Lesson objective
Write a target behaviour for each role in a session that names the trigger, the action, the source checked against and the evidence left behind.
Plain-language explanation
Why this matters now
Module 01 described the people in your room: experienced, busy marketers who own different outputs. This module is about what you build for them. Most AI session plans start from topics: what generative AI is, the main tools, some prompt techniques, a few risks. Each topic is defensible. None says what anyone will do differently when they are back at their desk.
This module teaches the opposite order. Start from the behaviour: the thing a learner must do on Monday. Then work backwards to the practice that rehearses it, and only then to the information the practice needs. People often call this backward design. In this programme it is taught as practice. This programme holds no research on teaching methods, so nothing here claims that planning this way is proven to produce better learning. It is recommended because it forces every minute of a session to answer a plain question: what will someone do because of this?
For an AI marketing session, four behaviours come up again and again, because each is where AI-assisted work meets the agency's own exposure:
- Check a claim against the file before an AI-drafted claim leaves the agency.
- Brief a creator with a disclosure written into the brief, not left to the creator.
- Stop a send when consent is missing from the record, rather than assuming someone else checked.
- Record a review, with the reviewer's name, before a deliverable reaches the client or the public.
The second behaviour shows why these are agency behaviours and not only client ones. Professional guidance (FTC staff, not binding): "FTC's Endorsement Guides: What People Are Asking". The FAQ says the Guides do not have the force of law, but practices inconsistent with them can result in enforcement under section 5 of the FTC Act. It says: "Delegating part of your promotional program to an outside company doesn't relieve you of responsibility under the FTC Act." And it says intermediaries that recruit influencers may be liable if they "play a role in creating or disseminating endorsements containing representations you know or should know are deceptive." An agency that writes the creator brief is playing a role.
The fourth behaviour carries this programme's standard. AI Coalition Network recommendation: a named, qualified professional reviews and owns every AI-assisted deliverable before it reaches a client, a regulator or a filing, and the review is recorded. That is our educational standard, not a law.
A question to sit with
Take the last session plan you wrote or attended. For each item on it, finish the sentence "On Monday, because of this, a copywriter will..." How many items end with a verb you could watch someone do?
What goes wrong
1. The topic list. The plan is a list of nouns: "AI and copyright", "AI and privacy", "prompting". Nouns cannot be practised.
2. The unobservable verb. "Learners will understand disclosure requirements." Nobody can watch someone understand. The instructor cannot tell on the day whether it happened.
3. The behaviour with no trigger. "Check AI output" is attached to no moment. A behaviour needs a trigger: before this draft leaves, before this brief is sent, before this list is uploaded.
4. The behaviour the learner cannot perform. The session tells a junior email marketer to "stop any send without consent", but in their agency only the account lead can pause a campaign. The behaviour has to fit the authority the learner actually holds. If it does not, the behaviour is "raise a stop request to the account lead", and the session teaches that.
5. Policy presented as law. The session tells learners that the law requires a particular disclosure wording or consent step for their client. The behaviour is the agency's process. What a statute requires of a particular campaign is a question for counsel.
The practical answer
Best practice: write the behaviour statement in four parts. A trigger, an action, the source the action is checked against, and the evidence the action leaves behind. The evidence matters most for the instructor, because it is what you will look at in the session to see whether the practice happened.
| Behaviour | Trigger | Action | Source | Evidence left behind |
|---|---|---|---|---|
| Check a claim | Before an AI-drafted piece leaves the copy team | Mark each factual claim and find it in the file | The client's substantiation file or approved fact sheet | The draft with each claim ticked or struck |
| Brief a creator | Before a creator brief is sent | Write the disclosure and where it appears into the brief | The agency's disclosure policy and the client's policy | The brief with a disclosure section |
| Stop a send | Before a scheduled email or text campaign goes | Confirm the list's consent record; stop or raise a stop if it is missing | The agency's consent record for that list | A stop note or a recorded confirmation |
| Record a review | Before a deliverable reaches the client or public | Review it and record the reviewer's name and date | The agency's review checklist | The review entry in the job record |
Best practice: match the behaviour to the learner's authority. Before the session, ask the agency who can actually approve, pause or send. Write each behaviour at the level the learner holds. A junior who cannot pause a send can still write the stop note and escalate it; that is the behaviour you teach them.
Best practice: name the human role, not only the tool's role. Voluntary framework: NIST AI Risk Management Framework 1.0 (NIST AI 100-1, January 2023). Its Appendix C says human roles in decision making and oversight "need to be clearly defined and differentiated", and that human-AI configurations can span "from fully autonomous to fully manual." The framework binds no agency and is being revised, as NIST's page stated in September 2026. The point is still useful for planning: each behaviour statement should say which part the person does, which part the tool does, and who decides.
Best practice: keep the law out of the behaviour and the referral in it. The behaviour is the agency's process. When a learner asks whether a particular campaign's consent or disclosure meets a particular law, the behaviour is to write the question down and route it to the account lead and counsel. Teach that referral as part of the behaviour.
