What the options actually are, how to install one in a few minutes, and the part most guides skip: where the answers come from.
"AI chat for WordPress" covers three different products. Picking the wrong category is the most common reason people end up disappointed with the result.
| Type | What it does | Best when |
|---|---|---|
| Live chat / support inbox | Routes visitor messages to a human, often with an AI first responder. The value is the inbox and the people behind it. | You have someone answering messages and need ticketing, office hours and routing. |
| General AI assistant | Puts a general-purpose model on your site. It will happily answer questions that have nothing to do with you. | You want a novelty assistant, or you are building something custom on top of it. |
| Site guide | Answers using your own published pages and links to them. Its job is navigation, not conversation. | You have a lot of content — reviews, guides, a shop — and visitors cannot find the right page. |
If your actual problem is "people land on my site and leave without finding the article that would have helped them", you want the third kind. Kivunly is that kind.
In WordPress, go to Plugins → Add New Plugin, search for the plugin by name, then click Install Now and Activate. Plugins in the official directory are reviewed by volunteers before publication, which is a meaningful safety filter — prefer them to zip files from elsewhere.
Kivunly's listing is at wordpress.org/plugins/kivunly.
There are two honest options.
You paste an endpoint URL and a site key into the plugin settings, and the service handles reading your content, calling the AI model, and paying that bill. It is the fastest path, and it means trusting a third party with the text your visitors type.
If you are technical, point the plugin at your own endpoint — a small server, a serverless function, or an automation tool such as n8n or Make. You keep control of the data and pay the AI provider directly. Kivunly's request format is deliberately tiny so this is realistic to build in an afternoon; it is documented on the docs page.
The plugin itself is normally free. The backend is what costs money, because every answer is a model call. Expect services to price per message or per month, and expect free tiers to be small. If you run your own backend, a small model answering a few hundred questions a month costs single-digit dollars.
| Symptom | Usually means |
|---|---|
| The bubble does not appear | The widget is not enabled, no endpoint is configured, or a caching plugin is serving an old page. Clear the cache and reload. |
| Answers are generic and never link to your pages | The backend is answering from the model's own knowledge instead of your content. |
| Links in answers lead to 404s | The backend is inventing URLs. Stop using it for recommendations until that is fixed. |
| The first reply is slow, later ones are fast | Normal. The backend is doing its first read of your site and caching it. |
| It forgets what was said earlier in a long chat | Every chat has a memory limit. Most widgets send only the last few messages to keep costs down. |