In short
- WhatsApp has billed per message, not per conversation, since 1 July 2025
- Replies inside an open service window are free; marketing templates never are
- Instagram gives you 24 hours to reply, and 7 days if a human does it
- The answer library is the project — the integration is the easy part
There is a particular frustration in researching AI chatbot development in India: most of what you will read is either a vendor's rate card or an article describing rules that have changed. Meta moved WhatsApp from conversation-based to per-message billing on 1 July 2025, and a surprising number of guides still explain the old model in detail.
So this piece starts from the platform documentation rather than from anybody's pricing page. The rules are not trivia — they decide what your bot is allowed to say, when, and at what cost, which means they decide the design.
What does AI chatbot development in India actually involve?
Four things, in descending order of difficulty: writing the answers your customers actually need, deciding what happens when there is no answer, connecting to WhatsApp, Instagram or your website within each platform's messaging rules, and reading the transcripts afterwards to fix what went wrong. The integration is the easy part.
The WhatsApp rules that shape the build
WhatsApp automation runs on the WhatsApp Business Platform, and Meta's own documentation sets out the terms. Since 1 July 2025 billing is per message rather than per conversation. Templates fall into three categories — marketing, utility and authentication — and the category decides whether you pay.
- Service messages are free. Non-template messages sent while the customer service window is open cost nothing. This is the window a customer opens by messaging you.
- Utility templates are free inside that window. Order updates and similar, sent while the window is open, are not charged.
- Marketing templates are always charged. There is no free tier for them, in any window.
- Ads can open a free window. If someone reaches you from a Click to WhatsApp ad or a Facebook Page button on the Android or iOS app and you reply within 24 hours, that reply is free and opens a free entry point window in which messages are free for 72 hours. Desktop and web users do not trigger it.
- Prices move on a calendar. Meta updates rates only on 1 January, 1 April, 1 July or 1 October, so a quoted rate has a known shelf life.
The practical reading: a chatbot built to answer people who message you first is cheap to run, because almost everything it sends sits inside a free service window. A chatbot built to push marketing messages out is a different product with a different budget, and should be costed as advertising rather than as support.
The Instagram rules are stricter
Instagram and Messenger automation run on Meta's Messenger Platform rules, and the governing constraint is time rather than money. A business has up to 24 hours to respond after a customer takes a qualifying action such as sending a message. Inside that window you may reply freely, promotional content included.
After 24 hours the door mostly closes. Message tags allow certain important one-to-one updates to go out later, but only for Meta's approved use cases, and misusing them can restrict your ability to send messages at all. Notably, some tags exist on Messenger but not on the Instagram messaging API, so a design that works for a Facebook page may not transfer.
One tag is worth knowing about: the human agent tag lets a person — not an automation — reply by hand for up to seven days after the customer's message. That is the formal mechanism behind what we would anyway recommend: a bot that hands over cleanly to a human, who then has a week of room to resolve things properly.
What the rules mean for your design
| Platform | Reply window | Cost shape |
|---|---|---|
| Service window opened by the customer; 72-hour free window from ad click-throughs | Free inside the service window; marketing templates always charged | |
| 24 hours, extendable to 7 days when a human replies | No per-message charge; the limit is time, not cost | |
| Your website | No platform limit | Your hosting and model usage only |
Two design conclusions fall out of this. First, build for inbound. A bot that answers questions people have already asked is both cheaper and better received than one that starts conversations. Second, design the handover before the automation, because on Instagram the handover is also what buys you the next seven days.
The content is the project
Everything above is a constraint, not a plan. The plan is the answers. A chatbot is only as good as the material it draws on, and the common failure is pointing a capable model at a thin FAQ page and expecting it to improvise the rest, which it will, confidently and wrongly.
The work that decides whether the thing helps is unglamorous: collecting the questions people genuinely ask, writing correct answers in your own voice, deciding the edge cases, and establishing what happens when there is no source. That last one matters most. The correct behaviour in the absence of an answer is to say so and fetch a person — never to fill the gap.
This is why we treat it as a writing job attached to a build, which is the approach set out on our AI chatbots page.
What to prepare before you start
- Pull the last three months of enquiries from WhatsApp, Instagram DMs and email, and count what actually repeats. This list is the specification.
- Write the correct answer to each, properly, as you would want a customer to read it.
- Decide the boundaries: prices, refunds, medical or financial specifics, anything legally sensitive. Mark what the bot must never answer alone.
- Decide where a handover lands, who sees it, and what happens outside working hours.
- Check your accounts: WhatsApp automation needs the Business Platform and Meta approval; Instagram needs a professional account.
- Agree how you will review transcripts after launch, and who owns fixing the answers.
How these projects fail
Warning signs in a chatbot proposal
- The content work is a line item, not the bulk of the project
- Nobody asks to see your real enquiry history
- Handover to a human is described as optional or 'phase two'
- The demo answers beautifully on questions chosen by the vendor
- Pricing is quoted without reference to Meta's own messaging charges
- Nobody mentions what happens when the bot does not know
The last two are the common ones in this market. Meta's charges are billed by Meta, so any quote that does not separate build cost from running cost is hiding something you will discover later.
Website chat is a different problem
On your own site there is no messaging window, no template approval and no per-message charge from a platform. What you gain in freedom you lose in context: a stranger on a web page has given you nothing, where a WhatsApp user arrives with a phone number and a history.
So the jobs differ. Website chat earns its place by answering pre-sale questions quickly and capturing the enquiry when it cannot — much the same job as a well-structured website, done conversationally. Messaging bots earn theirs by handling the repeat questions that otherwise consume somebody's afternoon.
Judging a developer
- Ask what they would do if the bot has no answer. If the reply is about the model rather than about the handover, keep looking.
- Ask how they will get your answers out of your head and onto the page, and who writes them.
- Ask which Meta charges you will pay directly and which are in their quote.
- Ask what the Instagram design does on day two, after the 24-hour window has closed.
- Ask to see a transcript review from a live deployment, with the failures in it.
If you are weighing up the build more broadly, the same logic we apply to app costs applies here: the rate is rarely the expensive mistake. Paying for the wrong thing at a good rate is.