WhatsApp AI assistants compared, including ours

Prices checked 31 October 2026

There are three genuinely different products being sold under one name here, and most comparisons mix them up. Working out which of the three you want narrows fourteen options to about four, and that is a more useful exercise than any feature list.

The three products, and which one you want

Broadcast and campaign tools. Their job is sending: bulk template messages, abandoned-cart sequences, click-to-WhatsApp ad funnels. AiSensy, Interakt and DoubleTick are built around this. If what you want is to message a list of customers, this is your category, and an assistant that only replies will disappoint you.

Shared inboxes with flow builders. Their job is routing: several agents, one number, with a visual flowchart to handle the opening questions. Wati, respond.io, Gallabox and ManyChat are here. The published implementation timelines are honest about the cost — a simple FAQ flow in hours, appointments and lead qualification in one to two weeks, anything with real integrations four to eight. The recurring expert criticism of the whole category is that businesses build elaborate flows trying to cover every case and end up with something that breaks the moment a customer phrases a question unexpectedly.

Assistants that answer from your own content. Their job is answering: you give them your website, menu or price list, and they reply in their own words, with no flowchart. ChatPal, Chatbase, Markleyo and Jabee are here. This category barely existed eighteen months ago and is the one growing.

Most businesses asking for "a WhatsApp chatbot" want the third thing and get sold the second.

The table

Entry price is the cheapest plan that includes WhatsApp and some form of AI reply, billed monthly where monthly is offered. Every one of these is before Meta's own per-message delivery fee, which is a separate bill in every case and is explained in what Meta charges.

"Partly" in the fourth column means the platform has a knowledge base but gates it, limits it on the entry tier, or treats it as an add-on rather than as the way the product works.

The meter matters more than the price

Two platforms at the same monthly figure can differ by a factor of five on the same business, because they count different things.

Active contacts is the model respond.io and ManyChat use, and it is the most divisive thing about ManyChat according to its own users. You are billed for how many people were in a conversation, whether or not anything was resolved. A quiet month with a large list costs full price.

Agent seats is Wati, Gallabox and DoubleTick. Wati charges $39 for each extra seat. Seats cost a vendor nothing marginal, which is why this one draws the complaints it does.

Prepaid credits is AiSensy, Interakt and Chatbase. Chatbase's expire. This is the model behind most of the refund complaints in the category, because money paid in advance for something unused is money there is an argument about.

Answered messages is Jabee. One meter, and it only moves when a customer got a reply. A quiet month is a cheap month. We are not claiming this is cleverer than the alternatives — it is simply the only one we could explain in a sentence.

The hidden line: whether Meta's fee is marked up

Since 1 October 2026 Meta charges businesses for service messages — replies inside the 24-hour window a customer opened — after the first 1,000 per phone number per month. That is the only kind of message an assistant like this sends, so it is a real line on your bill, and it is not the same number for everybody.

Wati publishes a per-country rate card at /pricing-by-country/, a page absent from its English sitemap, ending with the line that a 15% commission will be added, with the figure itself behind a login. Gupshup passes Meta's fees through at cost. Markleyo advertises no markup. Jabee cannot add anything: we connect you directly to Meta's WhatsApp Cloud API as a Meta Tech Provider with no solution provider or reseller credit line in between, so Meta invoices your own business account at its published rates and we never see that money.

That is a structural difference rather than a promise, and it is the one you can check yourself: look at who the WhatsApp invoice comes from.

Whether you can leave, which the reviews care about more than features

Reading the one and two-star reviews across this category, the complaints are overwhelmingly not "it cannot do X". They are about being stuck.

Cancellation is the single most common one. Wati has no self-service cancellation: a reviewer on 28 October 2026 writes that there is no way to cancel through the website and you have to go through a chat; one on 15 September that it is made "purposefully extremely hard" and charges continued; one in July that a failed downgrade defaulted to charging a full annual renewal. ManyChat carries the same complaint. Refusals of refunds for unused credit appear repeatedly against both Wati and AiSensy, whose Trustpilot profile is 46% one-star, concentrated there.

Support response times of one to three business days are the second theme, across Trustpilot, G2 and Capterra. One Wati reviewer asks the question the whole category deserves: "Why make them struggle with an AI instead of connecting them with an actual human?"

Jabee cancels in the billing portal, inside the dashboard, with nobody to email. It is monthly, so there is no annual term and no prepaid credit pool that can be stranded. We are a small company and a person answers support, usually within the hour during Gulf working hours. We have far fewer customers than any vendor above, which is the honest reason we can promise that and the honest reason you should discount it. All of it, in detail and with the competitors' own published behaviour beside it, is on how to leave, and what we charge while you stay.

