The AI features use your own API key, added on the Integrations and AI tab. You pay the provider directly at cost, with no markup and no credits to buy in the plugin.
Getting a key:
- Claude, from Anthropic: create a key at console.anthropic.com. Keys start with sk-ant-.
- OpenAI, ChatGPT: create a key at platform.openai.com. Keys start with sk-.
Pick your provider, paste the key, choose a model and click Save settings. Then click Test key: a green message confirms the connection and names the model in use, and any provider error is shown in full so you know whether the key is wrong, out of credit or restricted.
Which model should I choose?
- Claude Haiku 4.5 or GPT-4o mini: fast and cheap, and genuinely good enough for drafting surveys and summarising results. Recommended, and the default.
- Claude Sonnet 4.6: a balanced step up.
- Claude Opus 4.8 or GPT-4o: highest quality, for the best analysis of a large, messy set of written answers.
What it costs:
These are small, occasional calls rather than a constant drain: one when you generate a survey, and one each time you click Analyze. On a cheap model that is fractions of a cent per survey draft and typically a cent or two per analysis, scaling with how much written text your respondents leave. Providers publish current pricing and let you set your own spending caps.
How your key is stored:
Obfuscated, never in plain text, and the field only ever redisplays a masked version so it cannot leak back out through the screen. Leave the masked value alone to keep your key, or clear the field and save to remove it. The key is used for nothing except the AI features you trigger.