12 WhatsApp AI agent use cases that work, with examples

The AI agents that work best on WhatsApp are the ones people start themselves and return to on their own: a companion, a language tutor, a habit coach, a journal, a wellbeing check-in, a personal assistant, a travel concierge, a meal planner, a shopping helper, a study buddy, a community helper and an expert persona. None needs the agent to message first.

What makes an AI agent fit WhatsApp?

An AI agent fits WhatsApp when four things are true. The user can begin by texting. The value arrives in a few short messages. The user comes back because of a habit or a need, not because the agent nudged them. And everything works when the agent only ever replies.

That last test is not optional on a whasp number. A number cannot open a chat: a send to someone who never wrote returns 409 chat_not_open, because cold messages are what get personal numbers banned. It also has no groups, no outbound voice notes, no typing indicator and no replies to a specific message. Inbound it handles text, photos, files, locations and contacts, and a voice note arrives as an audio file.

Most of these agents also have to run on a personal number. Meta’s pricing page lists the markets where it allows AI providers on the official Business API because regulators required it. Elsewhere the January 2026 terms bar them. Every option for getting an agent on WhatsApp is mapped in its own article.

Which companions and coaches work well on WhatsApp?

Companions and coaches work well because the conversation is the product and the user decides when it happens. Five fit.

1. AI companion

An AI companion fits WhatsApp because daily small talk is what the app was built for. The user texts when they feel like it; the persona lives on the number’s name, about line and photo.

Maya: Rough day. My manager moved the deadline again.

Ava: Again? Is it the same project as last week?

Maya: Yeah. I don’t even know where to start.

Get right: memory and honesty. Remember what mattered, say it is an AI, and plan for users in distress. Launching a companion on WhatsApp covers the whole path, and why companions are moving to messaging has the numbers.

2. Language tutor

A language tutor fits WhatsApp because practice happens in short bursts, and the learner can send a sentence, a photo of a menu or a voice note. Voice notes arrive as audio files you can transcribe for feedback.

Leo: Ayer yo fui al mercado y compre manzanas rojos.

Tutor: Almost. “Manzanas” is feminine, so “rojas”. Also “compre” needs an accent: “compré”. Want to try the sentence again?

Get right: correct one or two things per message. The tutor cannot send a voice note, so pronunciation help has to be text, or an audio file that arrives as a file.

3. Fitness or habit coach

A habit coach fits WhatsApp because users log a workout or a meal right after it happens, in a chat they already have open.

Sam: Did 5k, 29 min. Knee felt tight at the end.

Coach: Nice, that is 40 seconds faster than last Sunday. Tight how: sharp, or just sore?

Get right: store workouts as structured data in your database, not only in the model’s context, so “faster than last Sunday” is a fact and not a guess. Make clear it is not medical advice.

4. Journaling partner

A journaling partner fits WhatsApp because typing to a chat is how many people already think out loud, and a voice note works for the days they do not feel like typing.

Dev: Wrote nothing for three days. Everything’s a blur.

Journal: Let’s do two minutes. What is one moment from today you would keep?

Get right: privacy wording. Messages reach your server, so do not promise end-to-end privacy. Say where entries are stored and how to delete them. whasp’s servers keep no messages, so the journal is yours to protect.

5. Wellbeing check-in

A wellbeing check-in fits WhatsApp because people open up in a chat more easily than in a form. It is also the use case with the most regulatory attention.

Nora: I’ve been feeling flat all week.

Check-in: I’m sorry, that sounds heavy. What has a normal day looked like lately?

Get right: do not present it as therapy or diagnosis, and have a protocol for self-harm that gives a crisis line for the user’s country. California’s SB 243 requires such protocols for companion chatbots, and the FTC asked seven companies in September 2025 how they test and monitor chatbots for harm. Rules on AI and mental health vary by country and state, so check yours. The companion safety rules list the main ones.

Which assistants and planners work well on WhatsApp?

Assistants and planners work well because the user asks a question at the moment they need the answer, and the chat is already in their pocket. Four fit.

6. Personal assistant and reminders

A personal assistant fits WhatsApp because texting it like a person is the whole product. Lists, notes, quick questions and photos of things to remember all fit.

Priya: Remind me to call the dentist tomorrow at 9.

