Which AI Provider Should Your WordPress Chatbot Use?

Quick answer: for most small businesses, start with one of the big general providers, OpenAI or Google Gemini, on a small, low-cost model. Both write clear, accurate answers from your website content, and both are cheap at small-business volumes. Choose Groq if speed matters most, Mistral if you prefer a European provider, OpenRouter if you want one account for many models, and Ollama only if you need the model to run on your own server. Then set a second provider as a fallback, so your chatbot keeps answering if the first one has a bad day.

This guide is for WordPress site owners setting up an AI chatbot that uses their own API key. It explains what the provider actually does, how the main options differ, what it costs to run, and how to choose without overthinking it. Prices and policies change often, so treat the specifics as a starting point and check each provider’s current pages before you commit.

What the AI provider actually does

A WordPress chatbot like Outview has two parts. The plugin lives on your site: it reads your pages, stores them in a searchable form, finds the passages that answer each question, captures leads and keeps the conversation history. The AI provider supplies the language model that turns those passages into a natural answer.

That split matters for your choice. Because the answer is grounded in your own content, the provider doesn’t need to “know” your business. It needs to read the passages it’s given and write a clear, accurate reply without adding things that aren’t there. Most current models from the major providers do that well. The differences that matter are cost, speed, reliability, data terms and language quality.

If you haven’t set up a chatbot yet, our step-by-step guide on how to add an AI chatbot to WordPress covers where the API key goes and what else to configure.

The main options at a glance

Provider	Best for	Worth knowing
OpenAI	A safe default with strong all-round answers	Wide choice of model sizes; small models are inexpensive
Google Gemini	Low cost and a generous starting allowance	Check whether your tier’s data can be used to improve Google’s products
Groq	Very fast replies	Runs open models on its own fast inference hardware
Mistral	A European provider	Useful if your clients or policies prefer an EU company
Grok (xAI)	Businesses already using xAI	Fewer small-business integrations than the larger providers
OpenRouter	One account, many models	Lets you switch models without new accounts; adds a middle layer
Ollama	Running a model on your own hardware	Needs a capable server and technical upkeep

How to choose: five questions

1. How many conversations will you have?

Providers charge per token, roughly per chunk of text read and written. Each chatbot answer involves reading the visitor’s question plus a few passages from your site, then writing a reply. For a small business handling a few hundred conversations a month on a small model, the monthly bill is usually modest. A busy online store with thousands of conversations will notice the difference between models far more.

The practical move is to start on a small, inexpensive model, watch your usage for a month, and only move up if answers fall short. Most providers let you set spending limits in their dashboard, and it’s worth doing that on day one. Our article on chatbot cost: subscription vs lifetime licence puts licence and usage costs side by side over three years.

2. Does speed matter for your visitors?

Most visitors will happily wait a couple of seconds for a useful answer. If your chatbot sits on a busy shop and people are asking quick product questions, faster replies feel better. Groq is known for speed. Smaller models from any provider are also quicker than large ones.

3. What are the data terms?

To write each answer, the visitor’s question and the relevant passages from your site are sent to the provider. That’s true of every hosted provider. What differs is what they may do with it. Some paid API tiers state that your data isn’t used to train their models by default, while some free tiers allow use for product improvement. Read the current terms for the tier you’ll actually use, and don’t index anything you wouldn’t publish.

South African businesses should read this alongside our POPIA-minded look at AI chatbots. If no data may leave your infrastructure at all, Ollama on your own server is the only option on this list that keeps the model in-house, at the cost of running and maintaining that server yourself.

4. Which languages do your customers use?

The large general models handle English, Afrikaans and other widely used languages well. Quality for less widely represented languages varies by model and changes as models are updated. If you serve customers in isiZulu, isiXhosa or another South African language, test the same twenty questions on two providers before you decide. Ask a fluent speaker to judge the answers rather than relying on how they look.

