Change How Your AI Assistant Replies by Asking

To change how your AI assistant replies, tell Maya what you want in plain words, such as "Stop offering the 10% discount" or "Don't answer on Instagram after 10 pm." She finds the setting behind it, shows you a card with the before and after, and by default changes nothing until you approve. Getting replies right is worth a few minutes of your time: in a Gartner survey of 3,566 customers published in September 2026, only 27% said they would try a chatbot again after a negative experience.

What can you change just by asking?

Almost everything about how your AI employees talk, and when. The replies section of the Maya page lists the areas. Here are requests that work, and what each one changes:

You say	What Maya changes
"Call yourself Lina."	The assistant's name, one setting used on every channel
"Sound warmer, and drop the emojis."	Its tone and behavior rules
"When someone just says hi, ask which treatment they're after."	The greeting: how it answers a plain "hello"
"Stop offering the 10% discount."	The rule that offers it, edited or removed
"If someone wants to negotiate the price, send them to me."	A handover rule, and who gets the alert
"Only reply in English and Arabic."	The reply languages
"Don't answer on Instagram after 10 pm."	AI working hours for that one channel
"If anyone asks about franchising, send this text and tell me."	An automatic reply for that topic
"Ignore personal messages sent to the clinic account."	A block rule for that kind of message
"Wait a bit longer, people send three messages in a row."	The message wait time, from 1 to 300 seconds
"Follow up with people who go quiet."	Follow-ups to quiet customers, on eligible plans

What does the approval card show?

Every request becomes a card. In the app it looks roughly like this (the rule is an example):

Update AI Behavior Rules Before: "Offer new customers 10% off their first visit." After: rule removed Conflict check: no clashes with your other rules [Approve] [Reject] [Edit]

Approve applies it. Reject drops it. Edit sends it back to the chat so you can adjust the wording with Maya. On WhatsApp, most of these changes arrive as "Needs your OK" with Approve, Not now and Details buttons, as shown in managing your business from WhatsApp. A few, like changing who gets handover alerts or rewriting a rule to settle a clash, open a review page instead.

Want fewer cards? "Ask me first" is the default (Approval mode in the app). "Show me the plan" (Plan mode) lists the steps first, and once you OK the plan she works through it. "Just do it" (Bypass mode) applies most reply changes on its own. A change that rewrites an older rule to settle a clash, or one that fails its test, still waits for you, and on WhatsApp every change waits for your tap. Once Maya has read anything on the web, every change waits for you.

How does Maya keep a new rule from contradicting an old one?

A common way assistants go wrong is boring. Someone adds "never give discounts" at the bottom while "offer 10% to new customers" still sits further up, and now the AI has two orders that disagree.

Maya doesn't pile new lines on top. She edits the exact rule your request touches, then runs a conflict check across all of your rules. If the change clashes with an existing rule, or repeats one, she has to fix both in the same change, usually by rewriting or removing the old line. If what you asked for is already how the assistant behaves, she says so and changes nothing. The card shows the result of that check.

More on this in how to write AI agent instructions for customer service.

Can you test a change before customers see it?

Yes, in two ways.

Verify changes. This switch is on by default. Maya test-runs each change on sample messages before it can go live, and flags any change that fails. It uses training credits, and you can turn it off.
Try it yourself. Say "Test it: a customer asks if we're open on Sunday," and Maya runs that message through your assistant, including "what if we changed this" versions. Nothing reaches a customer, and the test run itself isn't billed, though your chat with Maya uses credits as usual. Why did my AI say that? covers this in depth.

How should you set handover and languages?

Start with handover. In a Gartner survey published in August 2026, 87% of customers said companies using generative AI for service must offer a way to reach a person. Tell Maya which conversations go to your team (price haggling, complaints, anything you'd rather handle yourself) and who gets the alert.

Languages work the other way round. By default, your AI employees answer in whatever language the customer writes in, from more than 100. That fits what buyers want: a 2020 CSA Research survey found in a survey of 8,709 consumers in 29 countries that 76% prefer to buy products with information in their own language. If you'd rather keep replies to languages your team can follow, ask Maya to turn on "Only reply in these languages." A customer writing in any other language then gets your main language. Our guide to multilingual customer service explains how this fits with your team's replies.

What happens to messages that arrive after hours?

AI hours are set per channel, so "Don't answer on Instagram after 10 pm" leaves WhatsApp untouched. Outside those hours, the AI stays quiet on that channel.

If you turn on catch-up replies (off by default), the AI answers those messages once its hours start again, as long as each message is no older than a limit you pick, from 1 to 24 hours. Catch-up works on WhatsApp, Instagram, Facebook, Telegram and TikTok. It skips website chat, because by then the visitor has left the page.

Can you undo a change?

Yes. Maya keeps a version history of your assistant's rules and instructions: what changed, when, and whether it came from the settings page or from her. Ask "What did we change last week?" or "Go back to how it was on Monday," and she prepares the rollback as a card like any other change. Hours, languages, wait time, catch-up, automatic replies and block rules aren't in that history, so to undo one of those, ask Maya to change it back.

What won't Maya do here?

Answer your customers. She changes how your AI employees reply. She doesn't send messages in a conversation or rewrite one already sent.
Let anyone but the owner edit the language list. The allowed languages are shared by every business the owner runs, so only the owner can change which ones are on it.
Turn on follow-ups on every plan. Suggested and automatic follow-ups need an eligible plan.

Have a reply you'd like to change? Text Maya on WhatsApp and tell her in your own words.

FAQ

Do I need to know how to write AI prompts?

No. Describe the behavior you want, the way you'd brief a new employee. Maya picks the rule, writes the wording, and shows it to you before anything changes. If the wording isn't right, tap Edit and tell her.

Who on my team can change how the assistant replies?

The owner, admins, and team members you've given access to Maya. Every change still arrives as a card unless the person asking has switched their own Maya to a mode with fewer approvals. The language list stays owner-only.

How do I undo a change I approved?

For your assistant's rules and instructions, ask Maya to show the version history, then ask her to roll back to an earlier version. The rollback comes to you as a card first. Hours, languages, wait time, catch-up, automatic replies and block rules have no version history, so ask Maya to change those back.

Does testing a change cost credits?

The automatic "Verify changes" test uses training credits. When you ask Maya to run test messages yourself, including "what if" versions, the test run itself isn't billed. Your chat with Maya uses credits as usual.

Sources: Gartner press releases of September 2, 2026 and August 4, 2026, survey of 3,566 B2B and B2C customers, February and March 2026; CSA Research, "Can't Read, Won't Buy: B2C", July 2020, 8,709 consumers in 29 countries. Maya's behavior checked against the product, November 2026.

Change How Your AI Assistant Replies, Just by Asking