AI Provider & Models — SEO Copilot AI Docs
AI

AI Provider & Models

How to configure your AI provider, which models to use, what data is sent during an analysis, and how to use a compatible endpoint instead of OpenAI.

How the AI provider works

SEO Copilot AI does not include a built-in AI model. Instead, it connects to your own AI provider account using an API key you supply. All AI calls are made from your WordPress server directly to the provider’s API — your content is never routed through NordicDataTools.

This approach keeps costs transparent, gives you full control over which model and provider you use, and means your API usage is visible in your own provider account at all times.

The two provider modes

OpenAI

Standard OpenAI

Connects to api.openai.com/v1/chat/completions. Use this for any OpenAI model — GPT-4o, GPT-4o mini, GPT-4 Turbo, and future OpenAI releases. Your key must be from platform.openai.com.

Compatible endpoint

Any compatible API

Connects to any URL you specify that implements the OpenAI chat completions API format. Works with Azure OpenAI, Ollama, LM Studio, Mistral, Groq, and other compatible services. You provide both the endpoint URL and the API key.

Choosing a model

The model field accepts any model name available on your provider account. The default is gpt-4o-mini — a strong starting point for most sites.

Model Speed Output quality Cost per analysis Best for
gpt-4o-mini Recommended Fast Very good ~$0.01–0.02 Most sites — strong output at low cost
gpt-4o Moderate Excellent ~$0.05–0.15 Pages where richer, more nuanced suggestions matter
gpt-4-turbo Moderate Excellent ~$0.05–0.15 Alternative to gpt-4o with similar capability
Local model (via Ollama) Varies Varies by model No API cost Development, testing, or privacy-sensitive environments

Start with gpt-4o-mini. It handles the structured JSON output format the plugin requires reliably, produces useful recommendations for most pages, and costs significantly less than larger models. Only switch to a larger model if you find the output quality insufficient for your specific use case.

What gets sent to the AI

Every analysis sends a structured prompt to the AI provider. Understanding what is included helps you make informed decisions about privacy and data handling.

Sent to your AI provider in each analysis request

✓
Page title, meta description, and heading structure (H1, H2, H3)
✓
A plain-text content preview (approximately 220 words — not the full page content)
✓
Word count, internal link count, image count, and image alt-text coverage
✓
Your target keyword
✓
Existing FAQ questions and schema types (if present on the page)
✓
Competitor page data (title, headings, word count, FAQ/schema signals) — only when competitor URLs are added
✗
Full page HTML or complete page content — only a preview is sent, not the entire post
✗
Any data sent to NordicDataTools — all AI calls go directly from your server to your provider
✗
Personal user data, customer information, or order data from WooCommerce

Technical details

API format OpenAI chat completions — POST /v1/chat/completions
Default endpoint https://api.openai.com/v1/chat/completions
Default model gpt-4o-mini
Temperature 0.1 — kept low for consistent, deterministic output across repeated analyses of the same page
Request timeout 35 seconds — applies to all analysis requests
Authentication Bearer token — Authorization: Bearer {api_key}
Expected response format JSON object — the AI is instructed to return only valid JSON with no markdown formatting
Key storage WordPress wp_options table — local only, never transmitted to NordicDataTools

Using a compatible endpoint

Any service that implements the OpenAI chat completions API format can be used as a compatible endpoint. This includes:

  • Azure OpenAI — set the endpoint to your Azure deployment URL, e.g. https://your-resource.openai.azure.com/openai/deployments/gpt-4o/chat/completions?api-version=2024-02-01
  • Ollama — run a local model and point to http://localhost:11434/v1/chat/completions. Set the API key to any non-empty string (Ollama does not validate it)
  • Groq — use https://api.groq.com/openai/v1/chat/completions with your Groq API key and a supported model name like llama-3.1-70b-versatile
  • Mistral — use https://api.mistral.ai/v1/chat/completions with your Mistral API key and a model name like mistral-large-latest
  • LM Studio — run a local server and point to http://localhost:1234/v1/chat/completions

The plugin requires the AI to return valid JSON only — no markdown, no preamble. The system prompt instructs the model to return structured JSON. Some smaller or local models do not follow this instruction reliably, which can cause analysis failures or empty results. If you experience issues with a compatible endpoint, try a larger or more instruction-following model.

AI usage and operations

Each successful AI analysis request counts as one operation against your monthly quota. On the Free plan, you have 50 operations per month. On Pro, Business, and Agency, operations are unlimited.

  • One analysis = one operation, regardless of page length, keyword, or number of competitor URLs
  • If an analysis falls back to local (rule-based) mode because the AI call failed, no operation is counted
  • The Test connection button in Settings does not count as an operation
  • Viewing, re-loading, or re-reading saved analysis results never uses operations
  • Operations reset on the first day of each calendar month

Common questions

My API key is saved but Test connection returns an error

The most common causes are: the key has been revoked or expired in your OpenAI account, the model name entered does not exist on your account, or there is a temporary outage at the provider. Check your API key is active at platform.openai.com/api-keys, confirm the model name is correct, and try again.

The analysis returns empty or garbled results

This usually means the AI response could not be parsed as JSON. The most common causes are using a model that does not follow structured output instructions, a very short timeout (less than 35 seconds on a slow provider), or a temporary provider outage. Try switching to gpt-4o-mini if you are using an alternative model, or check your provider’s status page.

How much does each analysis cost in OpenAI credits?

With gpt-4o-mini, a typical analysis costs between $0.01 and $0.02 depending on page length and whether competitor URLs are included (which increases the prompt size). With gpt-4o, expect $0.05–$0.15 per analysis. Your exact costs are visible in your OpenAI usage dashboard.

Can I use this plugin without an OpenAI account?

Yes — use the Compatible endpoint option with any OpenAI-compatible service. Groq, Mistral, and Ollama (for local models) are all viable alternatives. Some have free tiers or lower costs than OpenAI. The plugin does not require OpenAI specifically — it requires any service that implements the same API format.

Is my page content stored by OpenAI?

By default, OpenAI may use API inputs for safety monitoring purposes, but does not use API data to train models. If data privacy is a concern, review OpenAI’s current API data usage policy at openai.com/policies/privacy-policy, or use a local model via Ollama or LM Studio where no data leaves your server.

Where to go next

  • Settings — where to configure provider, model, API key, and endpoint URL
  • License & Plans — understand AI operation limits per plan
  • Troubleshooting — common AI connection and analysis errors
← Google Search Console License & Plans →