91% of India Chats With a Business Every Week. Zero Percent of That Conversation Trains an AI.

The answer engine nobody built on purpose

An answer engine, in the plainest sense, is a system that takes a specific question and returns a specific, trusted answer. India already runs one at enormous scale, and nobody designed it as a product. It emerged as behavior on WhatsApp, one conversation at a time, because that is where Indian consumers already are and where small businesses already show up to meet them.

Every day, a buyer messages a shop or a service provider with a version of the same question an AI engine is now also being asked somewhere else: whether an item is in stock, what it costs, whether it can be delivered by Friday, whether it comes in a given size, whether the shop is open right now. The business owner, or a family member, or an employee, answers directly, correctly, and in real time, drawing on stock counts, prices, and service details that live nowhere except in that owner's head and inventory. Functionally, the loop is identical to what ChatGPT, Perplexity, or Google's AI Overviews attempt to do without a human in the middle: take a specific query, retrieve or construct the right fact, deliver one trusted answer.

The difference is what happens to that answer afterward. When a generative engine answers a question, the underlying content it drew on, if it drew on anything real, sits in public, gets indexed, and can inform the next answer to the next person who asks something similar. When an Indian MSME owner answers the same question on WhatsApp, the answer is correct exactly once, for exactly one buyer, and then it is gone. It never compounds. It never trains anything. It never helps the next customer who asks Google or ChatGPT the same thing about the same business.

The numbers behind the claim

The starting figure comes directly from Meta. A Kantar study commissioned by Meta, conducted across 22 markets between April and September 2025 with a base of 11,056 online adults aged 18 to 64, found that 91% of online adults in India message a business at least once a week, more than any other channel and more than phone calls or email combined. Meta has cited that figure directly to justify rolling out AI-assisted business messaging tools for small Indian businesses on WhatsApp.

On the business side, the same pattern holds. Meta reported in December 2025 that more than 92% of businesses using its platforms in India are micro, small, and medium enterprises, not large brands. A separate, India-specific survey, the MSME Digital Index 2024 by PayNearby, covering more than 10,000 retail MSMEs including kirana stores, medical shops, mobile-recharge outlets, and travel agents, found WhatsApp and WhatsApp Business in use by 97% of respondents, the single most dominant tool in the entire survey, ahead of accounting software, point-of-sale systems, and CRM tools combined.

The base this sits on is not small. India's active internet user base reached 886 million in 2024 and was projected to cross 900 million in 2025, according to the IAMAI-Kantar Internet in India Report 2024. On the enterprise side, more than 7.83 crore, 78.3 million, enterprises had registered on the government's Udyam Registration Portal as of the end of February 2026, according to the Press Information Bureau. Put those together and the shape is unambiguous: a very large share of a very large population is, every week, asking a very large number of small businesses direct questions and getting direct answers, entirely inside a channel that was never built to surface any of it publicly.

What actually happens inside the thread

A WhatsApp Business conversation with an Indian MSME rarely stays a single exchange. It functions as a rolling combination of storefront, order book, appointment desk, and customer-support line, all inside one thread the owner already checks throughout the day. The WhatsApp Business app's own feature set, catalogs, quick replies, labels, and automated greeting and away messages, was built around exactly this pattern of repeated, similar questions answered the same way each time.

Meta has been actively formalizing this. In May 2026, it launched Business AI on WhatsApp for small businesses in India, a no-code assistant that automates FAQs, product recommendations, and appointment booking, and works across native Indian languages, with the owner still able to step in on complex queries. Meta's own case studies from the launch cite a plant-nursery business reporting an 80 to 90 percent conversion rate on AI-handled 24/7 support, and another small retailer reporting a 40 percent sales increase after AI began handling routine order queries. Those are Meta's selected examples, not independently verified outcomes, and they should be read as illustrations of the mechanism rather than as a general result.

What matters for this piece is the mechanism itself, not the marketing. Whether a human or Meta's own assistant is typing the reply, the underlying knowledge, current stock, real prices, actual service radius, honest turnaround time, is generated fresh inside the chat and stays inside it. It is real, current, first-party business information, produced constantly, at a volume no outside researcher or crawler will ever see.

Why the corpus stays sealed

There are two separate barriers here, and it is worth being precise about which is which, because one is a privacy protection and the other is a business decision.

The asymmetry this creates

Generative answer engines do not invent what they say about a business. Peer-reviewed research on what actually earns a citation inside a generated answer, the 2024 Generative Engine Optimization study from KDD, found that sources gained visibility when they carried cited statistics, direct quotations, and corroborating authority, signals that are, by definition, public, structured, and checkable by something other than the two people in the conversation.

A WhatsApp thread cannot supply any of that, not because the information inside it is wrong, but because it is structurally invisible to anything outside the conversation. The business answers a real question correctly, but the answer is uncorroborated by any second source, unstructured, and permanently private. The next customer who asks an AI engine the identical question, is this shop open on Sunday, do they deliver to this pin code, is this the right size, gets whatever the engine can find in public: a Google Business Profile that may be thin or outdated, a directory listing nobody has touched in a year, or nothing at all. The business has already generated the correct answer, sometimes hundreds of times over. None of that effort discounts the work an AI engine has to do to guess the same answer independently, and worse, from thinner material.

What can be measured, and what is still an open question

It would be a stronger claim, and a false one, to say anyone has directly measured what share of a typical Indian MSME's total customer Q&A volume happens exclusively on WhatsApp versus what is also present somewhere public, a website FAQ, a Google Business Profile Q&A section, structured data on a landing page. No public dataset currently makes that comparison, and this piece does not claim to have run one. What the figures above do establish, reliably, is the scale of the private channel (weekly usage above 90% on both the consumer and MSME side) against the near-total absence of India-specific research on how much of that same content ever gets mirrored anywhere machine-readable.

A credible way to close that gap would not require reading anyone's private messages. It would mean sampling the categories of questions MSME owners report answering most often on WhatsApp, a survey instrument, not a scrape, and then checking, business by business, whether the same answers exist anywhere public: on a Google Business Profile, a website FAQ page, structured LocalBusiness schema, or a directory listing. That comparison, common WhatsApp question versus public machine-readable answer, is the specific, measurable gap this piece is naming as unresolved, not asserting as already quantified.

What is already actionable, without waiting for that research, is the mirroring itself. A business owner who already knows the ten questions buyers ask most on WhatsApp has, without realizing it, a ready-made FAQ page, a set of Google Business Profile Q&A entries, and the raw material for structured data markup. None of that requires exposing a single private conversation. It requires treating the knowledge already proven correct by real customers as a public asset instead of a private one.

How to read this

The evidence here should not be read as an argument against WhatsApp. It is the opposite: WhatsApp is doing exactly what it was built to do, moving real transactions and real answers between real people, at a scale few channels anywhere in the world can match. The problem is not the channel. It is the assumption, often unstated, that being highly active on WhatsApp is the same thing as being visible to the systems that increasingly decide which businesses get named when someone asks an AI engine a question instead of typing it into a chat.

Those are two different jobs, and doing one well says nothing about the other. A business can be answering hundreds of buyer questions a week on WhatsApp, accurately and profitably, while remaining functionally invisible to every AI engine a prospective new customer might ask instead. The fix is not to abandon the private channel that works. It is to stop treating it as the only place the knowledge lives, and start putting a public, structured, corroborated version of the same answers where a machine can find them too.