AI for a small law firm: intake, documents, and what stays with the lawyer
Three in four small law firms now use AI and fewer than a third have seen revenue from it. The gap is not the tools. It is that the hour is still the unit, and a faster hour bills less. This is the work that pays, the work that does not, and the one thing the surveys say separates the firms that grew.
Small law firms use AI most for intake and for documents: drafting, summarizing, and first-pass review. Clio's 2026 Legal Trends data puts adoption at 71 percent of solo practitioners and 75 percent of small firms, with fewer than 33 percent seeing meaningful revenue gains and 86 percent of solo firms making no pricing change (Crossing Report on Clio). The work pays when the firm bills by outcome or flat fee, checks every citation, and keeps advice, filings, and privilege with the lawyer.
The efficiency paradox
The adoption numbers moved fast. The ABA's Legal Technology Survey Report found 30 percent of lawyers using AI-based tools in its 2024 edition, up from 11 percent the year before, with solo practitioners at 18 percent and firms of two to nine lawyers at 30 percent (LawSites, reporting on 512 responses). By Clio's 2026 Legal Trends data, 71 percent of solo practitioners and 75 percent of small firms use AI (Crossing Report; Clio's report draws on more than 1,700 respondents).
The revenue did not move with it. Fewer than 33 percent of those firms report meaningful revenue gains, and 86 percent of solo firms and 78 percent of small firms have made no pricing change despite the efficiency (Crossing Report). The arithmetic is simple. A task that took three billable hours and now takes one bills one, unless the fee changed. AI did what it said. The business model did not.
Intake: the work that pays first
Intake is the first place AI pays because it is not billed by the hour in the first place. A prospect fills a form or calls; the system captures the matter type, the parties, the urgency, and the conflict-check inputs; it routes the lead to the right person and books the consultation. Nobody at the firm was billing for that, so every hour it returns is pure capacity, and every lead that used to go cold on a Friday afternoon is a client the firm did not have.
The rules are the same as for any intake automation: it captures and routes, it does not advise, and anything that sounds like a deadline or an emergency goes to a person now.
Documents: drafting, summarizing, review
The second place is the document work, and it is where the accuracy problem lives. The ABA survey found 54 percent of respondents named saving time as AI's primary advantage, and 75 percent named accuracy as the top concern (LawSites). Both are right. A first draft of a routine agreement, a summary of a deposition, a first pass over a document set for the clause that matters: each of those is hours, and each is a draft.
The operating rule is that nothing the system produces is a work product until a lawyer has read it, and nothing it cites exists until a lawyer has opened the source. Tools that return the citation beside the claim make that check fast. Tools that return prose alone make it a trap.
What stays with the lawyer
- Advice. The system drafts an answer. The lawyer gives it.
- Filings. Every citation opened, every fact checked, a lawyer's name on it.
- Privilege and confidentiality. Decided before the first document goes anywhere.
- Anything a client will act on. A draft can be fast. The advice is the lawyer's.
Find the hours nobody bills.
The ten questions ask what gets retyped, what gets chased, and which leads go cold because nobody had time. For a small firm the answer is usually intake, and the result says whether that is the first project.
The pricing change the surveys point to
The firms that turned efficiency into revenue changed the unit. The Crossing Report's reading of Clio's data has 80 percent of firms facing pricing pressure from clients, 71 percent of clients already preferring flat fees, and firms running five or more integrated AI workflows growing revenue at twice the rate of firms running fewer (Crossing Report). Read those together: the client wants a price, the firm can now deliver the work in fewer hours, and a flat fee lets the firm keep the difference. The firm that automates and keeps billing hours has handed the gain to the client without being asked.
Confidentiality and the consumer chatbot
A lawyer pasting a client document into a free chatbot has disclosed it to a third party. Whether that breaches a duty depends on the tool's terms, the jurisdiction, and the matter, and the firm should not be finding out in a grievance. The practical rule is the one every regulated profession has arrived at: a written policy that names the permitted tools, whose terms exclude your content from training and state where data is stored, and names the prohibited ones. Then a permitted tool that does the job, so the policy is not asking anyone to work slower.
