Can you bill a client for AI time? What ABA Opinion 512 actually permits

Yes, but only for the minutes you actually spent. ABA Formal Opinion 512 lets you charge an hourly client for the time you spend prompting the tool and the time you spend checking what it gives back. Nothing for the hours it saved. The two questions that matter more to a small firm are the ones most coverage skips: whether your monthly subscription can go on a client's bill, and what happens to a flat fee when the same work now takes an hour. On the first, the answer is usually no. On the second, Opinion 512 says your fee may have become unreasonable.

The hourly rule, stated precisely

Opinion 512 works through an example rather than a principle, which makes it unusually easy to apply. A lawyer uses a tool to draft a pleading and spends fifteen minutes inputting the relevant information. The lawyer may charge for those fifteen minutes, and for the time spent reviewing the resulting draft for accuracy and completeness. That is the whole of it.

The rule underneath comes from Formal Opinion 93-379, which Opinion 512 quotes: a lawyer who has agreed to bill on hours expended does not fulfil her ethical duty if she bills for more time than she actually expended. Efficiency does not create billable hours. It removes them.

Texas said the same thing in Opinion 705, issued in February 2025, borrowing the D.C. Bar's formulation that where hourly billing has been agreed, the economies must inure to the benefit of the client. If you bill by the hour and your tools genuinely work, your revenue per matter falls. Any adviser telling you otherwise is selling something.

Your subscription is overhead. A per-use charge is not.

This is where most of what is written about AI billing goes wrong. The test Opinion 512 sets is functional. To the extent a tool or service functions like equipping and maintaining a legal practice, its cost is overhead, and you do not charge the client for it absent a contrary disclosure made in advance.

The opinion gives both ends of the spectrum. A grammar checking feature built into your word processor is overhead. A third party service used to review thousands of contracts for one particular client, charged to you on a per-use basis, is ordinarily billable to that client as the actual out-of-pocket expense you incurred. Texas Opinion 705 lands in the same place and adds the ceiling: you generally may not recover more than the amount you actually paid the provider.

A $30 monthly seat that you use on every file is the first example, not the second. You can move it out of overhead by disclosing the arrangement in advance and getting agreement, which Opinion 512 permits, but then you own two problems. There is no honest per-matter figure to allocate. And Opinion 512 is explicit that you may not add a surcharge to actual cost and must pass along any discount the provider gave you. Most solo and small firms should book the seat as overhead, price it into their rates, and stop there.

You cannot bill for learning the tool

Opinion 512 is blunt on this. A lawyer may not charge a client to learn how to use a tool the lawyer will regularly use for clients, because maintaining competence in the tools you use is your own obligation. The opinion frames it as a species of the older rule that you may not charge for time necessitated by your own inexperience.

There is one carve out. If a client explicitly asks you to use a specific tool you do not know, it may be appropriate to bill for gaining that knowledge, provided you and the client agree the billing terms before you start. That fits a corporate client mandating its own platform. It does not cover your Saturday afternoon working out better prompts.

The flat fee trap, which is the one that will actually bite

Almost every article on this subject is about the billable hour. Most solo and small firm revenue is not. Wills, conveyancing, immigration applications, uncontested matters and fixed price commercial work are quoted as a number, and that is precisely where Opinion 512 points its sharpest paragraph.

The Rule 1.5(a) reasonableness factors apply to flat and contingent fees as much as hourly ones. If a tool lets you complete tasks much more quickly than before, the opinion says it may be unreasonable to charge the same flat fee as when you worked without it, and it cites a Maryland disciplinary decision for the proposition that a fee charged for which little or no work was performed is an unreasonable fee.

Read carefully, that is not an instruction to reprice everything downwards in line with your stopwatch. Rule 1.5(a) values skill, the difficulty of the question, the result obtained and the customary local rate, none of which are hours. What it does mean is that if your flat fee is only ever justified by the effort it used to take, you have lost the justification. Write the engagement letter so the fee describes what the client is buying and why it is worth that, and you have an answer when someone asks.

The failure mode nobody writes about: using AI to justify your fee

In February 2024, Judge Engelmayer in the Southern District of New York cut a fee application in J.G. v. New York City Department of Education from the $113,484.62 requested to $53,050.13. Part of the reduction came from ordinary billing problems: duplicate review of the same records, and 3.9 hours booked to a three page boilerplate complaint.

The part worth memorising is what happened when the firm cited ChatGPT-4 to support its hourly rates. The court called the invocation utterly and unusually unpersuasive, noted that the firm had not identified the inputs the tool relied on, had not revealed whether any of them were imaginary, and had not shown that the tool considered the line of authority in that district rejecting those very rates. The firm was advised to excise references to ChatGPT from future fee applications.

Statutory fee shifting is small firm territory. Special education, civil rights, employment, housing, consumer work. If your income depends on fee applications, the market evidence supporting your rate needs to be declarations and comparable awards, and the tool needs to be nowhere near it.

The English version of this risk is quieter and just as expensive. On the standard basis under CPR 44.3, the court allows only costs that were reasonably incurred and are proportionate, and resolves any doubt in favour of the paying party. A costs draftsman looking at eleven hours of drafting on a matter your opponent knows was substantially machine assisted has an easy submission to make.

What to put in the engagement letter

• One sentence stating that AI tools may be used, that time spent operating and reviewing them is billed as time, and that the cost of general subscriptions is absorbed as overhead.
• If you ever intend to pass on a per-use charge, say so before the engagement begins, charge it at cost, and pass on any discount you received.
• For flat fees, describe the scope and the outcome the fee buys. Do not describe it in hours you no longer spend.
• Record prompting and review time as ordinary time entries, with the same specificity you would give a phone attendance. Vague entries are what got the J.G. firm cut, before anyone reached the AI question.

The uncomfortable part of Opinion 512 is not the prohibition. It is that the honest application of it reduces what an hourly firm can charge for the same output, and the profession has not yet decided what to do about that. Our walkthrough of Opinion 512 covers the duties beyond fees, the disclosure question is separate and answered differently, and the policy template has the clauses to drop into your engagement terms.

We keep a running breakdown of which AI tools are worth a small firm's money in our guide to the best AI tools for lawyers. The free guide covers the confidentiality and verification workflow that sits underneath all of this.

FAQ