The failure nobody's feature list addresses

The expensive way this goes wrong is not a flow that breaks. It is an assistant that answers confidently and incorrectly in front of a customer.

A dental clinic put an AI assistant on WhatsApp so patients could ask about treatments and prices; it guessed, confidently, and was wrong. A travel site's bot quoted fares below the real ones and produced failed bookings. An e-commerce bot reapplied a promotional code until prices went negative, costing over $150,000 and 2,400 fraudulent orders before anybody noticed. The general version, as a practitioner put it: a general-purpose chatbot is fluent, confident, and capable of inventing a price, delivery date or refund policy the business never offered.

This is the thing to ask every vendor on this page about, and the answers are thin. What we built for it:

• Before it goes live, ten questions every business gets asked are run against your content, and the gaps are named on your setup page rather than discovered by a customer.
• Every answer records what it retrieved, so "how often does it not know?" is a question with a number behind it.
• When it could not find something, the customer's actual question is grouped and shown to you, and you answer it once — by replying teach from your phone, if that is where you are.
• When it is not sure, it says so and hands the conversation to a person, rather than producing its best guess.

We are not publishing a resolution-rate percentage, and the reason is worth saying. Intercom's Fin advertises 76% and Tidio disputes it at 62%; neither publishes a method, and with five businesses live ours would be noise. When there is a real number we will publish it with the query that produced it.

Where Jabee is the wrong choice

Five cases, and in all five one of the products above is a better buy.

You want to send, not answer. Jabee replies to customers who write to you first and has no broadcasts, no campaigns and no template sending. If you want abandoned-cart recovery or a promotional blast, AiSensy and Interakt are built for it and we are not.

You want a flow builder. There is not one, deliberately. If your team has designed a decision tree it wants implemented exactly, Wati or respond.io will do that and we will not.

You need voice notes understood. Jabee cannot read audio or video. It keeps the file and hands it to a person. We would rather say that than ship a transcription feature we have not tested.

You need ticketing and SLA reporting. There is a shared inbox, handover, team routing by the customer's language and a record of who answered. There is no ticket queue and no SLA dashboard. respond.io and Intercom are the grown-up tools here.

$49 is more than the problem is worth. If you get a handful of WhatsApp questions a week, ChatPal's free tier is a genuinely sensible place to start, and we would rather tell you that than take $49 from somebody who will cancel in six weeks.

Where Jabee is the right choice

Your customers write in Arabic. This is the one we would argue hardest. Most platforms' Arabic is machine-translated interface strings; respond.io has over a thousand Arabic pages with no Gulf currency and no local example anywhere in them. Jabee answers in the customer's own dialect and every Arabic line we publish is written and checked by a native speaker, because the founder is one. Arabic demand is also the only part of this market growing — searches for واتساب بزنس API are up 200% year on year while the English equivalents fell by half.

You have a sales floor, not a support desk. Several people answering, each selling, each speaking different languages: a customer's question routes to a team by what it is about, then narrows to who speaks their language; the first person to reply claims the thread and everybody else is told who took it; the owner is the escalation if nobody does. ChatPal says plainly it is not a support desk and Markleyo says it has no queues. This is the part of Jabee that is genuinely deeper than anything cheaper.

Your prices are on paper, not on a website. Photograph the board behind the counter or upload the PDF rate card, and the model reads it — including the badly scanned ones a PDF library gives up on, which are exactly the ones real businesses have.

You have your own systems. Give Jabee a URL, a key and a sentence describing what it does, and the assistant calls it — order status, stock, a booking system. No flowchart, no middleware.

You care who gets your WhatsApp number near Meta. There is a whole genre of article about cheap providers getting business numbers banned, and the advice is always to check your provider uses the official API. We connect directly to Meta's Cloud API, never an unofficial library, and we have written up what happened when one of our own numbers was restricted, which is not a thing a vendor writes unless it is true.

How this page was made, and what is wrong with it

Prices came from each vendor's own pricing page or their own published comparison, read on 31 October 2026. Complaint quotes came from dated, named reviews on Trustpilot, Capterra and G2. The AI failure cases are published incidents. Where a figure is a vendor's own claim rather than something we verified, it says so.

Three honest weaknesses in it. We have not run a paid account on most of these platforms, so the table is their published terms and not our experience of them. Geo-dependent pricing means your Wati quote may differ from both figures shown. And we sell one of the fourteen, which no disclosure fully fixes — the useful test is whether the where we lose section names things that would actually cost us a sale, and it does.