation |
| Title all-caps | Medium | Page titles written entirely in capital letters instead of sentence case |
| Title too long | Low | Page titles that exceed the recommended length |
| Missing confidentiality statement | Low | No confidentiality / proprietary / internal-use statement found on a page that has content |
| Bullet punctuation | Low | Inconsistent trailing punctuation across bullet list items |
| "Percentile" vs "%ile" | Low | Use of the full word "Percentile" where the style guide calls for "%ile" |
| Empty page | Medium | A page with no extractable text content |
Limitation: no font, color, or position checks on PDF input
PPTX Auditor can additionally check non-standard fonts, off-brand color palettes, title font size, and title position — because PowerPoint's XML format exposes per-run styling directly. PDF Auditor cannot run these checks: text is extracted from the PDF with pdf-parse, which returns plain text only and does not expose per-character font, color, or positional metadata. Only the text-based rules in the table above apply to PDF input. If font/color/layout compliance is a hard requirement for a document, audit the source PPTX before it is exported to PDF.
Accept, Dismiss & Notes
Every finding in the issues panel starts in a pending state. From there:
- Accept marks the finding as reviewed and intentional — it's kept in the record but flagged as acknowledged rather than an open issue.
- Dismiss marks the finding as not applicable or a false positive. Dismissed findings can be filtered out of the accepted-only PDF export.
- Add a note (the note icon on each finding) attaches a free-text annotation to that specific finding — useful for recording why a finding was accepted or dismissed, or flagging it for a teammate to double-check. Notes are included in both the PDF export and the JSON export.
A Dismiss all control is available per page to quickly clear low-priority findings on pages you've already reviewed.
Exporting the Report
- Download PDF (All Issues) — a formatted summary of every finding, grouped by severity and page, including any notes.
- Download PDF (Accepted Issues) — a summary containing only the findings marked Accepted, useful for a final sign-off document.
- Export JSON — the full structured findings set (accepted, dismissed, and pending), including notes, for programmatic use or archiving.
Frequently Asked Questions
.md file. A built-in default guideline set (styled after a generic corporate document standard) is provided as a starting template you can edit./api/pdf-audits, which is tenant-scoped — only members of your workspace can see your saved audit reports.