Lesson objective

Audit a session outline item by item, cut every item no target behaviour needs, and keep any retained reference material accurate and dated.

Plain-language explanation

Why this matters now

Every AI session grows. A colleague adds a slide on the history of chatbots. The client asks for "something on the EU". Someone remembers a framework that "we should mention". Each addition is small and true. Together they push the practice out of the session.

Backward design only works if you are willing to cut. Once the behaviours are written (Lesson 1), and the objectives, activities and evidence line up (Lesson 2), every remaining item has to answer one question: which behaviour does this serve? If the answer is none, it goes, however interesting it is.

Cutting also protects accuracy. Material that no behaviour needs is material nobody checks. It is often where out-of-date dates and overstated rules hide, because nobody uses them on Monday and nobody notices they are wrong. Two examples show how easily this happens.

Law / rule (EU, international): Regulation (EU) 2026/1744, the Digital Omnibus on AI, published in the Official Journal on 24 July 2026 and in force from 27 July 2026, amended the EU AI Act's timetable for high-risk requirements. Chapter III, Sections 1, 2 and 3, except Article 6(5), now apply from 2 December 2027 for systems that are high-risk under Article 6(2) and Annex III, and from 2 August 2028 for systems that are high-risk under Article 6(1) and Annex I. A slide built before July 2026 carries the old dates. If that slide is in a session for a US social media team, it is probably serving no behaviour at all, and it is also wrong.

Voluntary framework: NIST AI Risk Management Framework 1.0 (NIST AI 100-1, released 26 January 2023). It is intended to be "voluntary, rights-preserving, non-sector-specific, and use-case agnostic". Its Core has four functions, GOVERN, MAP, MEASURE and MANAGE. As of 15 September 2026, NIST's web page states: "The AI RMF 1.0 is being revised as part of the White House AI Action Plan." No revised version had been published. A ten-minute tour of the framework in a copywriting session serves no Monday behaviour, and a slide that describes a revision nobody has seen would be inventing it.

A question to sit with

Which slide in your current deck would you most hate to cut, and which Monday behaviour does it serve? If you cannot name one, why is it still there?

What goes wrong

1. The history lesson. Ten minutes on the development of AI. No behaviour needs it.

2. The tool tour. Five tools demonstrated in turn. The learners use one of them. The comparison also drifts toward ranking products, which this programme does not do.

3. The framework recital. A framework is presented in full because it seems authoritative. The learners need at most one idea from it, such as defining who does what.

4. The out-of-scope regulation. An EU timeline appears in a session for a US team with no EU work, because a client once asked about it. If it stays, it must carry the amended dates.

5. The unchecked leftover. A slide nobody uses carries a date or a rule that has changed. Because no behaviour depends on it, nobody notices.

6. Everything is kept "for completeness". The instructor fears a question they cannot answer. The fear is better met with a parking list and a dated reference handout than with twenty more minutes of slides.

The practical answer

Best practice: audit the outline in a table, one row per item.

Item Behaviour it serves Would the learner fail without it? Decision
History of chatbots None No Cut
Checking claims against the substantiation file Check a claim Yes Keep and practise
Tour of five drafting tools None directly No Cut; show only the tool the team uses
NIST AI RMF overview Record a review (roles) Only the idea that human roles are defined Keep one sentence; move the rest to the handout
EU high-risk timeline None for a US-only team No Cut; if a reference is needed, use amended dates
Writing a disclosure into a creator brief Brief a creator Yes Keep and practise
IAB V2 overview Brief a creator Only where to look Keep one sentence and the procedural behaviour

Best practice: move, do not delete, what some learners will need later. A reference handout can hold the material the session cannot. It must meet a higher standard than the slides, not a lower one, because it will be read without you. Every item on it carries its source, what the source binds, and the date you checked it. For the EU AI Act, that means the amended high-risk dates and a note that whether the Act reaches a given organisation is a question for counsel. For NIST, it means "voluntary", "1.0, January 2023", "being revised as NIST stated in September 2026", and no claim that anyone can be certified to it.

Best practice: keep a parking list. When a good question falls outside the session's behaviours, write it on a visible list and promise a written answer or a referral after the session. This keeps the practice time intact and shows the room the question was heard. Questions about what a law requires of a particular client or campaign go on the list as referrals to counsel, never as answers.

Best practice: cut the framework to the one idea the behaviour needs. For a review behaviour, the one idea from NIST's Appendix C is that human roles in oversight "need to be clearly defined and differentiated." That sentence earns its place. The four functions, the trustworthiness characteristics and the Playbook do not, for this room.

