All of the AI features need an API key from either Anthropic, who make Claude, or OpenAI. You supply your own key and pay them directly for what you use. Nothing is marked up and no AI usage is bundled into the plugin price.
Getting an Anthropic key:
- Go to the Anthropic console and create an account
- Add a small amount of billing credit
- Open the API keys section and create a new key
- Copy it, it starts with sk-ant-
Getting an OpenAI key:
- Go to the OpenAI platform and sign in
- Add billing credit
- Open the API keys page and create a new secret key
- Copy it, it starts with sk-
Adding it to the plugin:
Go to AI Mailer, then AI Studio. Paste your key into the matching field, choose a model, and click the test connection button. The test sends a tiny prompt and reports back within a second, telling you the response time and token counts. If the key is wrong you get a clear message saying so rather than a vague failure.
Your key is stored in your own WordPress database. AI requests go directly from your server to the provider, so your subscriber list and email content are never sent to us.
What it costs:
Typical usage is small. A heavy month, generating a hundred subject lines, thirty sequences and a couple of hundred segment translations, usually lands between five and fifteen US dollars. Light users pay pennies. The plugin records the cost of every AI call so you can see exactly where it went.
Choosing a model:
Use a higher quality model for copywriting where the output matters, and a faster cheaper model for high volume jobs. You can change model at any time.
Choosing your model:
Each provider has a model dropdown in AI Studio Configuration, refreshed with the newest models. On the Claude side that includes Claude Opus 5, Opus 4.8, Opus 4.7, Sonnet 5, Sonnet 4.6 and Haiku 4.5, with current pricing shown beside each. Opus models give the highest quality, Sonnet is the everyday workhorse, Haiku is fastest and cheapest for high volume.
Primary and fallback: pick which provider handles AI requests first. If it fails and your fallback provider has a key saved, the fallback takes over automatically. Error messages always name the provider that actually failed, so you are never debugging the wrong one.