Lesson objective
Write session objectives that name an observable accounting behaviour, and align each with an activity that practises it and a piece of evidence produced in the session.
Plain-language explanation
Why this matters now
Once the behaviour and the scope are set, three things must line up: the objective you state, the activity learners do, and the evidence that shows they can do it. When they do not line up, a session can be well delivered and still teach nothing that reaches the desk. The objective says "verify", the activity is a lecture, and the evidence is a feedback form.
The sources give you the standard the behaviour has to meet, and each binds its own population.
Professional guidance: IRS OPR Alert 2026-19, 24 June 2026, for practitioners before the IRS, says due diligence under section 10.22 "requires verifying the accuracy of facts, citations, and calculations produced by AI," and lists among its best practices documenting AI usage and verification processes. The alert creates no new rule.
Professional standard: AICPA SSTS section 1.4, effective 1 January 2024, for AICPA members providing tax services: judgment and care when relying on a tool (1.4.3), and the member remains responsible for the completed work product (1.4.7). It prescribes no verification steps.
Professional guidance: Texas State Board of Public Accountancy, Board Report vol. 168, August 2026, a newsletter article that applies to Texas only. It recommends considering documentation of how AI was used, how results were validated and how conclusions were reached, without documenting every interaction. For attest work it states that "AI-generated analyses do not constitute audit evidence" and do not replace required procedures.
These give you the "standard" part of an objective. The objective's wording, the activity and the evidence are yours to design, as practice.
A question to sit with
If every learner walked out holding one piece of paper they produced in your session, what would it need to show for you to be confident they could do the behaviour on Monday?
What goes wrong
1. The objective cannot be observed. "Understand the risks of AI" or "be aware of hallucinations." Nobody can watch someone understand.
2. The activity practises something else. The objective is to verify citations; the activity is a discussion of whether AI will replace accountants. The discussion may be lively, and it rehearses nothing.
3. The evidence measures the wrong thing. A multiple-choice question asks what "hallucination" means. The behaviour is catching one. Module 10 treats assessment design in depth; here the point is only that the evidence must match the objective.
4. The standard is missing. "Check the AI output" with no standard for what counts as checked. Checking one citation of five and checking all five look the same on a slide.
5. The record is forgotten. The behaviour includes a record, but the session never asks learners to produce one, so the first time they write it is on real client work.
The practical answer
Best practice: write objectives with an observable verb, a condition and a standard. The verb says what the learner does; the condition says with what; the standard says how well.
| Weak objective | Stronger objective |
|---|---|
| Understand AI hallucinations | Given a fictional AI-drafted research note with five citations, check each against the primary source and mark which do not support the statement made |
| Be aware of confidentiality risks | Given a fictional client file and the firm's approved tools list, identify which tool may receive which documents and state why |
| Know how to review AI output | Given a fictional AI-drafted variance commentary and ledger extract, recompute each figure and correct any that do not agree |
| Appreciate professional responsibility | Complete the firm's review record for one AI-assisted draft, naming what was checked and who reviewed it |
Best practice: build the alignment table before the slides. For each objective, write the activity that practises it and the evidence it produces. If a row has no activity, the objective is not being taught. If a row has no evidence, you cannot tell whether it was learned. If an activity has no row, it probably does not belong.
| Objective | Activity that practises it | Evidence produced in the session |
|---|---|---|
| Check five citations in a fictional research note | Pairs, then individually, timed | The note with each citation marked verified or not, and the reason |
| Recompute figures in fictional variance commentary | Individual, with a fictional ledger extract | The corrected commentary with each recomputation shown |
| Complete the review record | Individual, on one practised item | A completed review record naming the reviewer and what was checked |
Best practice: make the evidence the record the firm uses. The best evidence is often the very record the behaviour produces at the desk: the verification marks, the review sign-off, the documentation of how AI was used and checked. OPR's best practices and the Texas article both point toward documenting use and verification. When learners produce that record in the session, they rehearse the behaviour and you see whether they can do it.
Best practice: set the standard so it can be judged. "Every citation checked against the primary source" can be judged. "Output reviewed carefully" cannot. Where no source sets the standard, say it is the firm's standard or this programme's recommendation.
AI Coalition Network recommendation: no more than three objectives for a sixty-minute session. Each needs an activity long enough to practise and a piece of evidence you can look at. It is our recommendation, not a requirement, and it follows from the time each row needs.
Remember this
- An objective needs an observable verb, a condition and a standard.
- Build the alignment table of objective, activity and evidence before the slides.
- Make the evidence the record the firm uses at the desk: verification marks, the review record, the documentation of use and checks.
