ResultsAiX Forms uses your own AI account. As each provider card says, your key is "Stored in your WordPress database. Never sent to us."
Step by step:
- Open the Integrations & AI tab and scroll to "Your AI providers". There are three cards: Google Gemini, Anthropic Claude and OpenAI. One key is enough.
- Open "How To Get Your Key In 60 Seconds" on the card you want. Google Gemini is marked "Free Key Available", and its walk through notes that no credit card is required for the free tier. Claude and OpenAI ask you to add a little credit to your account with them first.
- Paste the key into "API Key". Gemini keys start with AIza, Claude keys with sk-ant, OpenAI keys with sk-.
- Pick a "Model". The best everyday model for each provider is listed first and marked recommended.
- Click "Save AI settings", then click the card's Test connection button. The test checks the saved key, so save first. Success reads "Connected to" followed by the provider and model.
After saving, the key shows as dots with its last four characters. As the hint says: "The dots are a mask, not your key. Paste a new key to replace it, or clear the field and Save to remove it."
Which model to choose:
- Google Gemini: "Gemini Flash, always current", which has a free tier and follows Google's latest Flash model, so it keeps working when Google retires an older one.
- Anthropic Claude: Claude Sonnet 5.5.
- OpenAI: GPT-5 mini.
Each list shows the price per million tokens beside every model, with faster, cheaper and premium choices underneath. A model you have already chosen and saved is never changed for you. Newer Claude models that think before they answer are set up automatically, so there is nothing to switch on or off.
AI configuration:
- "Enable AI features" is on by default. Untick it to switch both AI features off.
- "Primary provider" is set to "Auto (first provider with a key)" unless you choose one.
- "Fallback provider (optional)" is "Used automatically if the primary fails (rate limit, outage, error)."
- "Monthly soft budget (USD)": enter an amount, or leave it at 0 for off. When the estimated spend of the last 30 days passes it, a warning appears on the AI usage panel, in the builder's AI box and on the AI Insights tab. AI keeps working, because "Hard limits are enforced by your provider."
AI usage:
The panel shows "Total jobs", "Last 30 days", "Spend (30d)" and "Errors (30d)", and a "Recent AI jobs" table with the type, status, model, cost, duration and time of the last ten calls. The costs are estimates from published prices. Your provider's own dashboard is the source of truth. A model with no published price yet shows "price unknown" instead of a made up figure, and those calls are left out of the 30 day spend, with a note saying how many.
When something goes wrong:
- Every failed job in the table has a "Why:" line under it with the reason, in the provider's own words.
- Each provider card shows its "Last error" and how long ago it happened, until a call to that provider works again.
- If the most recent call failed, a "Last AI error" box sits at the top of the usage panel. It clears on the next successful call.
- AI errors are also written to the error log on the Diagnostics tab. Your keys are never written to it.
What works without a key:
Everything except "Describe it, AI builds it" and the written analysis on AI Insights.