Best practice: check what survives the cut. Everything left in the session is now something learners will use. That makes its accuracy matter more. Run every retained fact through the currency check in Module 03 before delivery.

AI Coalition Network recommendation: record what you cut and why. Keep the audit table with the session plan. When someone asks why the EU timeline or the tool tour is missing, the table answers. When a later instructor inherits the session, it stops the cut items creeping back in unchecked.

Remember this

  • Every item must name the behaviour it serves, or it is cut or moved to a reference handout.
  • Material nobody uses is where out-of-date dates hide; cutting protects accuracy.
  • EU high-risk dates were amended in July 2026: 2 December 2027 and 2 August 2028.
  • NIST AI RMF 1.0 is voluntary, carries no certification and was being revised as of September 2026; describe no revised content.
  • Keep a parking list and a dated reference handout instead of adding slides.

Marketing agency example

Tomasz Eberhardt, a fictional strategy director at the fictional Tallbridge & Rowe Communications, inherits a seventy-five-minute AI session for the agency's social team, which works only for US clients. The deck has thirty-two slides across twelve items. He writes the session's two behaviours first: check each claim in an AI-drafted post against the client's fact sheet, and record the named review before the post is scheduled. Then he audits the twelve items. The history of AI, a tour of five tools, a slide of prompt formulas and a ten-minute NIST AI RMF overview serve neither behaviour. He keeps one NIST sentence, that human roles in oversight need to be clearly defined and differentiated, because it supports the review behaviour. The EU AI Act slide gives high-risk dates of August 2026. That is doubly wrong for this room: no behaviour needs it, and Regulation (EU) 2026/1744 moved those dates to 2 December 2027 and 2 August 2028. He cuts it from the session and puts the amended dates in a reference handout with the note that scope is a question for counsel. A slide claims "NIST AI RMF 2.0 adds rules for marketing". No revised version had been published, so he deletes the claim entirely. He keeps five items: the two behaviours with their practice, a fictional-draft activity, the review record and the parking list. Five kept plus seven cut or moved makes twelve. The session now runs sixty minutes, fifteen under the slot, and he gives the spare time to a second practice round. He files the audit table with the plan so the next instructor can see why each item went.

Responsible-use warning

Never keep an item "for completeness" when no behaviour needs it, and never let a cut item survive unchecked in a handout. Never teach the 2024 EU high-risk dates; they were amended in July 2026. Never describe the content of a revised NIST AI RMF, never call the framework a legal requirement, and never suggest an agency can be certified to it. Never answer a client-specific legal question from the parking list.

Human-review requirement

The instructor owns the audit table, the retained content and the reference handout. Before delivery, a second reviewer reads the audit table beside the plan, confirms every retained item names the behaviour it serves, and checks every date and binding label in the handout against its source, including the amended EU high-risk dates and the voluntary, under-revision status of NIST AI RMF 1.0. Counsel owns questions of whether the EU AI Act reaches a particular client.

Practical exercise

Audit and cut a fictional session outline. Use a fictional agency, a fictional deck and fictional client work throughout.

  1. Write a fictional outline of ten to fourteen items for a seventy-five-minute AI session for one team, including at least one history item, one tool tour, one framework overview and one regulation from outside the team's work.
  2. Write the one or two Monday behaviours the session exists to install.
  3. Build an audit table with one row per item: the behaviour it serves, whether a learner would fail without it, and your decision.
  4. For each item you keep, write the evidence it produces or the behaviour it supports.
  5. For each item you move to a reference handout, write its source, what that source binds, and the date you checked it.
  6. Check any EU AI Act high-risk dates against the amended dates, and any NIST AI RMF statement against its voluntary, under-revision status.
  7. Count kept, cut and moved items and confirm they add up to the original total.
  8. Re-time the session and decide what to do with any minutes freed by the cut, giving priority to practice.
  9. Write a short parking list of three questions you expect, marking any that must be referred to counsel.

Knowledge check

  1. What two answers must an outline item give to stay in the session?

    Answer guidance: It must name the behaviour it serves, and a learner must be likely to fail at that behaviour without it.

  2. From when do the EU AI Act high-risk requirements apply after the July 2026 amendment?

    Answer guidance: From 2 December 2027 for Article 6(2) and Annex III systems, and from 2 August 2028 for Article 6(1) and Annex I systems.

  3. What may an instructor accurately say about the status of NIST AI RMF 1.0 as of September 2026?

    Answer guidance: That it is voluntary, offers no certification, and NIST stated it was being revised, with no revised version yet published.

  4. Why must a reference handout meet a higher accuracy standard than the slides?

    Answer guidance: Because it will be read without the instructor present, so every item needs its source, binding label and check date on the page.

Sources and further reading