AI-Generated Workshop Agenda: 5 Checks Before You Run It

Check five things before you run an AI-drafted agenda: the blocks add up to the session length with breaks, every fixed point of your brief sits where you put it, each method fits your group and is not used where its write-up warns against it, the agenda says where each method comes from, and the participant version stands without your notes. The first two usually hold. In the State of AI Facilitation 2026 agenda test, 8 of 9 agendas drafted by three AI assistants added up to the workshop length, breaks included, and the same eight kept every fixed point of the brief. The other three checks are where those drafts fell short.

Drafting with AI is common practice. In SessionLab's State of Facilitation 2025, a survey of 1,050 respondents run from September to November 2024, preparing sessions was the most common use of generative AI, named by 85.8% of the AI users who answered that question. In the same survey, 25.9% of those AI users used it to design the full agenda (SessionLab 2025).

The State of AI Facilitation 2026 tested AI assistants; it did not survey facilitators. The report sent 3 briefs to Claude, ChatGPT and Perplexity with web search off and had the 9 resulting agendas scored blind by an AI rater against a written sheet. Read them as an early signal from a small sample. The test instruction did not ask for sources, method warnings or questions, so the results show what the drafts contain by default. A second rating agreed with the first in 65% of 40 compared scores, below the protocol's 70% threshold (State of AI Facilitation 2026).

The five checks for an AI-generated agenda

Each check takes a few minutes with the draft and your brief side by side. Run them in this order, because the first two tell you whether the draft is worth editing at all.

1. Timings: add up every block, breaks and transitions included, and compare the total with your start and end time.

2. Fixed points: find every element your brief fixed (a sponsor slot, lunch, the end time) and confirm it sits exactly where you put it.
3. Method fit: for each method, compare the group size it was written for with your group, and read its "when not to use" list against your situation.
4. Sources: check that each method names where it comes from, so you can read the original write-up.
5. Participant version: read only the lines participants will see and check they make sense without your facilitator notes.

The report puts the method-fit question as "For each method, is there a 'when not to use' in its literature, and does it apply?" In its test, sources were the largest gap (0 of 9 agendas gave one) and method warnings the second (State of AI Facilitation 2026).

Do the timings add up?

Usually yes, and confirming it takes two minutes. In the report's agenda test, 8 of 9 agendas added up to the workshop length, breaks included, with every block timed. The ninth, Perplexity's answer to a deliberately vague brief, announced a four-hour workshop and planned three; it was the only zero among 60 scores across the 9 agendas (State of AI Facilitation 2026).

Write the duration of each block in a column next to the draft. Include breaks, the walk back from a break and any setup the method needs. Sum the column and compare it with the session's start and end time, then check that the agenda states the total itself, so whoever runs it can see the arithmetic.

Then look at each block against its method. A block shorter than the method's minimum is a timing problem even when the total is right. Decide in advance which block gives way when the session runs late, and write that into the facilitator notes. For how the blocks of a session usually fit together, see how to structure a workshop.

Are your fixed points where you put them?

In the test they were. 8 of 9 agendas kept every fixed element exactly as briefed, no agenda dropped or moved one, and the ninth contradicted its own stated length (State of AI Facilitation 2026). In the report's German brief the fixed elements were lunch at 12:30, a guest slot without discussion, an end time on the hour and the formal "Sie" (report).

The weaker spot is what the brief left open. Of 108 agenda answers the report collected from three assistants, 1 asked a clarifying question before it planned (State of AI Facilitation 2026). Given the vague brief in the agenda test, none of the 3 benchmark agendas asked anything; all three stated assumptions, a four-hour session and a team of 6-12, and planned on them (report). The report adds that its test instruction invited exactly that, so the stronger evidence is the assistant survey, where no instruction was given and 1 of 18 answers to the vague request asked before planning (State of AI Facilitation 2026).

So list what the draft assumed that your brief did not say. Length, group size, who decides and what the session must produce are the usual candidates. Correct each assumption in the draft, or in the brief if you redraft.

Does each method fit your group and your decision?

After sources, this is where the drafts fell short most often. In the report's test, 7 of 9 agendas used a method in a setting its library write-up warns against and did not say so. Five times it was a dot vote shaping a binding decision with the leadership in the room (State of AI Facilitation 2026). Twice it was a clustering run faster than its write-up allows, once alongside a one-word check-in round; in the published scores, that case is a six-minute silent affinity clustering plus a one-word round for a group that needed to process. Only 2 of 9 agendas scored full marks on respecting "when not to use" (report).

Group size is easier to check and less often wrong. Group size was stated for every block in 4 of 9 test agendas (report). Where a question to the assistants stated a group size, only 4 of 144 answers recommended a method outside its library size range (State of AI Facilitation 2026).

For each method, open its write-up and note three things: the group-size range, the duration range and the "when not to use" list. The Dot Voting write-up lists four settings in which not to use it, among them "The decision is high-stakes or binding" and "Strong hierarchy is present and votes are public". The Affinity Map write-up warns against running it in less than 10 minutes, because rushed clustering skips the discussion. Not every method in the library has such a list, so a missing one means you judge the fit yourself.

