Plaintiff PI firms are experimenting with AI for intake summaries, medical-record review, demand drafting, and issue spotting. The risky part is not that an attorney asks an AI tool to help organize a file. The risky part is sending privileged strategy, client communications, or unnecessary medical detail into a tool that was never designed for confidential legal work.
Privilege-aware prompting is the practical discipline of deciding what belongs in an AI request, what should stay inside the firm, and what must be reviewed by an attorney before any output becomes part of the demand workflow. For PI firms, that discipline matters because demand letters often sit at the intersection of medical facts, liability theory, negotiation posture, and attorney work product.
Why generic AI tools create a different risk profile
Most plaintiff firms do not need a lecture on attorney-client privilege. The harder problem is operational: lawyers and staff are under time pressure, case files are messy, and general-purpose chat tools make it easy to paste more than the task requires. A prompt that starts as “summarize these records” can quickly become a dump of intake notes, liability concerns, adverse facts, settlement posture, and private client context.
That matters because privilege and confidentiality are not abstract compliance concepts in a PI practice. California attorney-client privilege is reflected in Evidence Code sections 950 through 962, while attorney work product protections are addressed in Code of Civil Procedure section 2018.030. Firms also handle medical information that may create HIPAA, state privacy, contractual, or protective-order obligations depending on the file and the vendor relationship. The attorney remains responsible for protecting the client’s information and for reviewing any AI-assisted work before it leaves the firm.
Generic AI tools can be useful for low-risk drafting, internal brainstorming, or formatting tasks when no confidential information is involved. But they usually are not built around the realities of plaintiff demand work: medical chronology, specials review, causation analysis, lien context, adjuster-facing narrative, and attorney-supervised advocacy. That difference should shape what the firm allows into the prompt.
What plaintiff firms should avoid putting into generic prompts
A good internal policy starts with the obvious categories, then moves into the gray areas where mistakes actually happen. The following material should stay out of a generic AI tool unless the firm has completed vendor review, confirmed the data-use terms, and decided that the workflow is appropriate for confidential legal work.
Client communications and attorney impressions
Direct client messages, intake-call notes, and attorney comments about credibility, strategy, comparative fault, or settlement leverage should not be pasted casually into a generic chatbot. Even if the tool does not display the information publicly, the firm still needs to know how the data is processed, retained, secured, and used. A better pattern is to convert the task into a neutral instruction: “Create a checklist of questions an attorney may want to consider before finalizing a disputed-liability demand,” rather than uploading the attorney’s actual strategy note.
Unnecessary medical identifiers
PI demand work often requires medical facts, but not every AI task requires every identifier. A prompt asking for a timeline does not need a patient’s full identifying details if the goal is structure, sequencing, or issue spotting. Firms should strip unnecessary identifiers, provider account numbers, claim numbers, and file-specific labels before using any general tool. When medical-record content is necessary, the firm should use a workflow with clear security terms, access controls, and appropriate vendor commitments.
Negotiation posture and reserve assumptions
Demand drafting is not just a writing exercise. It may include internal valuation ranges, adjuster history, carrier-specific assumptions, mediation posture, or a plan for what to concede later. Those are exactly the notes that can become work-product-sensitive. A generic prompt should not include the firm’s internal bottom line, the attorney’s assessment of weak facts, or strategy about how to frame damages if litigation becomes necessary.
Unreviewed case documents in bulk
Bulk uploading is tempting because it saves time. It is also where firms lose control. Medical records, police reports, photographs, insurance correspondence, wage records, and prior demand drafts may each carry different confidentiality considerations. Before staff upload a batch of documents, the firm should know whether the tool is approved for that document type, whether uploads are retained, who can access them, and whether the task truly requires the full file.
A practical prompting framework for PI firms
Privilege-aware prompting works best when it is simple enough for staff to follow during real case work. The point is not to stop AI use. The point is to separate safe operational assistance from confidential legal judgment.
- Classify the task first. Is the firm asking for public-law research structure, generic checklist drafting, document formatting, medical chronology support, damages narrative review, or attorney strategy? The category determines the permitted tool.
- Use the minimum necessary facts. If the model only needs injury type, treatment sequence, and procedural posture, do not include full identifiers or private communications.
- Separate facts from legal judgment. Medical dates, treatment gaps, and billing entries can be organized differently from attorney impressions about liability, causation, and settlement leverage.
- Require attorney review before external use. AI output should not become a demand-letter fact, causation statement, or damages argument until an attorney verifies it against the file.
- Document the approved workflow. Staff should know which tools may be used for confidential file materials, which may only be used with sanitized hypotheticals, and which are off limits for client information.
This framework also helps firms train new staff. Instead of saying “be careful with AI,” the firm can give concrete examples: a sanitized checklist prompt may be fine in a general tool; a full medical-record upload belongs only in an approved workflow; a negotiation memo should stay under attorney control.
Where vendor review fits into privilege-aware prompting
Prompt discipline and vendor discipline have to work together. A firm can write careful prompts and still choose the wrong tool for confidential work. Before using AI for medical-record intake, chronology review, or demand drafting, plaintiff firms should ask vendor questions that connect directly to privilege, confidentiality, and work product.
Key questions include whether the vendor uses uploaded data to train general models, how long files and prompts are retained, whether the vendor supports appropriate BAAs for HIPAA-sensitive workflows, what access controls exist inside the product, how audit logs work, and whether attorneys can review source material behind an AI-generated conclusion. These are not cosmetic security questions. They determine whether the workflow fits the firm’s professional obligations.
For a deeper companion piece on defensible workflow design, see Privilege, Work Product, and Audit Trails: Building a Defensible AI Demand Workflow.
How Legal Power AI fits
Legal Power AI is built for plaintiff PI demand workflows rather than general legal brainstorming. That matters because the product focus is narrower: medical records, chronology-building, demand-letter drafting, attorney review, and firm-controlled use of case materials. The right goal is not to let AI replace legal judgment; it is to give attorneys a more structured way to review facts, preserve work-product discipline, and move from file review to demand drafting without pushing sensitive context into tools that were not designed for that job.
Conclusion
Privilege-aware prompting is a workflow standard, not a one-time warning. Plaintiff firms should define what can go into a generic AI tool, what requires an approved legal-tech workflow, and what belongs only in attorney-controlled analysis. The payoff is practical: fewer accidental disclosures, cleaner staff habits, better review trails, and AI-assisted demand work that stays anchored to the attorney’s responsibilities.
Ready to see a plaintiff-PI demand workflow built around attorney review and confidentiality?
