AI-Assisted Work and the Billable Hour: What a Firm Should Record Before It Bills

By Troels Johannesen

A note before I start: I am not a lawyer, and nothing in this article is legal advice. I run a software agency in Copenhagen and I build a time-tracking tool for agencies, so I have a commercial interest in how people record their working hours. Where I refer to bar guidance, I link the original text so you can read it yourself; for your own obligations, your bar’s ethics resources are the right place to go.

Most time records are built on a simple picture: one person, one task, one stretch of time. That picture held up for a long time because it matched how professional work happened.

It matches less well now. More and more legal work is drafted, summarised or reviewed partly by AI tools. The person still does real work, but the shape of that work has changed, and the time record often has not.

I see the same gap in agency work. It is common for a task to be started by a person, pushed forward by a tool, and finished by a person checking and correcting the result. If the time sheet only shows one line with one name and one duration, it hides most of what happened. In law, that gap touches billing disputes and professional ethics. My view, from the outside, is that much of the risk here is a record-keeping problem, and record-keeping problems can be fixed.

What the guidance actually says about fees

The American Bar Association addressed this directly in Formal Opinion 512, Generative Artificial Intelligence Tools, issued July 29, 2024. On fees, it is plain: lawyers who bill an hourly rate “must bill for their actual time.” It gives an example. If a lawyer spends 15 minutes putting the relevant information into a generative AI tool to draft a pleading, the lawyer may charge for those 15 minutes and for the time spent reviewing the resulting draft for accuracy and completeness. What the lawyer may not do is bill for more hours than were actually spent because the work went faster.

The same opinion says a lawyer may not charge a client to learn how to use a tool the lawyer will regularly use for clients, because keeping up with one’s own tools is part of competence. The exception it describes is when a client explicitly asks for a specific tool the lawyer does not know, and even then it suggests agreeing the billing terms first and preferably writing them down. It also treats tools that work like general office equipment as overhead, while a per-use charge from a third-party service for one client’s matter may ordinarily be billed as an expense at actual cost.

The Florida Bar reached similar conclusions in Ethics Opinion 24-1, dated January 19, 2024. It says a lawyer “may not ethically engage in any billing practices that duplicate charges or that falsely inflate the lawyer’s billable hours,” and that the efficiency AI brings “must not result in falsely inflated claims of time.” It also says a lawyer should be careful not to charge for time spent developing minimal competence with generative AI, and that if the actual cost of a tool for a particular client cannot be determined, the lawyer should not prorate subscription charges but should treat them as overhead.

The State Bar of California’s practical guidance on generative AI, updated in 2026 and now also addressing agentic AI, puts it in one line I find useful: a lawyer may charge for actual time spent, “e.g., crafting or refining generative AI inputs and prompts, or reviewing and editing generative AI outputs,” and hourly billing must reflect the time actually spent.

None of this is exotic. It is the old rule, bill what you actually did, applied to a new kind of tool. To follow it, you need records that show what you actually did.

What a useful record contains

From my side of things, a time entry for AI-assisted work should be able to answer four questions.

Who started the work? A named person, not “the system.” Even when a tool produces the first draft, someone asked for it and is responsible for the result. ABA 512 and the California guidance both make clear that responsibility stays with the lawyer.

Which tool was used, and for what? Not a technical log, just enough that a reader can tell a tool drafted, summarised or reviewed something. ABA 512 mentions, as a training suggestion, marking materials produced by generative AI as such when they are stored in a client or firm file. A short note in the time entry does a similar job for billing.

How much time did a person spend, and on what? This is the part that matters most for the bill. Preparing the input, reviewing the output, checking citations, correcting errors and rewriting passages are all human time. Write them down as separate activities where you can, with honest durations.

What was billed for each part? Human time at the agreed rate. Any tool cost that is passed on as an expense, at actual cost and only where the engagement allows it. Nothing for the time the tool saved.

A client or bar regulator reading that record a year later should be able to follow it without asking you to remember.

Review is real work, so record it honestly

There is a temptation in two directions: under-recording review because it feels like “just reading,” or padding it because the drafting went suspiciously fast. Both are mistakes. Careful review of AI output is real legal work. The California guidance describes it as part of the duty of competence: reviewing, verifying and correcting AI-generated outputs. Florida’s opinion compares it to reviewing the work of a paralegal. If reviewing a draft properly takes forty minutes, that is forty minutes of work, and recording it accurately protects both the client and the firm. If it took ten, record ten.

Flat fee or hourly when AI speeds things up

When work gets faster, hourly billing means a smaller bill for the same result. The guidance does not let a firm avoid that by recording phantom hours.

Florida’s opinion suggests that firms may want to consider contingent fees or flat rates for specific services “so that the benefits of increased efficiency accrue to the lawyer and client alike.” California’s guidance mentions flat fees for more predictable assignments. But a flat fee is not a free pass. ABA 512 says that if a tool lets a lawyer finish tasks much more quickly, it may be unreasonable to charge the same flat fee as without it.

Whichever model you choose, two things help. First, say it in the engagement letter: how fees are calculated, whether AI tools are used, and whether tool costs may be billed as expenses. Second, keep time records anyway, because a flat fee can still be questioned for reasonableness, and records show the work was done and reviewed by a person.

A one-page audit checklist for small firms

Take one recent matter where AI tools were used and walk through it:

1. Can you name the person who started each AI-assisted task?
2. Can you tell, from the record alone, which tool was used and what it produced?
3. Is human review time recorded as its own activity, with an honest duration?
4. Does the billed time match the time people actually spent, with nothing added for time the tool saved?
5. Is any time spent learning a routinely used tool kept off the client’s bill?
6. Are tool costs either treated as overhead or billed at actual cost, with the basis explained to the client in advance?
7. Does the engagement letter say how AI use affects fees and costs?
8. For flat fee work, would the fee still look reasonable to the client if they saw how the work was done?
9. Could a client or regulator follow the record a year from now without your help?

If you can answer yes to all nine, you can show what a person did, what a tool did and what you billed for each. If not, the fix is usually not a new policy document. It is a better time entry, written on the day the work happened.

Sources:

1. American Bar Association, Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512, “Generative Artificial Intelligence Tools,” July 29, 2024. https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf (Section F, Fees, pp. 11 to 14; Section E, footnote 52 on marking GAI-produced materials.)
2. The Florida Bar, Ethics Opinion 24-1, January 19, 2024. https://www.floridabar.org/etopinions/opinion-24-1/ (Full text checked in The Florida Bar’s own Technology Packet, https://www-media.floridabar.org/uploads/2025/11/Technology-Packet-updated-11-14-25.pdf, sections “Oversight of Generative AI” and “Legal Fees and Costs.”)
3. The State Bar of California, Standing Committee on Professional Responsibility and Conduct, “Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law” (2026 version, replacing the November 2023 guidance). https://www.calbar.ca.gov/Portals/0/documents/ethics/Generative-AI-Practical-Guidance.pdf (Sections on competence, and “Charging for Work Produced by Generative AI and Generative AI Costs.”)

About the author

Troels Johannesen is the founder of WeCode, a software agency in Copenhagen, Denmark, and of Hourtick, a time tracking and task tool for agencies that logs the hours of people and AI agents on one board. He writes about how teams record, review and bill work as AI tools become part of everyday practice. He is not a lawyer.

Featured image: cottonbro studio / Pexels. This is a contributed editorial article; the views are the author’s own and it is not legal advice.