Most of setting up IrisAI is waiting for your content to index. The configuration itself is short — and one step now takes a single click that used to involve signing up for an account and copying an API key between two browser tabs.
What you’ll do
Install, connect in one click, choose what to index, place the assistant, then check it with Search Eval. Most of the elapsed time is the index running in the background.
Step 1: Install and open the wizard
Install the plugin and activate it, then open IrisAI from the admin menu — you’ll be taken to the setup wizard the first time. It’s two steps: connect, then index.
Step 2: Connect
Click Start Free with IrisAI. Your account, API key and trial credits are created automatically — there’s no registration form and nothing to copy across. When it finishes, the wizard shows your site ID, plan and remaining credits.
The Connect step offers three routes. Option 1 is Start Free. Option 2, in the collapsed advanced panel, is for people who already hold an IrisAI proxy key. Option 3 — “Bring your own key (BYO)” — is the one to use if you already pay for ChatGPT, Claude or Gemini: click Use my own key, and requests go directly from your server to that provider, who bills you.
Don’t confuse Options 2 and 3. Option 2 wants an IrisAI key; a provider key pasted there will not work.
Step 3: Choose what to index
This is the step worth thinking about rather than clicking through. Posts and pages are selected by default, and the wizard shows how many items each content type holds.
Index what answers questions. Skip what doesn’t. The picker lists your public post types, so the choice is usually between posts, pages, and whatever custom types your site defines — a type full of thin, near-duplicate entries adds chunks that can only dilute your matches. If you run WooCommerce, product descriptions are usually worth including.
PDFs are handled separately, from the PDF picker under Knowledge Base, not from this grid. Add those once the basics are working.
Indexing runs in the background. You can leave the page — a few hundred items typically takes a few minutes.
Step 4: Put the assistant somewhere
You have two options and most sites use both.
- The floating widget appears on every page, bottom corner, available wherever the visitor happens to be. It’s on by default.
- The full-page chat is a destination — add the
[irisai_chat]shortcode to a page called “Ask us” or “AI Assistant” and link it from your navigation. Use a full-width template.
Step 5: Test it properly
Don’t test with questions you already know the answer to — you’ll unconsciously phrase them using your own page headings. Test the way a stranger would.
Better still, use the Search Eval screen. Save a list of real questions and run them against retrieval only — no answer is generated, so you skip the expensive half of a request. Each question still costs one embedding to match, and those are cached for an hour, so re-running the same set while you tune is free. For each question you’ll see the top matching passage, its similarity score, the keyword boost and whether the assistant would answer or decline.
Two settings worth revisiting later
The system prompt
The default is deliberately conservative — it tells the assistant not to state prices, dates, names or contact details that aren’t in your content, and to reply in the visitor’s language. Add your own rules on top: how to handle booking requests, when to point at a human, what tone to use.
Auto re-index
With this on, editing a post refreshes its chunks automatically, which keeps answers current between full indexes. It is not a replacement for one: the incremental path splits by length rather than by heading, so a page updated this way loses the section structure until the next full re-index.
Leave it on for day-to-day edits, and run a full re-index after publishing in bulk or restructuring a page.
After any significant content restructure, run a full re-index rather than relying on incremental updates.
Ready when you are — trial credits included.