WhatsApp Chatbot Guide: How to Build One That Actually Helps

What should a WhatsApp chatbot actually do on day one — answer FAQs, qualify leads, track orders, or stay out of the way? Most guides skip that question and jump straight to build instructions, which is how teams end up with a bot that impresses the demo and annoys every customer after it. This guide covers the four bot archetypes worth building, the two honest build paths, the design rules that keep bots tolerable, and what each path really costs in time and money.

TL;DR

• Pick an archetype before a tool: FAQ bot, lead qualifier, order tracker or AI conversational bot — each has different build needs.
• No-code builders cover most service and lead-generation bots in days; the API route is for deep integrations and custom logic.
• Menus beat free-text for anything structured; error recovery and an escape hatch matter more than cleverness.
• Test with hostile inputs before launch, not happy-path demos.
• Budget for maintenance, not just build — bots decay as your offers, catalog and policies change.

Four archetypes, four different bots

"Chatbot" hides four very different products, and deciding which one you're building settles half your tooling and design decisions before you draw a single flow.

The FAQ bot

Answers predictable questions — hours, locations, policies, starting prices — from a fixed knowledge set, via menus or keyword rules. It's the cheapest to build and audit, and boring on purpose. Its success metric is the share of inbound questions resolved with zero human touches.

The lead qualifier

Engages new contacts within seconds, asks three to five structured questions (need, timeline, budget band), tags the contact, and hands warm leads to humans with full context. This is the archetype with the clearest revenue line — the full sequence around it is covered in our lead nurturing flows guide.

The order tracker

Connects to your commerce stack — Shopify, WooCommerce, a sheet, an ERP — and answers "where is my order?" with live data, plus shipping exceptions and returns intake. All the value sits in the integration; the conversation itself is trivial by design, and that's fine. The pattern generalises: any record your systems can fetch — booking status, application stage, delivery ETA — becomes a bot answer with zero conversation design.

The AI conversational bot

An assistant backed by large language models that handles open-ended phrasing, answers from your docs, drafts replies for approval and escalates gracefully. Powerful and risky in equal measure: quality depends almost entirely on the scope you give it and the guardrails around it. Treat it as a staffed role, not a set-and-forget switch.

Build path one: no-code builders

Visual flow builders — trigger, message, wait, branch, handoff — with template management and inboxes bolted on. You draw the conversation, connect your WhatsApp Business Account, and ship. Popular options include dedicated WhatsApp platforms and the automation modules inside CRMs; Whaterakt's flow builder sits in this camp, aimed squarely at the operator who will also own the inbox afterwards.

The strengths are real: days instead of weeks, non-technical owners, template approvals handled in-app, and easy iteration as you learn what customers actually ask. The limits: unusual data lookups or complex logic get clunky — though native integrations (Sheets, Shopify, HubSpot, Google Calendar) absorb most of it. For a full map of what's possible without code, read our no-code WhatsApp automation guide.

Build path two: the API route

The alternative is building on the raw Cloud API — webhooks in, JSON messages out, your own state machine, your own datastore, your own admin panel. The Cloud API documentation on developers.facebook.com is genuinely good, and a competent developer can have a working proof-of-concept bot in a week or two.

Choose this path when the bot is a feature of a larger product (inside your app's own login), when you need lookups no connector covers, or when compliance requires data you must host yourself. Costs shift from subscription to developer time — almost always the more expensive line item for an SMB. A middle path exists too: build the flows in a platform and push custom events through webhooks and the REST API, which Whaterakt exposes alongside its builder.

Design principles that separate good bots from annoying ones

Tone: pick one and keep it

Decide whether the bot speaks as "we" for the brand or as a named persona like "Kabir from support", then hold that line everywhere — switching mid-flow feels like two strangers sharing one phone. Short sentences, one idea per message, emojis only if your human agents already use them. The bot's tone should be a slightly tidier version of your team, not a different species.

Menus for structure, free-text for depth

Anything with a known shape — choosing a service, checking status, booking a slot — should be buttons, because buttons can't be mistyped. Reserve free-text and AI for genuinely open inputs like "the strap broke after two weeks". A bot that makes people type what it could have offered as a button has already lost the user.

Error recovery is the real design work

Happy paths are easy; every serious hour of design goes into "what if they type gibberish at step three?" The rules that hold up in production: acknowledge the miss, restate the options, and count failures. Two failures in a row should offer a human, never a third identical attempt. Nothing torches goodwill faster than a loop.

The escape hatch is non-negotiable

One visible route to a person, from every step, answered within a stated SLA. If your bot's architecture makes that hard, the architecture is wrong — the handoff patterns in our automate WhatsApp replies ladder give you the exact rules.

Respect the 24-hour window

Everything above is ultimately shaped by one platform rule: free-form messages only inside a customer-initiated 24-hour session, approved templates outside it. Design the bot so session-time replies carry the conversation while template-time messages carry the ceremony — confirmations, reminders, re-engagement. A bot that tries to stay chatty through templates alone will keep failing template review; one that knows which clock it's on feels seamless to the customer.

Cost and effort, compared honestly

Rough planning numbers for an SMB-scope bot (as of writing — always check current pricing):

Two footnotes. First, Meta conversation fees apply on every path — they're charged per conversation, not per tool, as our API pricing breakdown explains. Second, platform subscriptions range from a few hundred rupees a month at the low end to enterprise pricing at the top; a freelancer-built bot often feels cheaper until the second change request arrives.

A pre-launch testing checklist

Run this before customers meet the bot:

1. Walk every menu path to its end, including the boring exits: back, repeat, main menu.
2. Type gibberish at every step; confirm the recovery pattern and the two-failure handoff rule.
3. Verify the escape keyword and button reach a human who sees full conversation context.
4. Check template rendering on Android, iOS and WhatsApp Web — formatting and buttons differ.
5. Confirm 24-hour window behaviour: what the bot sends outside a live session is an approved template, not free text.
6. Run one hostile user past it: complaints, sarcasm, refunds, voice notes, mixed languages.
7. Validate analytics before launch day: resolution rate, handoff rate and the step where people drop off.

FAQ

How much does a WhatsApp chatbot cost?

A no-code bot typically costs a platform subscription plus Meta's per-conversation fees — from a few hundred rupees monthly for small Indian teams. A custom API build shifts the cost to developer time, which usually dominates. Get a range, not a quote, before committing.

Can I build a chatbot on the regular WhatsApp Business app?

No. The app supports away messages and canned quick replies, but not flows, menus or AI. Any real chatbot runs on the official WhatsApp Business API.

How long does an FAQ bot take with a no-code builder?

A focused FAQ bot — ten to twenty questions, a menu, a fallback and a handoff — is a couple of working days including template approvals. Budget more time for content than for configuration.

Can a chatbot actually sell?

It can move sales along: qualify, share catalogs and storefront links, answer pre-sale questions and book calls. Closing still benefits from a human, and bots that push too hard train contacts to ignore the number.

What if the AI gives a wrong answer?

Scope it to your documents, log every conversation, review transcripts weekly and tune the escalation triggers. AI nodes are best where a wrong answer is recoverable — not for pricing promises or medical specifics.

Do I need the green tick to run a bot?

No. The verified badge is a credibility signal for notable brands and is separate from API access. Bots run fine on ordinary verified business accounts.

The right chatbot is smaller than you think and maintained more than you planned — that's true on every path above. Pick the archetype that matches your bottleneck, pick the build path that matches who will own it, and test like a hostile customer. Whaterakt ships the no-code path end to end — flows, template approvals, AI nodes and a shared inbox — as you'll see on the features page.