What we have and have not built
We have not built a system for a law firm, and this article does not claim otherwise. It is written from the ABA and Clio data and from the pattern in our own work: a repeated process with a trigger, one action, an exception path to a named person, and a log. The intake automation above is that pattern. Our patient presentation automation is the same pattern in a clinic, where the rule was that missing inputs stay visible instead of being filled in, which is the rule a law firm wants too.
What it costs
Intake and document tools are subscriptions: published guides put software at $50 to $300 a month per system, with usage adding $100 to $500 a month when a tool calls a model on your documents (AIessentials). Set that against the intake hours nobody billed and the leads that went cold, and the case makes itself. The pricing change is free and it is the one most firms have not made.
How to start
- Count the leads that went cold last quarter. That is the intake project's number.
- Write the confidentiality policy before any tool sees a client document.
- Automate intake first. Capture, route, book. No advice.
- Add document drafting with citations, and the rule that nothing is work product until a lawyer has read it.
- Change the fee. Flat or by outcome on the matter types the system made faster. Otherwise the client keeps the gain.
The tools work. The firms that grew are the ones that stopped selling the hour the tools removed.
What to hold on to
- Adoption is up, revenue is flat. Three in four small firms use AI; fewer than a third see revenue from it; most have not changed a price.
- Intake pays first. Nobody billed for it, so every hour returned is capacity and every saved lead is a client.
- Documents are drafts until a lawyer reads them. Accuracy is the top concern for 75 percent of lawyers surveyed. Open every citation.
- The lawyer keeps advice, filings, and privilege. The system drafts and routes. It does not advise, and it does not file.
- Change the fee or give away the gain. Clients prefer flat fees; firms with five or more integrated workflows grew twice as fast.
Questions owners ask us
How are small law firms using AI?
Mostly for intake and for documents: capturing and routing new matters, drafting routine agreements and correspondence, summarizing records, and first-pass review. Clio's 2026 data puts adoption at 71 percent of solo practitioners and 75 percent of small firms (Crossing Report). The ABA's survey found saving time was the primary benefit named by 54 percent of lawyers (LawSites).
Why do most small law firms not make more money from AI?
Because the hour is still the unit. A task that took three billable hours and now takes one bills one unless the fee changed, and 86 percent of solo firms and 78 percent of small firms have made no pricing change (Crossing Report). Clients already prefer flat fees, and firms running five or more integrated workflows grew revenue at twice the rate.
What is the best AI setup for law firm intake?
A form or phone flow that captures matter type, parties, urgency, and conflict-check inputs, routes the lead to the right person, and books the consultation. It captures and routes only. It does not give advice, and anything that sounds like a deadline or an emergency goes to a person immediately.
What should a lawyer never let AI do?
Give advice, file anything unreviewed, or cite anything a lawyer has not opened. Accuracy was the top concern for 75 percent of lawyers in the ABA survey (LawSites), and the operating rule follows from it: nothing the system produces is work product until a lawyer has read it, and no citation exists until the source has been opened.
Is it safe to put client documents into ChatGPT?
A free chatbot is a third party, and pasting a client document into one is a disclosure. Whether it breaches a duty depends on the tool's terms and the jurisdiction. The practical rule is a written policy naming permitted tools whose terms exclude your content from training and state where data lives, and naming the prohibited ones.
Has Prometheus built AI for a law firm?
No, and this article says so. It is written from the ABA and Clio data and from the pattern in our shipped work: one repeated process, a trigger, one action, an exception path to a named person, and a log. The intake automation described here is that pattern applied to a firm.
Sources
Survey figures are quoted from coverage of the ABA and Clio reports; Clio's full report is form-gated and its figures are cited through the Crossing Report. This is not legal advice. Checked September 5, 2026.
- LawSites, "ABA Tech Survey Finds Growing Adoption of AI in Legal Practice". March 7, 2025. The 2024 ABA Legal Technology Survey Report: adoption by firm size, top benefit, top concern, 512 responses.
- The Crossing Report, "Clio 2026 Legal Trends: AI and Revenue Data". April 18, 2026. Adoption, revenue gains, pricing changes, flat-fee preference, and the integrated-workflow comparison.
- Clio, "2026 Legal Trends Report for Solo and Small Law Firms". The source report, drawing on more than 1,700 respondents; full figures are form-gated.
- AIessentials, "AI Consultant Cost (2026)". Updated August 21, 2026. Software and usage costs.
- Patient presentation automation, the visible-gaps pattern from our record.