How AI is used in Review
A clear look at what the AI does, how to spot it, and where your data goes.
Overview
Review uses AI to help you find and understand your captures, but it never changes the underlying evidence. AI surfaces, ranks, suggests, and explains. You decide what matters, what to mark up, and what to produce.
Most of Review's AI features are guided by your case context, a short description of your case that you write. See Case Context and Relevance.
Which features use AI
- Relevance ratings: Each capture is rated high, some, or low relevance to your case, with an explanation.
- Suggested marks: Review suggests the pages most relevant to your case, with a drafted explanation for each. See Suggested Marks.
- Suggested searches: Review suggests searches based on your case context and each capture's content.
- Image and relevant search results: Review describes the pictures on each page, finds content that matches the meaning of your search or appears in images, and writes a short description of each result.
- Entities: Review identifies people, companies, places, dates, and links in your captures. See Entities in Review.
- Export cover pages: The Review Summary cover page in exports with Marks or Suggestions includes an AI summary of each capture.
Which features don't use AI
Your captures themselves, your case context, case organization, filters and sorting, and the highlights, notes, and redactions you add all work without AI.
How to tell them apart
Everything in the list above is AI-generated. A few places carry an extra label:
- Suggested marks are labeled Suggestion and marked with a robot icon until you accept them.
- Relevant search results are marked with a robot icon.
- The first time you see suggestions, Review reminds you that they're AI-generated and may be incomplete or inaccurate.
When you accept a suggestion, it becomes your highlight. Its note starts as the AI-drafted explanation, and you can edit it. If you ever need to know whether something came from the model or directly from your capture, check the capture itself.
Where your data goes
Review's AI models run on Amazon Bedrock, AWS's enterprise AI service, inside Page Vault's own secured AWS environment. Your content isn't sent to a third-party AI vendor, and the companies that build the models can't see your prompts or the model's responses. Processing is encrypted in transit and at rest, and AWS doesn't keep your prompts or responses. The AI only processes captures in your Review cases: ones you send, and ones added automatically when you turn on auto-add by Case ID. Nothing else in core Page Vault is analyzed.
To make your captures searchable, Review turns them into embeddings: mathematical representations that let it find the most relevant content quickly. These are stored in a search database provided by Qdrant, a managed service that Page Vault lists as a sub-processor. Qdrant runs on AWS, in Qdrant's own AWS environment rather than Page Vault's. Your data there is encrypted in transit and at rest, protected by contractual data-protection commitments, and never used to train AI models or shared for any other purpose. We're moving this database into Page Vault's own AWS environment.
Your content is not used to train any AI models, and your original captures are never changed by anything the AI does. If your firm needs more detail for a security review, ask your Page Vault contact for the "AI and Your Data with Page Vault Review" one-pager.
Frequently asked questions
Does the AI change my captures? No. Your original captures are never altered. Everything the AI adds sits on a separate layer.
Can I use Review without the AI features? Yes. Skip case context and search and mark captures yourself. Relevance ratings and suggestions need case context to appear. You can also hide suggestions in an open capture with the Suggestions switch; it turns back on the next time you open a capture.
Is my case context used to train AI models? No. Your content, including your case context, is not used to train any AI models.