Assistant: Done. I will text you tomorrow at 9:00 about the dentist. Which time zone are you in?

Get right: reminders are the one place the agent sends without a fresh inbound message. A whasp number can send to anyone who has written to it, so do this only for things the user asked for, at the time they set. Ask for the time zone once and store it.

7. Travel concierge

A travel concierge fits WhatsApp because people ask while they are on the move, and a shared location arrives with its latitude, longitude and name.

Tom: (shares a location) Any good place for dinner nearby? Not too pricey.

Concierge: Three options within ten minutes on foot. First: a small trattoria, mains around 12 to 16 euros. Want vegetarian only?

Get right: use live data from a places service, not the model’s memory, for hours and prices. Say when you are unsure.

8. Recipe and meal planner

A meal planner fits WhatsApp because the input is often a photo of a fridge or a shopping receipt. Media arrives as a signed link valid for 24 hours, so copy the file if you need to keep it.

Ana: (sends a photo of a fridge) What can I cook tonight? Two people, 30 minutes.

Planner: I can see eggs, spinach, feta and half a lemon. Baked eggs with spinach and feta, 20 minutes. Want the steps?

Get right: store allergies and dietary rules as hard constraints and check them on every plan, not as a line the model might forget.

9. Shopping helper

A shopping helper fits WhatsApp because “which one should I buy” is already a chat question, usually asked with two links and a doubt.

Jo: Torn between these two vacuums. (sends two links)

Helper: Both fit your budget. The first is lighter, the second has a bigger bin. You said the stairs are the problem, so I would pick the first. Want me to check the return policies?

Get right: fetch live prices and stock, never guess them, and disclose any affiliate link.

Which learning, community and expert agents work well on WhatsApp?

Learning, community and expert agents work well because members already hold the number as a contact and ask one question at a time. Three fit.

10. Study buddy

A study buddy fits WhatsApp because students already live there, and a photo of a notes page is a natural input.

Lina: Quiz me on the French Revolution, five questions.

Buddy: Question 1 of 5: what event on 14 July 1789 became a symbol of the revolution?

Get right: age and accuracy. WhatsApp’s Terms of Service set a minimum age of 13, or higher where local law requires, so some users will be minors. Ground questions in the student’s material, and track what they get wrong.

11. Community helper

A community helper fits WhatsApp when a club, course or creator puts one number on the welcome page and every member texts it privately. Groups are not supported, so this is a one-to-one helpdesk, not a group bot.

Kai: When does the next cohort start, and is there a refund policy?

Helper: The next cohort starts on 4 November. Refunds are available until the end of week one. Want the full policy?

Get right: answer only from your own documents and hand over to a human when unsure. If the agent mainly serves your own business’s customers, the official Business API is often the right home, and the options article compares them.

12. Expert persona

An expert persona fits WhatsApp when a coach, chef or trainer puts their know-how behind a number and followers text a question.

Chris: Is it too late to start strength training at 52?

Coach Mia (AI): Not at all. Most people start with two short sessions a week. Any injuries I should know about?

Get right: consent and scope. Use a real person’s voice and likeness only with their approval, label the agent as an AI, and have it decline questions outside the expert’s field.

What do all twelve have to get right?

All twelve rest on the same foundations.

• The user starts. Show a wa.me link in your app, site or email, with a prefilled message that carries a signup code.
• You keep the memory. whasp passes messages through and its servers store none, so history is yours. Conversation memory shows how.
• STOP works. STOP, UNSUBSCRIBE, OPTOUT, CANCEL, END, QUIT and any spelling of “opt out” opt the chat out; polite requests are yours to catch.
• Capacity is finite. A number holds about 100 active users, 40 new chats a day and 1,500 sent messages a day. Add numbers before 80%.
• What is left off. Price alerts, “good morning” broadcasts, re-engagement blasts and cold lead generation all need the agent to speak first. A number cannot, so they are not here.

Where should you start?

Start with the use case where you can finish this sentence: “The user texts first, and what they send is ___.” If you can, build that one. Write the persona, add a number, set a webhook, and use the quickstart to get your first reply working. Then read how to launch a companion on WhatsApp, which covers onboarding, memory, safety and scale in order, and apply it to your use case.