# Literature search protocol: AI in workshop and meeting facilitation

Status: written 2026-09-26 before the first query. Changes after that are dated amendments at the bottom.

## Why

The report's first draft reviewed six hand-picked sources and said it had read "every public survey". That was not true. This protocol makes the review systematic, so the report can say exactly what was searched, what was found and what was left out.

## Question

What does published evidence say about (a) how facilitators use generative AI, (b) how well AI plans, runs or supports facilitated group sessions (workshops, meetings, deliberations, design sprints, retrospectives), and (c) what AI does to group idea generation and decision quality?

## Sources searched

1. **OpenAlex** (api.openalex.org), all works, 2022-01-01 to 2026-09-26.
2. **Semantic Scholar** (api.semanticscholar.org, graph API), same window.
3. **Google Scholar**, first 50 results per query, as a cross-check for items the two APIs miss.
4. **Grey literature**: web search for practitioner surveys and industry reports from facilitation associations (IAF, INIFAC, IAP2), training firms, meeting and whiteboard vendors (e.g. Miro, Mural, SessionLab, Butter, Klaxoon), consultancies and research institutes, plus German-language sources.
5. **Backward and forward citation check** on every included academic study (references and "cited by" in OpenAlex).

## Queries (run in English; grey literature also in German)

- ("facilitation" OR "facilitator" OR "workshop" OR "meeting") AND ("large language model" OR "LLM" OR "generative AI" OR "ChatGPT" OR "AI assistant")
- ("group ideation" OR "brainstorming" OR "collective creativity") AND ("large language model" OR "generative AI")
- ("deliberation" OR "deliberative") AND ("large language model" OR "AI facilitator" OR "AI moderator")
- ("agenda" OR "meeting planning" OR "workshop design") AND ("AI" OR "LLM")
- ("design thinking" OR "design sprint" OR "retrospective") AND ("generative AI" OR "LLM")
- German grey literature: ("Moderation" OR "Workshop" OR "Facilitation") AND ("KI" OR "ChatGPT") AND ("Umfrage" OR "Studie")

## Inclusion

- Published 2022 or later (generative AI era), in English or German.
- Empirical: a survey of practitioners, an experiment, a field study, a systematic review, or an evaluation of an AI system in a facilitated group setting.
- Reports its method and sample size.
- Peer-reviewed papers, preprints (flagged as such) and practitioner surveys with a stated method are all eligible; each is labelled by type.

## Exclusion

- Opinion pieces, vendor marketing without data, how-to guides.
- Studies of one-to-one tutoring, customer service chatbots or individual writing that say nothing about groups or facilitators (Doshi & Hauser stays in as the closest published test of group-level effects, labelled as such).
- Education studies of classroom use, unless the setting is a facilitated group session.

## Screening and recording

- Every hit from sources 1–3 is logged in `literature/search-log.csv` with query, database, date, title, year, venue, DOI/URL, type, and a screening decision (included / excluded + reason).
- Included items are read in full where available; each usable number goes into the claim ledger with page, method and sample size.
- Items that could not be accessed in full are listed as such and not quoted beyond their abstract.

## What the report may then say

- The number of records found, screened and included, per source, and the date of the search.
- That the search covers the listed databases and queries, not everything that exists.

## Amendments

(none)