The American Bar Association issued Formal Opinion 512 on July 29, 2024, making clear that lawyers must independently verify any AI-generated output for accuracy and reliability before it’s used in a matter. The opinion applies Model Rule 1.1 (competence), Rule 1.6 (confidentiality), and Rule 5.3 (supervisory responsibilities) to generative AI tools, and frames the verification obligation as factually specific: the level of review required depends on what the tool did and what the document will be used for. That guidance lands squarely on PI firms, where AI-generated demand letters, medical chronologies, and case summaries are increasingly part of everyday practice.
Knowing the standard is one thing. Knowing what to look for when reviewing a draft before it goes out is another. AI legal drafting has gotten good enough that outputs can look polished and complete. Still, it may contain errors that an adjuster, opposing counsel, or judge will catch. The danger isn’t obvious wrongness. It’s subtle errors that only surface when someone checks carefully against the source.
This article covers the specific things to verify, section by section, before any AI-generated PI document leaves the firm.
WOMEN IN BUSINESS: The Most Influential Women in Arizona for 2026
WOMEN IN REAL ESTATE: The Most Influential Women in Commercial Real Estate for 2026
Why Verification Is Harder Than It Sounds
Verification feels straightforward until you’re doing it under deadline pressure on the fifteenth file of the week. Several categories of AI error don’t announce themselves, which is what makes a structured review habit more reliable than a quick read. This is especially true in AI legal drafting workflows, where outputs can look complete even when key details are subtly wrong.
The Types of Errors That Slip Through
A few patterns show up repeatedly when AI-generated PI documents go out with mistakes:
- Date errors: AI tools occasionally transpose dates or compress timelines, making treatment appear to start earlier or later than it did
- Unsupported claims: a sentence that asserts a specific diagnosis or wage loss figure that doesn’t trace back to anything in the uploaded records
- Confident paraphrasing: the output restates a medical note in cleaner language that subtly changes its clinical meaning
- Missing gaps: a chronology that lists treatment without flagging a two-month period of no treatment that the adjuster will notice immediately
Why These Errors Are More Dangerous in PI Than in Other Practice Areas
In PI, every factual claim in a demand letter or medical chronology has a direct line to settlement value. An adjuster who finds a factual error in a demand package uses it to question everything else, which costs time and often costs money on the final number.
What to Check in a Demand Letter Before It Goes Out
A demand letter built with AI for legal drafting starts from organized case data rather than from scratch, reducing structural errors that come from manual compilation. But the sections requiring attorney judgment still need close attention regardless of how the draft was generated.
The Structural Sections
The parts of a demand letter that AI handles most reliably are also worth checking most carefully, because errors carry through the entire document:
- Treatment chronology: read every date and provider against the underlying records
- Damages total: add up the attached bills yourself and confirm the stated total matches line by line
- ICD codes: verify any referenced diagnosis codes directly against the billing records
- Gap explanations: flag any period of interrupted treatment the draft ignores
The Sections That Need the Most Attorney Judgment
The liability narrative and the settlement figure are where a weak or inaccurate sentence does the most damage. Read these as if seeing the case for the first time, because that’s how the adjuster will read them.
What to Check in a Medical Chronology Before It Gets Used
A medical chronology feeds into depositions, demand letters, and trial prep, so an error compounds across everything downstream. The same approach to AI for drafting legal documents applies here: trust the tool for extraction, verify the output before it goes anywhere.
A Practical Verification Process for Chronologies
| Element | What to Verify | How to Check |
| Dates | Every treatment date in the chronology | Spot-check against source records for 5–10 entries |
| Provider names | Correct provider attributed to each entry | Cross-reference billing and treatment notes |
| Diagnosis accuracy | Clinical terms not paraphrased into something different | Compare against actual medical notes |
| Source citations | Each entry links to a specific page | Click through on any entry that will be used at deposition |
| Gap periods | Unexplained breaks in treatment | Note any gap over 30 days and confirm it’s addressed |
What to Do When the Chronology Has Errors
Flag errors as you go rather than correcting them inline, so you can do a second pass to check whether the same type of error appears in multiple places. A pattern of errors usually points to a systematic issue rather than a one-off mistake.
How to Build a Pre-Send Checklist Into the Firm’s Workflow
A pre-send checklist reduces the chance that any single review step gets skipped under deadline pressure. The value comes from making verification routine rather than optional.
The Minimum Viable Check for Any AI-Generated PI Document
A few checks apply to every AI-generated document the firm sends:
- Every factual claim should trace to a specific page in the underlying records
- Damages figures should match the attached or referenced billing documents exactly
- The signing attorney should be able to explain every key assertion, not just read it from the AI draft
A one-page demand letter review checklist built into the firm’s workflow is a practical way to standardize this step without making it cumbersome.
What the ABA Verification Standard Means in Practice
ABA Formal Opinion 512 states that the required verification “will necessarily depend on the GAI tool and the specific task that it performs,” meaning there’s no universal answer. A demand letter sent to an adjuster requires closer review than an internal case summary used for strategy.
The practical takeaway is that AI legal drafting in a PI context means the attorney’s review is still the last line of defense before a document creates a legal or professional consequence. The AI produces the draft. The attorney is responsible for what goes out. This applies whether the tool was used for AI for drafting legal documents like demand letters and chronologies, or for internal work product like case summaries.
Applying These Standards Across a Full Caseload
The firms that handle this well don’t review AI documents differently from manually drafted ones; they review all outgoing documents with the same habit. The difference is that AI drafts can be reviewed faster because the structural work is already done, freeing up review time for the judgment calls that deserve more attention.
FAQ
Does the ABA verification requirement apply to all AI tools or just generative AI? Formal Opinion 512 focuses on generative AI, but the underlying competence obligations under Model Rule 1.1 apply to any technology a lawyer uses in client work. The attorney remains responsible for verifying output before it creates a consequence.
How detailed does a pre-send review need to be for a routine demand letter? It depends on the case’s stakes. A straightforward soft-tissue demand with a single provider warrants a lighter review than a catastrophic injury demand with multiple providers, disputed causation, and a large settlement figure.
Can a paralegal conduct the pre-send verification, or does it need to be an attorney? A paralegal can run the initial structural check, verifying dates, totals, and citations. The attorney still needs to review sections involving legal judgment, specifically liability framing and the settlement figure, before signing off.
What happens if an error slips through in an AI-generated document? The attorney remains responsible regardless of how the document was produced. Consequences are the same as any other drafting error: potential malpractice exposure, reduced settlement value, or damage to the firm’s credibility with adjusters.
Are there specific document types where AI verification is more critical? Court filings carry the highest stakes since Rule 3.3 requires candor toward the tribunal. Demand letters follow closely because factual errors directly affect settlement outcomes.
