Keyword vs button vs AI chatbots on WhatsApp: which one should you use?

The short answer

Keyword vs button vs AI chatbot is not really a choice between three products. They are three tools for three different moments in a WhatsApp conversation:

If you are starting out, build a button bot first, start it with keywords, and add AI only where customers keep typing questions your buttons do not cover.

Keyword chatbots: triggers that decide which flow starts

A keyword bot replies when a message matches a word or pattern. On the WhatsApp Business API, where you can automate anything, keywords are best used as triggers: "FEES" starts the fee flow, "ORDER" starts order tracking, the prefilled text in your QR code starts the admission bot.

In Broadchat's chatbot builder, the Match Keyword trigger compares the incoming message with conditions you build from operators such as is, Contains, Start with or Match pattern (a regular expression), joined with AND or OR. Matching ignores capital letters and surrounding spaces. Hot Keywords go one step further: they restart their bot even while the contact is in the middle of another flow, which is what you want for words like menu or agent.

Where keywords break:

• Synonyms and spelling. "fee", "fees", "fess", "kitni fees hai" and "₹?" are five ways of asking one thing. Contains fee catches four of them; the fifth needs AI or a menu.
• Over-matching. Contains hi also matches "this" and "which". Use is for short words.
• Hidden commands. Customers do not know your keywords unless you print them on the QR code, the ad or the menu.

Use keywords for entry points and a few memorable commands. Do not build a whole conversation on them.

Button and list chatbots: predictable and measurable

Button bots ask a question and offer the answers. WhatsApp supports up to three reply buttons per message (titles up to 20 characters) and list messages with up to ten rows in total (row titles up to 24 characters, with an optional description). In Broadchat they are an Ask Question step of type Interactive, with Send Choices as Button or List, and every button or row gets its own output on the canvas.

That last point is why button bots are the workhorse of WhatsApp automation:

• Branching is visual. Each option leads somewhere you chose. There is no ambiguity about what the bot will do.
• Data is clean. A budget saved from a list row is always one of your five ranges, which makes segments and tags reliable.
• It works across languages. Customers tap rather than type, so a Hindi-speaking parent and an English-speaking student give the same answer.
• Reporting is easy. You can see which option people choose and where they stop.

Where buttons break: the customer's need is not on the menu, or they ignore the buttons and type a sentence. Always add a last option such as "Something else" or "Talk to a person", and handle typed replies with the step's No Match output, a retry message, or an AI step.

AI chatbots: free text, at a price

An AI step sends the customer's message and your instructions to a large language model and returns its answer. It understands "weekend batch hai kya Vaishali Nagar mein?" without any keyword, and it can detect intent ("this person wants a refund") or pull a name and date out of a sentence. Broadchat's builder has Trained AI (answers with an AI agent you configure), AI Intent Detector (routes by intents you describe in plain words, with a Fallback branch) and AI Data Collector (asks follow-up questions until it has every field you need).

What you pay for that flexibility:

• Accuracy depends on you. The model knows only the facts you wrote in its instructions. Out-of-date fees produce wrong answers delivered politely.
• Answers vary. Two customers asking the same thing may get differently worded answers. For legal, pricing or medical wording, use fixed text.
• More messages. Free-text chats run longer than taps, which uses more of the 1,000 free service messages each number gets per month; after that Meta charges per service message (₹0.115 per message to India, Meta rate card effective 1 Oct 2026, excl. GST).
• Policy. AI that serves your own customers is fine; a general-purpose assistant as the main purpose of the number is not. The AI chatbot guide covers Meta's rules and the hand-off design.

Decision table by use case

The hybrid most businesses end up with

This gives you the predictability of buttons for the questions you can foresee and the flexibility of AI for the rest, while every sensitive topic reaches a person. To build the button part step by step, follow the no-code chatbot guide; for ready-made flow outlines see WhatsApp chatbot templates by industry.

How to tell if you chose right

• Drop-off per step. If many people stop at a typed question, turn it into buttons.
• No Match rate. Frequent No Match on a button question means a missing option; add it.
• Repeated free-text questions. If AI answers the same question 50 times a week, make it a button with a fixed answer.
• Hand-off share and blocks. A rising share of "talk to a person" or users blocking the number means the bot is in the way; shorten it.

Whichever mix you choose, the platform rules are the same: bots reply freely only inside the 24-hour customer service window, and anything you send later must be an approved template. See what the WhatsApp Business API is for the basics and the chatbot builder for what Broadchat includes. Agencies can package these flows for clients under their own brand with the white-label platform.