How to choose a WhatsApp bot in 2026

Which concrete problem do you solve with a bot

Before comparing products, define the problem. A bot solves one or several of these four: 1) missed calls while you serve a client (interrupts work, frustrates the client), 2) out-of-hours messages that go unanswered and walk to your competitor, 3) repetitive admin tasks (replying with prices, hours, address), 4) appointment change management costing team time.

If your number-one problem is just "answering hours and prices", a simple menu bot is enough. If your number-one problem is "converting messages into real bookings in my schedule", you need a conversational AI bot connected to a schedule. The product choice flows from the problem choice, not the other way around.

How to test a bot in 30 minutes before paying

The protocol: 1) sign up for the free plan or trial, 2) configure three services with different durations (15, 30, 60 min) and two professionals, 3) give a relative the bot WhatsApp and have them write a real case ("hi I wanted to book for cut and beard on Saturday late morning"), 4) check that it understands intent, proposes a slot and confirms, 5) have your relative ask to move it and check the bot handles the change, 6) cancel and see if the slot is released immediately, 7) write something not related to appointments ("hey, do you have parking?") and see how it handles it.

If all 7 steps pass well, the bot is serious. If it fails at any of them, look for another candidate. This costs half an hour per product and discards the ones that only work in guided demos.

Real costs of WhatsApp Business API worth understanding

WhatsApp Business API is not free for Meta: it charges per conversation initiated by the business outside the 24-hour window after a client message. Prices vary by country and category (utility, marketing, authentication). In Spain, a utility conversation costs a few cents; marketing ones cost more. Products that advertise "WhatsApp free" usually absorb this cost up to some volume and then pass it to you.

Ask exactly: 1) how many initiated conversations (reminders, templates) are included in my plan, 2) what happens if I exceed that limit, 3) is the per-conversation cost beyond the limit clear or does it vary by category. If the salesperson does not give concrete numbers, bad sign: you probably end up with an unwanted monthly overage.

How to measure real AI quality

Do not trust perfect demos. AI quality is measured with real cases that break the script. Concrete tests: 1) write to the bot with serious typos ("i wnt apt cut sat aftrnoon"), 2) use Spanish colloquialisms ("for la peque", "for the older one", "Friday next week, not this one"), 3) change your mind midway ("oh no, better Thursday"), 4) speak in plural ("we wanted an appointment for both"), 5) ask something unrelated to the booking ("is there parking?").

A serious AI handles at least 4 of the 5 cases without getting lost. Marai is trained on real Spanish cases across the 14 industries it covers, which shows from the first atypical message.

Meta rules that matter

Three rules your bot must respect without you having to think about them: 1) the 24-hour window (after the client writes you, you have 24 hours to reply with any message; outside that window you can only initiate with pre-approved templates), 2) initiated templates need explicit client consent to receive them (GDPR compliance + Meta rules), 3) mass spam is forbidden and Meta can suspend your Business API account if it detects it.

A serious product manages these three rules internally: warns you if you are about to initiate outside the window, records consents, avoids spam-like patterns. If you have to learn the API yourself to avoid breaking rules, the product is incomplete. Marai manages all three internally as part of the standard flow.