At a law firm’s front desk, an unanswered enquiry can leave someone unsure whether they have reached the right place. AI may help organize an enquiry or draft an acknowledgement. It should not decide whether a case is worthwhile, calculate legal deadlines or tell someone they have been accepted as a client.
This article is for people evaluating an administrative workflow in a law firm, including experienced staff learning new tools. It is not an implementation approval or legal advice. A lawyer responsible for the practice should decide what is permitted under the rules that apply in the firm’s jurisdiction.
Choose one administrative task
Begin with a narrow question: can the assistant turn an approved, fictional enquiry into a structured note for a person to review? That is easier to evaluate than an automated system that chats with prospective clients, screens cases and books consultations without oversight.
List what the task must never do. Examples include promising representation, predicting outcomes, assessing credibility, quoting unapproved fees or assuring a visitor that a deadline can wait. A refusal to provide such an answer should lead to the firm’s approved human process.
People asking for help should have an understandable way to contact a person. Do not make access depend on fluent typing, a particular language or willingness to use a chatbot.
Decide what information is appropriate first
A detailed life story can contain sensitive information before a firm knows whether it can act. Let the supervising lawyer specify the minimum information needed at each stage, including the conflict-check process. Do not invite unrestricted document uploads simply because a tool accepts them.
The ABA’s Model Rule 1.18 addresses duties to prospective clients even when representation does not follow, and certain conflicts arising from information received. Local rules and circumstances matter; a generic website disclaimer is not a substitute for a properly designed intake process.
Keep contact records in the firm’s approved system. Removing a name from a narrative may not make it anonymous when its events, location or other details identify someone.
Review the tool before using real enquiries
Ask who can access submitted information, how long it is retained, whether it is used to train models, what connected services receive it and how deletion works. Check the actual account configuration and contractual terms rather than a general marketing statement.
ABA Formal Opinion 512 discusses competence, confidentiality, communication and supervision when lawyers use generative AI. It is guidance under the ABA Model Rules, not a single rule automatically governing every jurisdiction. Have the responsible lawyer assess the tool, applicable obligations and any necessary client communications.
Test a fictional enquiry
Use an invented example containing no real client information: “I received a letter about a tenancy issue. It mentions a hearing next week. I would like someone to call me on Thursday afternoon.” Do not ask the assistant to decide whether Thursday is soon enough.
Organize this fictional enquiry into: type of enquiry stated by the writer, dates or timing exactly as stated, requested contact method, and information missing from the approved intake fields. Preserve uncertainty. Do not infer a legal deadline, offer advice, decide urgency or accept the matter. Mark the note for staff review.
The expected note preserves “a hearing next week” without converting it to a calendar date. The request for a Thursday call is a preference, not a safe response timetable. A trained person must handle the enquiry through the firm’s established escalation process.
Make the handoff visible
For each enquiry, identify the responsible person, when it entered the queue and whether someone has reviewed it. A polished acknowledgement is not evidence that a lawyer has seen the message. Use wording approved by the practice about receipt, next steps and the limits of the contact.
Keep reminders and escalation checks independent of an AI summary where possible. If the tool fails, produces an empty note or cannot read an attachment, the original enquiry still needs attention. Staff should be able to find the source and correct the summary.
Check common failures before a limited trial: an ambiguous date, two people with the same name, a message outside normal hours, an unsupported language and a request outside the firm’s work. Use invented scenarios and the practice’s approved handling rules.
Judge the trial by reliability
Count omissions, invented details, inappropriate responses and the total time staff spend checking and correcting. If a standard form and a human review are clearer, keep that simpler process. Speed is useful only when the handoff remains dependable.
If you are seeking legal help yourself rather than running intake, see using AI to prepare questions for a lawyer. An assistant can help you organize questions, but the professional relationship and legal advice come from the qualified person you consult.
Reviewed September 30, 2026. Prepared with AI assistance and checked against the sources linked in this article. Examples are illustrative. Read our editorial and information disclaimer.
Illustrative photo by Nasser Eledroos on Unsplash.
