If you see answers not streaming word by word, the search still works. Something between your server and the visitor is holding the reply back, or ExactFind has switched to its fallback.
What's normal
- A short wait first. "Searching…" shows while ExactFind picks articles, which usually takes a few seconds. Nothing can stream until then.
- Some replies arrive at once. When your site doesn't cover a question, or the assistant asks for more detail, the whole reply appears in one go.
How the fallback works
ExactFind tries to stream first. If that fails, it asks again and shows the full answer when it's ready. Visitors get the same answer, just with a longer wait. Your AI provider may be asked twice in that case, so it can use more tokens.
Check what's happening
- Open your browser's developer tools and go to the Network tab.
- Ask a question.
- Find the request to
answer, orchatwith ExactFind Knowledge. - Check the response's
Content-Type.
| What you see | What it means |
|---|---|
text/event-stream, but the text appears all at once | Something is buffering the response |
Two requests, and the second is application/json | Streaming failed, and the fallback ran |
Common causes
- A proxy or CDN holding the response until it's complete.
- Server compression or output buffering.
- A security or optimisation plugin that changes REST API responses.
ExactFind already asks servers not to buffer or compress these replies, but not every host follows that. Exclude these paths from any caching, compression or optimisation rule:
/wp-json/exactfind/v1/answer
/wp-json/exactfind-knowledge/v1/chatIf you can't find the cause, ask your host whether they buffer server-sent events. For other caching and optimisation settings, see Use Caching and Optimisation Plugins With ExactFind.
What we tested
We tested streaming on shared hosting with PHP on LiteSpeed behind Cloudflare. It worked with no changes, including a stream that ran past the 30-second PHP time limit. Other hosts may differ.