The report's advice for the block that decides something is direct: write the method's "when not to use" into the facilitator notes yourself (State of AI Facilitation 2026). The method can often stay; the warning and the fallback need to be on the page you will read during the session.

Does it say where each method comes from?

Expect it not to. In the report's test, 0 of 9 agendas wrote down where a method comes from, although all nine named recognisable methods. The instruction did not ask for sources, so this is a default (report).

A source line matters because the original write-up holds the conditions the method was designed for. 1-2-4-All is a Liberating Structure developed by Henri Lipmanowicz and Keith McCandless, and the original write-up sets a minimum group size of four and times the whole sequence at about 15 minutes. The affinity diagram is attributed to Jiro Kawakita in the 1960s, according to Gamestorming. In the Sociocracy 3.0 pattern, a consent proposal passes when it is "good enough for now and safe enough to try until the next review".

Add one line per method: its name, where it comes from and a link to the write-up you checked in step three. If an assistant supplies sources when you ask, open each one and confirm it describes the method you are running.

Does the participant agenda stand on its own?

Often it does not. Participant text was cleanly separated from facilitator notes in 5 of 9 test agendas (State of AI Facilitation 2026).

Copy only the lines participants will see into a separate document and read them as someone who has never heard of the methods. Each block should say what people will do and what comes out of it, in plain words. Method names, facilitator cues ("hold the silence", "watch the energy") and timing notes for you belong in the facilitator version. Times, breaks and the end time belong in both.

A hypothetical worked check

Picture a 3-hour quarterly planning session for 14 people. The sponsor has fixed a 15-minute context slot right after the opening, stays for the whole session, and the group has to leave with three binding priorities. An assistant returns the draft below. The library columns come from each method's write-up in the 750+ method library on this site; the draft and the session are invented for this example.

Minutes	Block in the AI draft	Group size in the library	Duration in the library
0-10 min	One Word Check-In	2-50	5-15 min
10-25 min	Sponsor context (fixed by the brief)	n/a	n/a
25-40 min	1-2-4-All: what got in the way last quarter	4 or more	12 min or more
40-75 min	Brainwriting: candidate priorities	5-15	30-45 min
75-90 min	Break	n/a	n/a
90-110 min	Affinity Map of the candidates	1-20	10-90 min
110-125 min	Dot Voting for the top priorities	2-100	5-20 min
125-160 min	Consent Decision-Making on the shortlist	3-12	15-60 min
160-180 min	Round Robin of owners and next steps	3-20	5-30 min
Total: 180 min

Run the five checks on it.

1. Timings pass. The blocks sum to 180 minutes with a 15-minute break, and every block sits inside its method's duration range.
2. Fixed points pass. The sponsor slot sits at minute 10, as briefed, and the session ends on time.
3. Method fit fails twice. Consent Decision-Making is written for 3 to 12 people, and its write-up warns against groups larger than about a dozen, so 14 is outside its range. Dot Voting fits 14 people, but two of its warnings apply: the priorities are binding, and the sponsor is in the room while dots go up in public.
4. Sources fail. The draft names seven methods and says where none of them comes from.
5. The participant version fails. It reads "1-2-4-All, then affinity clustering", which means nothing to a participant.

The fixes stay inside the 180 minutes. Swap Consent Decision-Making for Gradients of Agreement, which the consent write-up names as one option for larger groups (the other is delegating to a smaller circle) and which the library sizes at 3 to 30 people and 10 to 30 minutes. Run it for 30 minutes and give the freed 5 minutes to the closing round, which brings Round Robin to 25 minutes, still inside its range. For the dot vote, write two lines into the facilitator notes. Dots go up at the same time, and the sponsor votes last or not at all. The dot picture is a shortlist; the decision is made in the next block. Add a source line per method. Rewrite the participant line as "Think alone, then in pairs, then in fours: what got in the way last quarter?"

When to stop editing and redraft

Edit when the failures are local: one method to swap, one warning to write down, participant wording to fix. Redraft when a failure sits in the brief the assistant worked from. A wrong session length, a moved fixed point, or a deciding block whose method warning applies with no alternative that fits the time all mean the draft was built on the wrong premises, and patching blocks will not fix that.

For a redraft, give the assistant a fuller brief: the length, the group size, every fixed point, the decision the session must produce and who makes it. Ask for a source and a "when not to use" note for each method, since the report's test instruction asked for neither (State of AI Facilitation 2026). The workshop planning guide walks through the objective, the audience and the session length step by step, which is most of such a brief. The ChatGPT prompts for workshop planning post covers how to write that request.

A draft that passes all five checks has a timed column that sums to the session length and your fixed points where you set them. Each method carries its group-size range, its warning where one applies and one source line. The participant agenda stands apart from the facilitator notes. Workshop Weaver exports a plan as a facilitator handout and as a separate participant agenda, so the fifth check is a read of the participant export.

Take the next agenda an assistant drafts for you, put a column of durations next to it, and run the five checks before anything goes to participants.