Your AI agent’s first message: what to say on each page

A visitor opens your chat and reads: “Hello, how can I help?” It is polite, but leaves them to discover whether your AI agent can help with product selection, delivery or contacting a person.

Name one useful thing the agent can help with and suggest a simple next step. In Breezaro, welcome messages, clickable starter questions and page-based scenarios each enter the conversation at a different moment.

All wording below is illustrative copy to adapt and test on your website. These are not transcripts from customers or messages with measured commercial results.

Decide where the message belongs

The Welcome message is the greeting when someone opens chat. It should make sense across your website. For example: “Hello, I can help you choose from our range and find delivery information. What would you like to know?” Mention only subjects covered by the agent's sources.

Starter questions are clickable prompts shown while the conversation is empty. Visitors can send their first question without typing it. Write these from their perspective: “How do I choose the right size?” is a customer's question, while “We can help with sizing” is your offer.

A proactive scenario message appears according to your rules, such as time spent on a particular page. Its wording can be more specific because you have defined the situation. Breezaro also has a separate Welcome bubble, managed in Assistant. When checking the website, identify which setting produced a bubble; switching off scenarios does not change that separate welcome setting.

Write for the decision being made

Someone browsing a category may be narrowing down options. On a product page, they may be checking a detail. On a pricing page, they may need to understand the difference between plans. Your opening line can address that step.

Choose one page where visitors ask a recurring question. Read existing conversations to find a decision you can support with concrete information.

Avoid assuming how the visitor feels. Time spent reading pricing does not establish that they think it is expensive. “Does this seem too expensive?” introduces a judgment you do not need. Offering to explain the differences leaves the visitor free to tell you why they are there.

Offer different ways to begin

Under Starter questions, you can configure up to four prompts per language. They appear while the conversation is empty. An illustrative store might use:

• “Can you help me choose a product?”
• “How do I find the right size?”
• “Where do you deliver?”
• “How can I return an item?”

Each opens a different subject. Four variations of a delivery question would use all the space without covering more needs. You can attach a fixed answer to a question, which Breezaro returns without AI. This suits stable information you want to state exactly, but you are responsible for keeping that text current.

For product selection or a context-dependent answer, leave the optional answer blank so the agent responds normally using its available sources. A fixed “Everything is in stock” cannot substitute for product information. Prepare each language's set separately and test it on the website.

Choose between your wording and an AI-written opener

Custom text gives you control over the exact wording. You can target a pricing page or a specific category. The page does not need to be in the knowledge base just to display your custom message. The agent's subsequent answer still needs appropriate information.

An AI-written message can refer to the content of the page currently open. Breezaro needs to find that page among its processed sources or catalog products. Test the scenario on a specific URL: a broad rule covering your whole website does not supply content for every page.

Outreach based on previously viewed pages requires Visitor context to be enabled. Keep the offer useful: “Comparing a few options? Tell me what matters most to you.” The existing scenarios guide explains the available conditions in detail.

Start with one message and a measured frequency

For an initial test, choose Custom rules, the address of one category and Time on page of at least 20 seconds. This is an example to evaluate, not the right delay for every page. Time on page counts while the browser tab is visible.

Save the mode, enable the scenario and switch on Reach out to visitors. You could start with a maximum of one message per visit and disable another offer after dismissal. Add cart and checkout pages to the excluded addresses. Observe responses before expanding the rules.

A dry run checks scenario conditions. Actual display also depends on visit limits and feature availability for the account. Test in a new tab because reloading the same tab may preserve visit state. See the documentation for the controls.

Check the wording with a short trial

Before publishing a change, run these checks:

1. Open the page as a new visitor. Is the offer clear from the first sentence?
2. Accept that exact offer. If the agent offers help with dimensions, ask for a specific measurement and check its answer.
3. Click every starter question. Check both the submitted question and any fixed answer.
4. Dismiss the bubble and continue browsing. Check frequency limits and excluded pages.
5. Try another language. The message and questions should make sense together.

In Analytics, review proactive message displays, chat openings and dismissals. For starter questions, look at clicks. An opening shows interest in continuing; it does not by itself establish that a question was resolved or a purchase followed.

Change one thing at a time, such as the wording on one category. Record the period and timing, then read the resulting conversations. If visitors respond but the agent cannot meet the need it offered to address, adjust that offer or improve its sources. The measurement guide explains how to interpret the other figures.