- Where no source sets the standard, label it as the firm's standard or this programme's recommendation.
- For Texas licensees, the Board's article says AI-generated analyses are not audit evidence; never model otherwise.
Accounting example
Ines Varga, a fictional CPA and firm trainer at the fictional Calloway Brandt CPAs, inherits a session whose objectives read "understand AI risks", "be aware of confidentiality" and "appreciate the importance of review". The activities are a video, a discussion and a ten-question quiz on definitions. She rewrites the objectives. First: given a fictional research note with five citations, check each against the primary source and mark which do not support the statement. Second: given a fictional client file and the firm's approved tools list, identify which tool may receive which documents. Third: complete the firm's review record for one practised item. She builds the alignment table. The video has no row, so she cuts it. The discussion becomes a ten-minute debrief of the citation exercise. The quiz is replaced by the three pieces of evidence the activities produce: the marked note, the tools decision sheet and the review record. She times the rows: twenty minutes for citations, fifteen for the tools decision, ten for the review record, and fifteen for the opening demonstration and the debrief, which makes sixty. The fictional note contains five citations, two of which do not support the statement made. Of 18 learners, 13 mark both, 4 mark one and 1 marks neither, which makes 18. Ines does not present this as a success rate; she uses it to decide which learners the tax manager should pair with on their first AI-assisted research notes. She labels the standard "every citation checked against the primary source" as the firm's standard, consistent with OPR Alert 2026-19 for practitioners before the IRS.
Responsible-use warning
Never state an objective nobody can observe, and never let an activity practise something other than the objective. Never use a recall quiz as evidence of a behaviour. Never write a standard that cannot be judged, and never attribute to SSTS section 1.4 a verification standard it does not contain. Never design an audit activity in which an AI output is treated as audit evidence; for Texas licensees the Board's article says it is not.
Human-review requirement
The instructor owns the alignment of objectives, activities and evidence. Before delivery, the commissioning partner reviews the one-page alignment table and confirms that each objective leads to the behaviour the firm wants and that the evidence uses the firm's real records. A second reviewer confirms that each cited source is correctly labelled and that the standards attributed to the firm are the firm's. The engagement owner or signing preparer remains responsible for the real work the behaviour will touch.
Practical exercise
Build an alignment table for a fictional session. Use a fictional firm, fictional learners and fictional material throughout.
- Write three weak objectives of the kind often seen in AI sessions for accountants.
- Rewrite each with an observable verb, a condition that names fictional material, and a standard that can be judged.
- For each rewritten objective, write the activity that practises it and the evidence it produces in the session.
- Mark any activity in your draft session that has no row, and decide whether to cut it or give it a row.
- Time each row and check that the total fits a sixty-minute session.
- For each standard, write where it comes from: a named source with its label and bound population, the fictional firm's procedure, or AI Coalition Network recommendation.
- Confirm that at least one piece of evidence is the review record the fictional firm uses at the desk.
- Write the two sentences you will say to the commissioning partner when you show them the one-page alignment table, explaining what each column is for.
- Check that you have no more than three objectives for sixty minutes, and if you have more, decide which one to move to a later session and why.
Knowledge check
-
What three elements make an objective observable and judgeable?
Answer guidance: An observable verb saying what the learner does, a condition naming the material, and a standard saying how well it must be done.
-
What does the alignment table reveal when a row has no activity or no evidence?
Answer guidance: No activity means the objective is not being taught; no evidence means you cannot tell whether it was learned.
-
What documentation does the Texas board article recommend considering?
Answer guidance: Documentation of how AI was used, how results were validated and how conclusions were reached, without documenting every interaction; it applies to Texas only.
-
Why is the firm's own review record good evidence for a session?
Answer guidance: It is the record the behaviour produces at the desk, so producing it in the session rehearses the behaviour and shows whether the learner can do it.
Sources and further reading
- Internal Revenue Service, Office of Professional Responsibility (OPR) — Introductory Guidelines for Responsible AI Use in Federal Tax Practice (Alerts from the Office of Professional Responsibility, Issue Number 2026-19) (2026-06-24). Accessed 2026-09-15.
- American Institute of Certified Public Accountants (AICPA), Tax Executive Committee — Statement on Standards for Tax Services No. 1, section 1.4, Reliance on Tools (2023-11-13). Accessed 2026-09-15.
- Texas State Board of Public Accountancy — AI and the Texas CPA: Professional Obligations in a Changing Environment (Texas State Board Report, Vol. 168, August 2026) (2026-08-01). Accessed 2026-09-15.
- AI Coalition Network — Certification Educational Standard: Human Review of AI-Assisted Professional Work. Accessed 2026-09-15.