The hourly answer is fairly clear and unwelcome. The non-hourly answer is genuinely unsettled, and that is where the argument is.
Start a conversation with the AI Adoption Concierge, already scoped to billing for AI-assisted work. Pick a starting point, or describe your situation directly.
For hourly billing the principle is not really in dispute: a lawyer bills time actually expended, and work that took one hour is one hour regardless of what it would have taken before. ABA Formal Opinion 512 addresses fees among the duties generative AI engages, and its treatment reflects that principle. The difficulty is that applying it honestly cuts revenue on precisely the work firms do most of. That pressure is what pushes the question toward non-hourly arrangements — and there the position is genuinely open, with jurisdictions already reaching different conclusions about whether a flat fee may remain unchanged when the effort behind it has collapsed.
The impact is very uneven, which is why firm-wide answers tend to fit nobody.
Among the most exposed. High volume, previously time-intensive, now dramatically faster.
Substantially accelerated, with the verification step partly offsetting the saving.
Standard correspondence and first drafts of familiar instruments.
Far less exposed. The time is in the thinking, which does not compress the same way.
Largely unaffected in the moment, though preparation compresses.
Unaffected, and arguably more valuable as the routine work commoditises.
How firms are working through it.
Both an ethics exposure and a revenue question, and they pull in opposite directions.
Virginia has indicated that charging the same non-hourly fee for AI-assisted work is not per se unreasonable. ABA Formal Opinion 512 cautions that a flat fee may be unreasonable where AI drastically reduced the effort. A multi-state firm cannot adopt one position and assume it travels.
Verification is real work actually performed, and firms generally treat it as billable on the same basis as reviewing a junior's draft. What draws scrutiny is verification time that looks disproportionate to the task, or that appears to have been inflated to recover the time the tool saved. The narrative matters here: a bill showing substantial review of AI-generated research invites a question, and the answer that it was necessary is much stronger when the firm has a documented verification process it applies consistently rather than an explanation constructed afterwards.
Practice varies and the ethics guidance does not resolve it universally, but the trend among firms that have decided is toward transparency, for a practical reason as much as an ethical one: a client who discovers it later, particularly during a fee dispute, reacts far worse than one who was told at the outset. Where disclosure has been made in the engagement letter, the bill narrative can be ordinary. Where it has not, a bill that reveals it for the first time is an awkward place for the conversation to start.
Take it seriously as an instruction and get it in writing, including its scope — clients frequently mean "not for drafting my documents" rather than a blanket prohibition covering spell-check and legal research platforms. Then work out whether the firm can comply, which is harder than it sounds if AI is embedded in research tools and the document system. A firm that agrees to a restriction it cannot actually honour has created a much larger problem than the one it avoided.
Estimates vary widely and should be treated as directional. Some analyses put the figure on the order of 240 hours per lawyer per year, though that depends heavily on practice mix and on how completely the tools are adopted — and firm-level realisation is consistently lower than per-task studies suggest, because verification, adoption gaps and uneven use absorb much of it. The number matters less for planning than the distribution: it is concentrated in specific matter types, and that is where pricing needs attention first.
Describe your practice mix and how you bill. The Institute will help you map it.