Lesson objective
Write a behaviour statement that names a trigger, an observable action and a standard, and choose which behaviour a session should target.
Plain-language explanation
Why this matters now
Module 1 argued that a slide lecture changes little of what a lawyer does on Monday, because it delivers information and the working failure sits in decisions and routines. This module is the constructive answer. It sets out a way of designing a session that begins at the end: with the specific thing a participant will do differently at their desk, and works back from there to the practice, the explanation and the evidence.
The method is not a research result. It is Best practice in the plain sense: a way of working whose logic can be followed and checked by anyone who uses it. Its value for an instructor of lawyers is that every element of the session has to justify itself against a named behaviour, which is the same discipline a lawyer applies to every paragraph of a brief: what is this doing, and for whom?
The stakes are practical. An AI session that does not name its behaviour tends to become a tour of tools and a list of risks. The duties that make the training necessary are about conduct. Professional guidance: ABA Formal Opinion 512 (29 July 2024) addresses what lawyers do when using generative AI, including competence, confidentiality, supervision and fees. It is guidance interpreting the ABA Model Rules, which each jurisdiction adopts in its own form, and it does not bind a lawyer by itself.
A question to sit with
Finish this sentence for the last AI session you delivered or attended: "After this session, when ______ happens, a participant will ______, instead of ______." If you cannot fill all three blanks, what was the session for?
What goes wrong
1. The topic stands in for the behaviour. "A session on AI and confidentiality." A topic tells you what the slides will be about. It says nothing about what anyone will do.
2. The behaviour is invisible. "Participants will think critically about AI output." Thinking cannot be seen. If you cannot see it, you cannot rehearse it in the session or know later whether it happened.
3. The behaviour has no trigger. "Participants will verify citations." When? A behaviour with no trigger has no moment in the working day to attach to, and so no moment at which it is recalled.
4. The session targets eight behaviours. Everything seems important, so everything is included. A sixty-minute session that rehearses eight behaviours rehearses none of them properly.
5. The behaviour contradicts the firm. The session teaches a routine the firm's tools, policy or deadlines do not support. Participants who try it on Monday meet a wall and conclude the training was theoretical.
The practical answer
Best practice: write a behaviour statement in five parts.
When [trigger], [who] will [observable action] [to what standard], instead of [what they do now].
Each part does work.
| Part | Its job | Weak | Strong |
|---|---|---|---|
| Trigger | Ties the behaviour to a real moment | "When using AI" | "When an AI-assisted draft contains any cited authority and is about to leave the team" |
| Who | Names the role, because routines differ by role | "Lawyers" | "The associate who prepared the draft" |
| Observable action | Something another person could see or check | "Will be careful" | "Will open each authority in the firm's research service and compare the cited passage with the draft" |
| Standard | What counts as done | "Properly" | "Every authority, with the check recorded on the matter file" |
| Instead of | Names the current habit the session competes with | (left out) | "Instead of relying on the citation format looking correct" |
The "instead of" clause is the one most often omitted and the one that matters most. It forces the designer to find out what people actually do now, which is usually a sensible habit under time pressure rather than a failure of character.
Apply five tests to the statement.
- Observable. Could a colleague watching, or a record left behind, show it happened?
- Triggered. Is there a recognisable moment in the working week when it applies?
- Within their control. Can this role actually do it, with the access and authority they have?
- Supported. Do the firm's tools, policy and deadlines allow it? If not, the session is the wrong fix and the instructor should say so to whoever commissioned it.
- Accurate. Does the behaviour reflect the rules and guidance in the participants' jurisdiction, stated with the right label?
Best practice: choose the behaviour by consequence, frequency and gap. Most sessions can target one behaviour well and, at most, three. To choose, score each candidate. The scores below are illustrative, for a fictional firm; yours come from asking.
| Candidate behaviour | Consequence if not done | How often the trigger occurs | Size of the gap between now and the standard |
|---|---|---|---|
| Verify every authority in AI-assisted drafts before they leave the team | High | Weekly for litigators | Unknown until you ask |
| Check the approved-tool list before pasting client material | High | Daily for some roles | Unknown until you ask |
| Record AI-assisted time as time actually spent | Moderate to high | Every billing cycle | Unknown until you ask |
| Use the firm's prompt library for first drafts | Low | Daily | Small |
High consequence and a frequent trigger with a real gap is where a session earns its time. Low consequence with a small gap may be better served by a one-page guide.
Best practice: find the gap before you design. Ask three or four people in the target role what they do now when the trigger occurs. Look at the firm's policy and at what the approved tools actually allow. Ask a supervising lawyer what goes wrong. This takes an hour and it changes the design more than any amount of slide preparation, because it tells you what the "instead of" clause really says.
Best practice: write the behaviour statement on the first line of the session plan, above the title, and keep it visible while designing. Every later decision in this module is tested against it.
