Who is responsible for this AI legal service?
For legal teams and consumers evaluating an AI service.
When a provider says “lawyer-reviewed,” who reviewed what?
Ask whether lawyers helped design the product, reviewed a sample, or reviewed your particular result. Identify the organization providing the service, the person responsible for the relevant decision and a contact who can investigate and correct an error.
Separate three roles
List the company offering the tool, the people who tested or designed it, and anyone who will review an individual output. One person may hold more than one role. Record exactly what the provider says each person does, when the review happens and whether a user may rely on it for the proposed task.
If the service says a lawyer will represent you, look for that lawyer’s name, jurisdiction, current status and written scope of the engagement. Verify the professional through the relevant licensing authority and independently sourced contact details. For a drafting tool, ask who on your own team must approve the output before it reaches anyone else.
Ask for one result you can trace
Use a fictional sample. Ask the provider to identify what the tool generated, what materials a reviewer checked, what the reviewer could change and what happened when the reviewer found an error. If the provider describes only general quality testing, record that limit on the individual-review claim.
The Federal Trade Commission’s final order involving DoNotPay addressed deceptive claims about a specific AI chatbot’s ability to perform like a lawyer. Its published account illustrates why evidence for a professional-equivalence claim matters. Assess the particular service and claim in front of you.
Find the correction path
Before sharing sensitive material, identify the authorized submission method, data handling terms and support contact. Ask who receives a complaint, how affected users learn about a correction and who can stop a flawed output from being reused. If a critical answer is missing, limit testing to fictional material and record the open issue.
Worked example
Illustrative example. A vendor says an attorney reviewed its templates. When asked about a generated lease summary, it confirms that no attorney checks each result. The buyer records template review as the supported claim, assigns its own reviewer for individual summaries and asks how a faulty template would be corrected.
Primary sources
Related work at Agentic Attorney
This service-evaluation guide is part of the Agentic Attorney resource collection. The main site describes AI consulting and implementation for legal teams. Explore Agentic Attorney’s approach to AI implementation.