Legal Tech and AI

Training Staff to Work Alongside AI Tools

A four-week plan to train law firm staff on AI tools: role-based permissions, a one-page policy, supervised drills and how to check the training worked.

AssistCall research teamPublished Reviewed by Faruque Amin, VP of Sales and Marketing6 min read

On this page
  1. The training gap
  2. What the ethics rules expect
  3. Train by role
  4. A one-page AI policy staff will actually read
  5. A four-week rollout
  6. Drills that build the right habits
  7. Practical skills worth teaching
  8. Train the supervisors too
  9. Handle the people side
  10. Keep the policy current
  11. Check that training worked
  12. Frequently asked questions
  13. References
Legal Tech and AI: Training Staff to Work Alongside AI ToolsArticle coverLEGAL TECH AND AITraining Staff to WorkAlongside AI ToolsAva for LawyersInsightsLegal Tech and AI: Training Staff to Work Alongside AI ToolsArticle coverLEGAL TECH AND AITraining Staff toWork Alongside AIToolsAva for Lawyers

Key takeaways

  • Most professionals see uses for generative AI, yet 64% have had no training on using it for their work[1].
  • ABA Formal Opinion 512 ties supervisory duties to training staff on the ethical and practical use of AI tools and their risks[2].
  • Train by role: what each person may use AI for, what they must check and what they must never enter into a tool.
  • Judge training by review results and error rates, not by attendance.

The training gap

Thomson Reuters found that 89% of professionals see use cases for generative AI, but 64% have received no training on using it for their industry work[1]. Among legal professionals who already use it, the top tasks are document review, legal research and document summarization, each cited by more than 70% of users[3]. Those are exactly the tasks where an untrained user can pass an error along without noticing it.

Professionals and generative AI training, 2025
Professionals and generative AI training, 2025Comparison. See use cases for generative AI: 89%; Received no generative AI training for their work: 64%.89%See use cases for generative AISee use cases for generative AI: 89%64%Received no generative AI training for theirworkReceived no generative AI training for their work: 64%Professionals and generative AI training, 2025Comparison. See use cases for generative AI: 89%; Received no generative AI training for their work: 64%.89%See use cases for generative AISee use cases for generative AI: 89%64%Received no generative AI trainingfor their workReceived no generative AI training for their work: 64%

Source: Thomson Reuters, 2025[1]

Show the data
ItemValue
See use cases for generative AI89%
Received no generative AI training for their work64%

The gap is not only about skill. Staff who have not been told what is allowed tend to fall into one of two habits: avoiding the tools entirely, or using unapproved ones quietly on their own devices. Both are worse for the firm than a clear policy and a few hours of practice.

What the ethics rules expect

ABA Formal Opinion 512 says managerial lawyers must set clear policies on the permissible use of generative AI, and supervisory lawyers must make reasonable efforts to ensure that lawyers and nonlawyers comply with their professional obligations when using it. Supervision includes making sure staff are trained on the ethical and practical use of the tools relevant to their work and on the risks. The opinion suggests training can cover the basics of the technology, its capabilities and limitations, ethical issues and good practice for secure data handling, privacy and confidentiality[2].

State guidance adds detail. California's bar says lawyers must not put confidential client information into a generative AI tool that may present material risks to confidentiality or security, absent informed client consent[4]. Your training needs to turn that principle into concrete rules for each tool the firm actually uses, because staff cannot apply a principle they have never seen applied.

Train by role

Role-based AI permissions and checks
RoleMay use AI forMust always checkNever
Receptionist or intakeReading call summaries and drafting replies to booking requestsCallback numbers, names for conflicts and urgency flagsAnswer legal questions or paste caller details into unapproved tools
ParalegalFirst drafts of routine letters, document summaries and chronologiesEvery date, name, amount and citation against the sourceSend AI output to a client without attorney review
AssociateResearch starting points, outlines and issue spottingThat every authority exists and says what the draft claimsFile or serve anything with unverified citations
Partner or supervisorReviewing output and approving templates and toolsThat staff follow the policy, by sampling work regularlyApprove a new tool without a confidentiality review

Keep the table to one page and pin it where people work. When a new tool is approved, add a row rather than writing a separate memo, so there is always one place to look.

A one-page AI policy staff will actually read

  • The approved tools list, with what each one may be used for.
  • What never goes into any tool unless it is approved for that data: Social Security numbers, medical records, privileged strategy, and anything under a protective order.
  • The disclosure rule: when clients or callers must be told that AI is involved.
  • The verification rule: a person checks every fact, date, name and citation before anything leaves the firm.
  • The billing rule: bill actual time spent, including reviewing AI output, and never time saved.
  • How to report an AI error and who to tell, with a promise that reports are welcome rather than punished.

A four-week rollout

  1. Week 1: why and what

    A short session on the policy, the approved tools and real examples of AI errors in legal work. Cover the limits, not only the features.

