A poorly configured AI widget is a conversion killer. Pop-ups that interrupt reading, assistants that answer confidently but wrongly, and voice inputs that fail silently on mobile all damage trust faster than having no widget at all. Getting it right requires deliberate choices about placement, behaviour, and scope — and Aurrus gives you fine-grained control over all three.

Placement: Where the Widget Lives on the Page

Bottom-right is the industry standard position for interactive page elements, and there is a good reason for that: it is the furthest corner from primary content, so it does not compete for attention during reading, but it is within easy reach of a right-handed mouse or a thumb on a phone. Aurrus defaults to bottom-right, but lets you move to bottom-left from the dashboard settings if your layout already places a cookie banner, a chat bubble, or a sticky CTA in that corner.

Avoid floating the widget in the middle of the page, near the top, or overlapping navigation. These placements routinely score poorly in usability tests because they interrupt the natural reading flow before visitors have formed a question.

Auto-Greet: When to Open Automatically and When Not To

The biggest mistake most teams make is enabling auto-greet on every page. A visitor who lands on your homepage from a search result is orienting themselves — they do not want a chat window demanding attention before they have read a single sentence. Reserve auto-greet for high-intent pages where visitors are already in decision mode:

  • Pricing pages — visitors comparing plans often have specific questions
  • Checkout pages — hesitation at checkout is frequently question-driven
  • Contact pages — someone visiting contact is already trying to reach you
  • Long-form product pages — complex products generate complex questions

In Aurrus, you configure auto-greet per-page using data-autogreet="true" on the widget script tag, or by specifying URL patterns in the dashboard. You can also set a delay (in seconds) before the greet fires, which prevents it from interrupting a visitor mid-scroll.

Writing a Greeting That Gets Responses

The default Aurrus greeting is "Need help?" — two words, a clear question mark, no pressure. This outperforms longer greetings in A/B tests consistently. Visitors who have a question respond; those who do not, ignore it. If you customise the greeting, keep it under ten words and phrase it as a question. Statements ("Our assistant is here to help!") convert worse than questions ("Have a question about pricing?").

You can also use dynamic greetings in Aurrus. By setting page-specific greeting text via the data-greeting attribute on the script tag, you can tailor the opening line to the page context without touching your dashboard settings for every change.

Scoping What the Assistant Can and Cannot Answer

Claude is capable of generating plausible-sounding answers about almost any topic. That is a feature for general use, but a risk for business widgets. Your Aurrus system prompt is the mechanism that keeps the assistant on-scope. Write explicit negative instructions — "Do not speculate about timelines you do not know", "Do not discuss competitor products", "If asked about account-specific billing, direct the visitor to billing.aurr.us" — alongside the positive context about your business.

Good scope definition dramatically reduces the chance of a visitor receiving a confidently wrong answer and then sharing a screenshot of it. Review the conversation logs in your Aurrus dashboard weekly during the first month. Edge cases you did not anticipate in the system prompt will surface quickly, and each update narrows the scope tighter.

Handling Fallbacks Gracefully

No assistant handles 100% of queries well. Aurrus lets you configure a fallback message for situations where the assistant cannot confidently respond. A good fallback does three things: acknowledges the limitation, does not apologise excessively, and provides a clear next step. Example: "That one is outside what I can answer here — reach us directly at [email protected] and a human will get back to you within one business day."

Avoid fallbacks that loop the visitor back into another question ("Can you rephrase that?"). If the assistant could not handle the first phrasing, it likely cannot handle the rephrasing either, and the repeated failure erodes trust.

Testing Before You Go Live

Before enabling the widget on live traffic, run through a structured test matrix:

  • Test the five most common support questions your team fields manually
  • Test at least one out-of-scope question to verify the fallback fires correctly
  • Test on iOS Safari with microphone permission granted and denied
  • Test on Chrome Android with microphone permission granted and denied
  • Test on a desktop browser with no microphone to confirm text input works

Each of these is a five-minute check from a real device. Skip any of them and you risk a live visitor hitting a failure mode you did not know about. The Aurrus dashboard shows conversation logs in real time, so you can monitor the first hours of live traffic and catch anything the test matrix missed.