Remember this
- Design starts from the behaviour, not the topic: when [trigger], [who] will [observable action] [to what standard], instead of [what they do now].
- The "instead of" clause is the most often omitted and the most important; find out what people actually do before designing.
- Test the statement: observable, triggered, within their control, supported by the firm, and accurate for the jurisdiction.
- Choose one behaviour, at most three, by consequence, frequency and gap.
- A behaviour statement that misstates a rule multiplies the error; a licensed attorney confirms it.
Legal example
The fictional regional firm Corriveau Asplund LLP asks its training partner, Neve Halloran, for "a session on AI and confidentiality" for paralegals and legal assistants. Before opening any slide software she spends an hour finding the gap. She asks four paralegals what they do when they need to summarise a long document quickly. Two use the firm's approved tool. One uses a consumer chatbot on her phone, because the approved tool is not available on the laptop she uses at the courthouse. One did not know there was an approved tool. She reads the firm's policy and finds that it names the approved tool but does not say what material may go into it. She asks a supervising partner what worries him, and he says he never knows whether AI was used on work that reaches him. Neve drafts her behaviour statement: "When a paralegal needs to summarise, extract from or redraft client material with AI help, the paralegal will use only the firm's approved tool, for material the policy permits, and will note on the task that AI was used and for what, instead of using whatever tool is to hand and not recording it." She applies the five tests. It is observable, because the note on the task is a record. It is triggered. It is within their control, except at the courthouse, where the approved tool is unavailable, so she reports that to the IT director before the session rather than training people to do something they cannot do. It is not yet supported, because the policy is silent on what material may go in, so she asks the general counsel to add a three-line table to the policy. And it is accurate, confirmed by the general counsel as consistent with the firm's jurisdiction and its policy. The session title, written last, is "Before You Paste".
Responsible-use warning
Do not begin a session plan with a topic, and do not write a behaviour that the firm's tools, policy or deadlines make impossible. Training cannot fix a missing tool or a silent policy, and participants who try a routine that fails on Monday will treat the whole session as theory. Never write a behaviour statement that states a rule more broadly than it is, such as a universal duty to disclose AI use. Say which jurisdiction and which policy it reflects, and have it confirmed.
Human-review requirement
A licensed attorney confirms every behaviour statement that touches a professional duty before any session is built around it. The reviewer checks that the behaviour is consistent with the adopted rules and guidance of the participants' jurisdiction rather than with a model rule alone, that it matches the firm's current policy and approved tools, that any disclosure or consent element reflects when those are actually required, and that the named-reviewer element is described as the programme's educational standard or as firm policy, not as law.
Practical exercise
Write and test a behaviour statement for a fictional session. Use a fictional firm, fictional participants and fictional matters.
- Write down the topic you were asked, or would be asked, to teach, for example "AI and research".
- Interview, or imagine interviewing, three people in the target role. Record what each does now when the relevant trigger occurs.
- Draft a behaviour statement in the five-part form: when [trigger], [who] will [observable action] [to what standard], instead of [what they do now].
- Apply the five tests: observable, triggered, within their control, supported by the firm, and accurate for the jurisdiction. Write one line on each.
- For any test it fails because of tools or policy, write a short note to the person who commissioned the session explaining what needs to change first.
- List three other candidate behaviours and score all four on consequence, frequency and gap. Choose one, or at most three.
- Mark any element of the statement that depends on a professional rule or guidance, and write the question you will ask a licensed attorney to confirm it.
- Only now, write the session title.
- Show the statement to one person in the target role and ask whether the "instead of" clause describes what they really do. Revise it if not.
Knowledge check
-
What are the parts of a behaviour statement?
Answer guidance: A trigger (the real moment), who (the role), an observable action, the standard for done, and what the participant does now instead: when [trigger], [who] will [observable action] [to what standard], instead of [what they do now].
-
Why is the "instead of" clause so important?
Answer guidance: Because it forces the designer to find out what people actually do now, which is usually a sensible habit under pressure; the session has to compete with that habit, and designing without knowing it aims at an imagined problem.
-
What should an instructor do when a behaviour fails the "supported" test because the firm's tool is unavailable in some settings?
Answer guidance: Report it to whoever commissioned the session or can fix it before delivery, rather than training people to perform a routine they cannot perform; training does not fix a missing tool or a silent policy.
-
How should a session choose among several candidate behaviours?
Answer guidance: By scoring each on the consequence if it is not done, how often its trigger occurs, and the size of the gap between current practice and the standard, then targeting one, or at most three, where those are high.
Sources and further reading
- American Bar Association Standing Committee on Ethics and Professional Responsibility — Formal Opinion 512: Generative Artificial Intelligence Tools (2024-07-29). Accessed 2026-09-14.
- American Bar Association — Model Rules of Professional Conduct. Accessed 2026-09-14.
- AI Coalition Network — Certification Educational Standard: Human Review of AI-Assisted Professional Work. Accessed 2026-09-14.