In short
- Some features use AI models from Anthropic (Claude) and, for search in q, OpenAI.
- AI never decides anything about a person on its own.
- AI-drafted text is shown to a person to review before it is published, or — in the q assistant and the gslk coach — is labelled as AI-generated.
- The q assistant always says it is an AI assistant.
- Neither we nor our AI providers use your data to train AI models.
Where AI is used
| Product | Feature | What is sent to the model | Who sees the result |
|---|---|---|---|
| vox | Drafting event descriptions and social posts | Event details; speakers' names, roles, companies and short biographies; the organiser's sample texts | The organiser, who edits it before publishing |
| vox | Translation | Text the organiser chooses to translate, such as event content and speaker biographies | The organiser; published after review |
| vox | Tone-of-voice analysis | The organiser's own sample texts | The organiser |
| vox | Report insights | Aggregated statistics, event titles and feedback question labels — no free-text answers | The organiser |
| q | Answering questions | The visitor's question and the closest entries from the organisation's knowledge base | The visitor; AI-drafted answers are labelled |
| q | Search | The visitor's question, and the organisation's knowledge-base entries when they are created or edited, turned into numeric vectors by OpenAI | Nobody — used only to find matches |
| q | Preparing knowledge-base entries | Web pages the organisation chooses; emails it forwards, sent to Anthropic as received so that the model can remove or generalise personal details — only that de-identified text is then turned into a search vector by OpenAI | Staff, before anything is published |
| scito | Translation | Form and email texts staff choose to translate | Staff |
| gslk | AI coach | The user's name, club, the page they are on and the conversation (up to 20 turns), including what the page adds to it — such as the form the coach is helping to fill in | The user; programme administrators can read conversations held in a club context |
| gslk | Competency suggestions | The experience the student wrote (title, organisation, type and description) and the club's list of competencies | The student, who keeps or removes each suggestion before saving; the exchange is kept in the coach's log like any other |
| gslk | Translation drafts | Text a user chooses to translate | The user, as a draft |
In vox, attendees' registration data is never sent to AI models.
Providers and your data
| Provider | Used for | Where | Training on your data | Retention by the provider |
|---|---|---|---|---|
| Anthropic Ireland, Limited | Drafting, translation, coaching, answers | USA | No | Up to 30 days; longer only for content flagged for abuse |
| OpenAI Ireland Ltd | Search vectors in q | USA | No | Abuse-monitoring logs up to 30 days |
Transfers to the USA rely on the Standard Contractual Clauses in each provider's data processing addendum; see Sub-processors.
How we meet the EU AI Act
- Telling people they are talking to an AI. The q assistant identifies itself as an AI assistant from the first message, and the gslk coach is named and labelled as an AI coach wherever it appears (Article 50(1)).
- Marking AI-generated content. Text an AI wrote carries a machine-readable mark in the page code, next to a visible AI label: q's answers, the gslk coach's replies and competency suggestions, and vox's drafts, tone-of-voice analysis and report insights. The gslk coach launched after 2 August 2026, so it is marked from launch; for vox's features, which were on the market before that date, the deadline is 2 December 2026 (Article 50(2)). Translations are not marked: a translation keeps the meaning of the person's own text, and the European Commission's guidelines on Article 50 treat AI translation as standard editing, which the marking duty does not cover. Once an organisation edits and publishes a draft, the published text is its own and carries no mark.
- Human review. Drafts in vox, scito and gslk are proposals that a person reviews, edits and publishes; the publishing organisation holds editorial responsibility.
- No high-risk uses. Our features do not score, rank or make decisions about people in education, employment or access to services, and our Terms (section 9.6) do not allow customers to use them for the high-risk purposes listed in Annex III to the AI Act without our written agreement.
- AI literacy. We take measures to support the AI literacy of the people who build and run these features (Article 4).
What organisations must do
Customers decide whether to use AI features, review AI output before publishing it, keep the q assistant's AI disclosure visible, and label AI-generated text they publish without human review (Terms, section 9).