AI Billing Ethics: Can Lawyers Bill for AI Time?

The one duty with no real disagreement

Across the sixteen formal opinions and ten guidance documents on the tracker, billing is where the authorities speak with one voice. The hourly bargain is time for money; AI changes how much time work takes, not whose gain that is. Under Rule 1.5 and its state analogues, an hourly bill that includes time a machine eliminated is a bill for hours not worked, which is to say a false statement, before it is anything else.

What is billable is the human layer: composing prompts for the matter, reading and correcting output, running the verification workflow, exercising the judgment that turns a draft into work product. Texas Opinion 705 and ABA Opinion 512 both describe this affirmatively; the time you actually spend supervising the machine is real lawyer time. What is not billable is the counterfactual: the six hours the research used to take. North Carolina’s 2024 FEO 1 gives the formulation every training session should quote, because no one forgets it: a lawyer cannot charge three hours for a task AI helped complete in one.

Learning time and tool costs

Two second-order questions recur. First, who pays for the learning curve? The ABA and Kentucky’s E-457 agree: maintaining technological competence is the lawyer’s professional obligation and therefore overhead, not billable, unless a client specifically asks for tool-specific work and agrees to pay for it. Michigan’s FAQs, updated February 2025 specifically to add fee questions, apply the same logic with a concrete example: fifteen minutes of AI-assisted drafting is fifteen minutes on the bill.

Second, can the tools themselves be charged to clients? The pattern follows existing expense rules. General-purpose subscriptions (the firm’s AI drafting assistant, like its research platform) are overhead baked into rates. Matter-specific, per-use costs may be passed through only where the engagement letter permits, the amount is reasonable, and the client agreed in advance; Kentucky requires advance written agreement, and Arizona’s guidance adds advance written disclosure of AI costs. The quiet trap is double recovery: charging a client for an AI tool while also billing the hours the tool was bought to save invites exactly the fee-reasonableness scrutiny Rule 1.5 exists for.

Virginia LEO 1901: the fee-design answer

The most economically interesting instrument on the tracker is Virginia’s Legal Ethics Opinion 1901, effective November 24, 2025, and unusual nationally for carrying the Supreme Court of Virginia’s approval. It confronts the question firms actually worry about: if AI makes us faster, must our fees shrink proportionately? Its answer separates billing mechanics from fee structure. Bills tied to time must reflect actual time, full stop. But Rule 1.5 does not mandate proportionate fee reductions when AI accelerates work, and value-based, flat, and other alternative arrangements may legitimately capture efficiency gains.

The strategic consequence is bigger than Virginia: the ethical way to monetize AI efficiency is to change the unit you sell. A flat fee priced on value, disclosed and reasonable at the outset, lets the firm keep speed gains; an hourly arrangement hands them to the client by design. Firms that adopt AI seriously and keep pure hourly billing are choosing margin compression; the instruments do not forbid that choice, they just forbid hiding from it on the bill.

What this means for your firm’s policy

The billing section of a firm AI policy needs four sentences, enforced: bill actual time only; never bill time saved; learning time is overhead absent specific client agreement; AI costs pass through only with advance written agreement. Then have the fee-design conversation separately, with LEO 1901 on the table, because the answer to AI economics lives in engagement letters, not timesheets. The policy template ships with the four sentences; the engagement-letter decision is yours and your counsel’s.