An AI timesheet for lawyers that suggests, and never bills on its own

What the rules ask of a time record

Model Rule 1.5(a) says a lawyer may not charge an unreasonable fee, and the time actually spent is one of the factors that decides reasonableness. A time record is evidence of that time. It has to be accurate, contemporaneous enough to be trusted, and specific enough that a client or a court can tell what was done. Two things follow for any automated timesheet.

• It must not invent work. A suggestion should come from something that happened in the matter: a document opened and read, a draft revised, an email or a calendar entry. If the software cannot point to the activity, it should not suggest the time.
• A lawyer must decide what is billed. Software can measure minutes; only the lawyer can judge whether those minutes were reasonably necessary for the client.

Generative AI adds one more rule of thumb. ABA Formal Opinion 512 addresses fees for AI-assisted work: when you bill hourly, you bill the time you actually spent, including time reviewing and correcting AI output, not the time the tool saved you. A timesheet that marks AI-assisted work helps you show that later.

What every Draftiro account does today

When you ask the AI a question or request a draft inside a matter, Draftiro creates a draft time entry for that matter, tagged "AI-assisted". The minutes are the elapsed time of the AI exchange, which may differ from the time you actually spent, so treat it as a reminder, not a measurement. From the matter's Time panel you can edit the description and minutes, mark the entry billed after review, or delete it and log your real time as a manual entry. Quick research with no matter attached creates no entry. The time tracking guide walks through it.

The AI daily timesheet (limited pilot)

The pilot replaces those per-chat drafts with a daily view built for the end-of-day reconstruction most solos dread.

1. Activity, not guesses. While a matter's documents, drafts or chat are open, Draftiro records active minutes per matter, at most one surface per minute. Reading time, the part of review work that leaves no other trace, is counted only while you are actually in the matter.
2. Suggestions with evidence. At the end of the day you see suggested entries per matter and task, each listing the activity behind it: the files read, the drafts revised, the calendar entries. Nothing is a billable entry yet.
3. Plain narratives. Each suggestion gets a short narrative from fixed templates, such as "Draft/revise motion to compel", built from the evidence labels. The narrative is not written by a language model, so it cannot describe work that did not happen. Edit it before approving.
4. Tenth-hour rounding. Time rounds up to the next 0.1 hour, once per entry, never per activity.
5. Codes and rates. Suggestions carry UTBMS task and activity codes you can change. Rates resolve from the matter and the timekeeper, and the rate is fixed on the entry when you approve it.
6. Approval is the only way in. An entry becomes billable only when an attorney approves it. Approved and billed entries export to CSV or LEDES 1998B for your billing system.

Firm owners see approved entries, not a member's raw activity minutes or pending suggestions. If you want to try the pilot, write to support@draftiro.com.

A daily routine that keeps the record honest

1. Before you close the day, open the suggestions or the Time panel while the work is still fresh.
2. Reject anything you do not recognize. A suggestion you cannot explain is not billable.
3. Rewrite narratives so a client can tell what was done, without disclosing privileged detail in an invoice line.
4. Bill review of AI output as your time; never bill time the tool saved.
5. Keep a note of time that was real but not chargeable instead of erasing it, so the record stays complete.

For the wider ethics picture, read Formal Opinion 512, explained and the Opinion 512 glossary entry.