  2. Week 2: supervised practice

    Each person completes three tasks from their own role using the tool while a supervisor reviews both the output and the checks they made.

  3. Week 3: error-finding drills

    Give staff AI output with planted errors, such as a wrong date, an invented case or a misspelled party name, and score how many they find.

  4. Week 4: sign-off

    Each person shows they can complete their role's checks unassisted. Record the sign-off and the date.

  5. Every month after that

    A short slot in a team meeting on one new error, one new use and any policy change.

Drills that build the right habits

Example drills for a fictional six-person family law firm:

  1. Summary check: staff compare an AI call summary with the call recording and mark every difference.
  2. Citation check: associates verify the citations in an AI research memo, one of which does not exist.
  3. Redaction check: paralegals prepare a document for an AI summary tool and remove the details the policy bars.
  4. Escalation check: intake staff read three summaries and decide which one needs an attorney today.
  5. Client explanation: each person explains in two sentences how the firm uses AI and what a person still checks.

Score the drills and repeat the weakest one a month later. The goal is a habit of checking, which only forms through repetition on realistic material.

Practical skills worth teaching

Policy tells people what is allowed. Skills make the allowed uses safe and worth the effort. Cover these in the supervised practice week, using the firm's own templates and anonymized past matters:

  • Give the tool the source material and ask it to work only from that, rather than asking open questions from memory.
  • Ask for output in a fixed structure, such as dates in one column and sources in another, so checking is quick.
  • Ask the tool to list what it was unsure about, then check those points first.
  • Keep a record of the prompt and the source documents with the matter, so a reviewer can retrace the work.
  • Stop and ask a supervisor when the output looks confident about something the sources do not say.

Train the supervisors too

Partners and senior associates often skip training because they assume they will only review. Reviewing AI-assisted work is its own skill. A reviewer has to know what kinds of errors these tools make, such as invented citations, confident summaries of documents that say something slightly different, and dates pulled from the wrong place. Give supervisors the same error-finding drill as everyone else, plus a short checklist for sampling work: one AI-assisted item per person per week, checked against its sources, with the result recorded.

Handle the people side

Attitudes are mixed. Half of law firm professionals say they are excited or hopeful about generative AI, while 10% see it as a significant threat to revenue[1]. Some staff will also worry about their own jobs. Be direct: say which tasks are changing, which are not, and what new work you want from the time freed up, such as faster callbacks or better client updates.

Pick one or two respected staff members as tool leads. They answer quick questions, collect errors and suggest policy changes, which keeps adoption practical rather than top-down. Give them a little protected time each week; a tool lead who has to do it on top of a full caseload will quietly stop.

Keep the policy current

  1. Review the approved tools list every quarter and remove anything nobody uses.
  2. Recheck vendor terms on data use and retention at each renewal, since they change.
  3. Add every reported error to the drill library so new staff learn from it.
  4. Rerun the sign-off for anyone whose role changes or who starts using a new tool.
  5. Record each policy change with its date and the reason, and tell staff in the next team meeting.

Check that training worked

Attendance proves nothing. Track what supervisors find when they review AI-assisted work, and look at the trend over the first quarter.

  • Errors found in supervisor reviews, by type.
  • Share of AI-assisted work that passes review the first time.
  • Policy breaches reported, and how quickly they were reported.
  • Time from finding an error to fixing the template, prompt or process behind it.

If your firm uses an SI Employee such as Ava on the phones, include its call summaries in the summary-check drill so staff learn which details to rely on and which to verify. For new intake hires, fold this training into a 30-day onboarding plan. The billing rule is explained in billing for AI-assisted work under Opinion 512, and teams that are not ready for AI can start with rules-based intake automation.

Frequently asked questions

Do nonlawyer staff need AI training?

Yes. ABA Formal Opinion 512 ties supervisory duties to making sure both lawyers and nonlawyers are trained on the ethical and practical use of the AI tools they use.

How long should initial AI training take?

A few hours spread over four weeks works better than one long session: a policy briefing, supervised practice, error drills and a sign-off.

What should staff never put into an AI tool?

Anything the firm's policy bars for that tool, typically confidential client details, identification numbers, medical records and privileged strategy, unless the tool is approved for that data.

How do we know training is working?

Watch review results: fewer errors found, more work passing first review, and faster reporting of problems.

References

  1. Thomson Reuters (2025). Generative AI Adoption Nearly Doubles as Professional Services Reach Crossroads (opens in a new tab) โ†‘

  2. American Bar Association, Standing Committee on Ethics and Professional Responsibility (2024). Formal Opinion 512: Generative Artificial Intelligence Tools (archived copy) (opens in a new tab) โ†‘

  3. Thomson Reuters Institute (2025). 2025 GenAI report: Executive summary for legal professionals (opens in a new tab) โ†‘

  4. The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) โ†‘

Book a demo

See Ava handle your calls.

Book a short demo. Weโ€™ll walk through your calls and workflow, and agree what Ava should take care of.

Book a demo