What your customers experience

Your customer never meets OMAI. They open WhatsApp, write to the business number they already have for you, and get an answer. Everything in this article is about that experience — because it is the only part of the product your customers ever judge.

What the thread looks like

It is an ordinary WhatsApp conversation with your business. There is no second number, no chat widget, no branding from OMAI, and nothing in the thread that identifies which software wrote the reply.

• Replies arrive from the same business number the customer wrote to.
• The business name and picture the customer sees are the ones Meta has approved for that number. They are set in Meta’s tools, not in OMAI.
• Replies are short and conversational, because the assistant is instructed to write WhatsApp style rather than in paragraphs. A reply longer than 900 characters is never sent — it is discarded and the conversation is handed to your team instead.
• Emoji follow your Emoji usage setting. The shipped default is Light, which means at most one emoji when it fits.
• The assistant is instructed not to announce that it is not a person unless it is asked. If a customer asks directly, it is told to say so.
• Your team’s replies arrive from the same number too, so the customer sees one continuous conversation rather than a transfer.

How fast a reply arrives

Usually within seconds, at any hour — evenings, weekends and holidays included. The message is picked up by a background worker that retrieves your business information and calls the AI provider, so the exact time depends on that provider. There is no delivery guarantee and no uptime commitment.

Two timing behaviours are worth knowing. If a customer sends several messages in a row, an answer that was still being written when the next message arrived is dropped, and the newer message gets the reply — so the thread does not fill up with near-duplicate answers. And if more than 24 hours have passed since the customer’s last message, WhatsApp itself does not allow a free-form reply: the assistant stays silent, and your team can only re-open the conversation with an approved template.

After hours, weekends and holidays

Closing time does not stop the assistant. It keeps answering at 2am and on Saturday exactly as it does at 10am on Tuesday, and nothing is queued for the morning. That is the point of it — but it also means the answers it gives at night are the answers you wrote, with no one reviewing them.

What your opening hours do change is one instruction. When today is marked closed in Business hours — or a special closure covers right now — and you filled in the After-hours message field, the assistant is told that the business is closed, to say when you are back, and to keep collecting useful details. Your text is guidance the assistant rewrites in its own words; it is not sent to the customer word for word. A day with no hours saved at all is treated as open.

When a person takes over

Pressing Take over on a conversation assigns it to that team member and silences the assistant in that thread until someone presses Resume AI. The customer is told nothing: there is no “an agent has joined” notice, no change of number, and no visible break. The labels Customer, AI assistant and Agent exist only inside your inbox.

If an agent simply types a reply without pressing Take over, the assistant’s answer to that same message is dropped before it is sent, so the customer never receives two answers at once. The assistant may still answer the customer’s next message, though — pause it or take the conversation over if you want the thread to stay human. On a number that also runs in the WhatsApp app on your phone, replying from the phone has a similar effect: the assistant stands by in that conversation and resumes automatically after the quiet window, which is 6 hours by default.

None of this is visible on the customer’s side. From their phone the thread simply continues: same number, same chat, one message after another, whether the words were written by the assistant or typed by your team.

When the assistant hands the conversation over

The assistant hands over when a customer asks for a person, when the message sounds frustrated, or when it raises a sensitive topic — medical, legal, financial or emergency wording. It also hands over when its own draft reply fails validation, and when your plan’s included AI answers are used up. It does not hand over because it found nothing useful in your business information; in that case it asks a clarifying question instead.

Thanks! I'm connecting you with a team member who will get back to you shortly.

תודה! אני מעביר/ה אותך לנציג/ה מהצוות שיחזור/תחזור אליך בהקדם 🙏

That acknowledgement is sent only while the 24-hour reply window is open. After a hand-over the assistant stays quiet in that conversation, and the conversation is marked in your inbox as waiting for a person — that entry is the notification your team works from. A best-effort email also goes to the organization owner saying a conversation is waiting; it carries a link to the inbox and no conversation content. Nothing promises the customer a response time — how fast a person answers is up to you.

What a customer never sees

• The instructions the assistant works from, and the business text it retrieved to answer. It is instructed never to reveal either, and a reply that leaks them is discarded and handed to your team instead.
• Anything you type in the test panel on the Assistant Settings screen. Those messages are processed by the AI provider and count towards your usage, but they never reach a customer.
• Internal notes, tags, lead status, or the reason recorded for a hand-over.
• Any other customer’s conversation. Each conversation is scoped to one contact and one organization.

What stays your responsibility

OMAI answers the people who write to you. Deciding who you may message, and being able to show that they agreed to it, remains with your business — as does telling customers, where your local rules require it, that an automated assistant answers first. Customer consent covers this in more depth.