AI Coalition Network recommendation: one behaviour per role per session. A session that tries to install four behaviours in every role installs none. Choose the behaviour that matters most for each role in the room, and let the others wait for a later session.
For PR rooms, check whose code applies. Professional guidance (PRSA, for members): Ethical Standard Advisory #22, "AI Agents in Public Relations", July 2026. It states that "Public relations professionals are accountable for how the AI tools they use are designed, deployed, and supervised." It is advisory guidance for PRSA members, not law. If your room includes PRSA members, it is a useful anchor for a review behaviour; if not, describe it as one professional body's guidance for its members.
Remember this
- Start from what the learner will do on Monday, then work back to practice, then to information.
- A behaviour has a trigger, an action, a source and evidence left behind.
- Fit the behaviour to the authority the learner actually holds.
- The behaviour is the agency's process; what a law requires of a campaign goes to counsel.
- Backward design is practice in this programme, not a proven method.
Marketing agency example
Hollis Brandt, a fictional training lead at the fictional Northlight Creative Co., is asked to run a ninety-minute AI session for the influencer and email teams. The request arrives as a topic list: "AI and influencers, AI and email, AI risks, prompting." Hollis writes one Monday sentence per role instead. For the four creator managers: "When I finish an AI-assisted creator brief, I write the disclosure and where it must appear into the brief, against the agency's disclosure policy, and I leave a brief with a disclosure section." For the three email marketers: "When a scheduled send is loaded, I check the list's consent record, and I leave a confirmation or a stop note." Before finalising, she asks the agency who can pause a send. Only the two account leads can. She rewrites the email marketers' behaviour: "When the consent record is missing, I write a stop note and send it to the account lead before the scheduled time." For the two account leads she adds: "When a stop note arrives, I pause the send and record the decision." Four plus three plus two makes nine learners, each with one behaviour. During the planning call, an email marketer asks whether a fictional client's old sign-up form gives valid consent for a text campaign. Hollis notes that this is a question about what the law requires of a particular campaign, and she builds the answer into the behaviour: write the question down and route it to the account lead and the agency's counsel. She deletes "prompting" from the plan, because no Monday sentence needed it, and tells the requester why. She labels the plan's structure as her practice, not a proven method.
Responsible-use warning
Never plan a session as a list of topics with no behaviour behind them, and never write an objective with a verb nobody can observe. Never teach a behaviour the learner has no authority to perform; teach the escalation instead. Never present the agency's disclosure or consent process as what the law requires of a particular campaign. Never claim backward design is proven to produce better learning.
Human-review requirement
The instructor owns the behaviour statements and the session plan built from them. Before delivery, a second reviewer, such as the agency's head of training or an account lead, confirms that each behaviour fits the authority of the role it is written for, that each names a real source the agency holds, and that no behaviour presents agency policy as a legal requirement. Counsel owns any question of what a law requires of a particular campaign. The agency's named reviewers own every deliverable.
Practical exercise
Write the Monday behaviours for a fictional room. Use a fictional agency, fictional roles and fictional source documents throughout.
- Write a fictional request for an AI session that arrives as a topic list of four to six nouns.
- List the roles that will attend, with the number of people in each, and check that the numbers add up to the total attendance.
- For each role, write one Monday sentence: "When [trigger], I [action] against [source], and I leave [evidence]."
- For each behaviour, name the fictional source document the agency holds, such as a substantiation file, a disclosure policy or a consent record.
- Write down who in your fictional agency can approve, pause or send. Rewrite any behaviour a role has no authority to perform as an escalation.
- Find any behaviour that could be mistaken for a legal requirement and rewrite it as the agency's process, adding the referral to counsel.
- Delete every topic from the original request that no behaviour needs, and write one sentence for the requester explaining each deletion.
- Label the plan's approach as best practice or as an AI Coalition Network recommendation, and check that it claims no research finding about how people learn.
Knowledge check
-
What four parts does a Monday behaviour statement have?
Answer guidance: A trigger, an action, the source the action is checked against, and the evidence the action leaves behind for the instructor to see.
-
What should an instructor do if a learner has no authority to perform the behaviour the session was going to teach?
Answer guidance: Rewrite the behaviour at the level the learner holds, usually as an escalation such as writing a stop note to the person who can act.
-
Why is briefing a creator with a disclosure an agency behaviour and not only a client one?
Answer guidance: Because the FTC staff FAQ says delegating promotion does not relieve responsibility and intermediaries that help create deceptive endorsements may be liable.
-
Is backward design presented in this programme as a proven method?
Answer guidance: No. It is taught as planning practice; this programme holds no research on teaching-method effectiveness and makes no such claim.
Sources and further reading
- Federal Trade Commission — FTC's Endorsement Guides: What People Are Asking. Accessed 2026-09-15.
- National Institute of Standards and Technology — Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (2023-01-26). Accessed 2026-09-15.
- Public Relations Society of America (PRSA), Board of Ethics and Professional Standards — AI Agents in Public Relations, Ethical Standard Advisory (ESA) #22 (2026-07-01). Accessed 2026-09-15.
- AI Coalition Network — Certification Educational Standard: Human Review of AI-Assisted Professional Work. Accessed 2026-09-15.