AI disclosure — how this survey handles AI#
Survey responses are stored in EU-hosted infrastructure (Scaleway, Warsaw region) on servers operated directly by Crystallized Intelligence. They are never sent to third-party AI services. Analysis and aggregation for the published report run on a leading open-source model hosted on a Scaleway GPU (Paris region, EU), rented for a few hours and torn down once the work is done. The website itself was built collaboratively using Claude Code (Anthropic), but no respondent data passes through any external AI tool.
Where AI was used#
Building this site. The Astro pages, the survey form, and parts of the copy were drafted with the help of Claude Code, a coding assistant from Anthropic. A human (Aisling) reviewed and edited every line before it went live. No respondent data is sent to Anthropic or to any other AI provider, ever.
Where AI is not used#
Storing, transmitting, or processing your responses. Your answers go from your browser directly to our Scaleway database. No AI vendor sees them. No cloud AI API is called during submission or storage.
Where AI will be used during analysis#
When the field period closes (31 July 2026), aggregating the open-text responses — the workflow stories you share — is the most labour-intensive part of writing the report. For that step, we rent a Scaleway GPU instance in the Paris region (EU) for a few hours and run a leading open-source model on it. The model runs entirely on that rented instance: your data stays inside Scaleway’s EU infrastructure, is never sent to any commercial AI API, and the instance — along with any temporary working data on it — is destroyed once the analysis is complete.
Audit trail of AI-assisted decisions#
Every interpretive judgement the open-source model makes during analysis — how it groups workflow stories into themes, how it labels disappointments, how it aggregates similar responses — is logged. The log records the prompt sent, the model’s response, and any human edit applied before it reached the report. We publish this log alongside the report on 14 September so any reader can see exactly which decisions were ours, which were the model’s, and where we overrode it.
A report summarising what small teams told us about AI shouldn’t itself be opaque about how AI was used to write it.
Why this matters#
Many small organisations, especially NGOs handling sensitive work, cannot route their data through commercial AI platforms. We believe a survey about AI use in small teams should be held to the same standard those teams hold their own systems to. So we are.
Questions: info@crystallized.lu.
The journey your response takes
Three stages, one explicit exclusion, and a separate channel for your email.
If you ask for the report, your email is stored separately. Even if we wanted to, we could not link it back to your response.
- EU infrastructure (Scaleway)
- Open-source model
- Excluded — commercial AI