5. What happens when the provider has an outage?

Every provider has occasional slowdowns or outages. Outview lets you set a second provider as a fallback, so if the first one fails, the chatbot switches automatically instead of going silent. We explain how that works in what happens if your AI provider goes down. A common pairing is OpenAI as the main provider with Gemini as the fallback, or the other way round.

A realistic setup for a small business

Imagine a physiotherapy practice in Rondebosch with a WordPress site of about thirty pages. It expects a few hundred chatbot conversations a month, mostly about hours, fees, medical aid and booking.

A sensible setup: a small OpenAI or Gemini model as the main provider, the other as a fallback, a monthly spending cap in both dashboards, and a check of the usage page after the first month. If answers are accurate and the bill is small, there’s nothing to change. If a few answers are vague, the fix is usually clearer content on the relevant page, not a bigger model.

When a bigger model is worth it

Upgrading to a larger, more expensive model can help when questions are complex and need reasoning across several pages, such as comparing three service packages against a visitor’s situation, or when your content is technical. It rarely helps when the real problem is missing or outdated content. Before paying more, look at a handful of weak answers and check what the chatbot was given to work with.

Managing keys across several sites

If you run chatbots on more than one site, or for clients, use a separate API key per site where the provider allows it. That way you can see each site’s usage, set separate limits and revoke one key without affecting the others. Agencies will find more on this in usage caps and API keys for client chatbot sites.

How to compare two providers in an afternoon

You don’t need benchmarks or a spreadsheet of model scores. A short, practical test on your own site tells you more.

1. Collect twenty real questions from your inbox, WhatsApp and phone notes, written the way customers write them.
2. Run them on provider A and save each answer.
3. Switch to provider B in the settings and run the same twenty.
4. Score each answer as correct, partly correct or wrong, and note which felt clearer.
5. Check the usage dashboards to see what the test cost on each.

If both score about the same, pick the cheaper or faster one and make the other your fallback. If both get the same questions wrong, the problem is almost certainly on your website, not in the model.

South Africa and the US: a few differences

• Billing currency. Most providers bill in US dollars, so South African businesses pay in rand at the going exchange rate and may see bank fees on foreign card transactions.
• Privacy law. POPIA applies in South Africa; in the US, state privacy laws and sector rules such as HIPAA may apply. Either way, keep sensitive personal information out of chat.
• Languages. US businesses often need English and Spanish; South African businesses may need Afrikaans or an African language. Test the languages you actually serve.

Common mistakes

• Starting on the biggest model. It costs more and is often slower, with little gain for typical business questions.
• No spending limit. Set a monthly cap in the provider’s dashboard before you go live.
• No fallback. One provider means one point of failure.
• Blaming the model for content gaps. Most wrong answers trace back to a page on your site.
• Skipping the data terms. Know whether your tier’s data can be used for training or product improvement.
• Testing only in English. If customers write in other languages, test those too.

Frequently asked questions

Which AI provider is best for a WordPress chatbot?

For most small businesses, OpenAI or Google Gemini on a small, low-cost model is a sensible start. Choose based on cost, speed, data terms and language quality, and add a second provider as a fallback.

Is AI usage included in the Outview licence?

No. You connect your own API key and the provider bills you directly for usage. The licence covers the plugin itself.

Can I switch providers later?

Yes. Your website content and settings stay in WordPress, so you can change the provider or model in the settings and test the new answers straight away.

Can I run the AI model on my own server?

Yes, with Ollama, if you have a capable server and the skills to maintain it. For most small businesses, a hosted provider is simpler and more reliable.

Key takeaways

• The provider writes the answer; your website content decides what it says.
• Start with a small model from OpenAI or Gemini, set a spending cap and review after a month.
• Pick Groq for speed, Mistral for an EU provider, OpenRouter for flexibility, Ollama for self-hosting.
• Read the data terms for the tier you actually use.
• Always set a fallback provider.

Bring your own provider

Outview AI Chatbot works with Gemini, OpenAI, Groq, Mistral, Grok, OpenRouter and Ollama, with automatic fallback between them. You buy the plugin once and pay your provider only for what you use, with a 14-day money-back guarantee before you go live. See features and pricing.