The Share-of-Answer Index: Are India's MSMEs Even in the Room When AI Engines Answer?

The question no public dataset answers yet

There is a specific, answerable question hiding inside the phrase "AI search visibility": when a buyer in Pune or Coimbatore asks an answer engine to recommend a supplier, an accountant, or a boutique, how often is a genuine micro or small business named inside the reply? Call the measure a business's share of answer: the fraction of relevant answers the engines generate, across the surfaces buyers actually use, in which that business appears as a cited or recommended source.

For the United States and Europe, a thin but growing body of research has started to estimate versions of this. For India specifically, and for Indian MSMEs in particular, no public dataset yet measures it. That absence is the starting point of this article. What follows is not a proprietary test result. It is a measurement framework and a reading of the surrounding evidence, so that a claim about Indian MSME visibility can be stated as something testable rather than asserted as a slogan.

The stakes are not abstract. India's MSME sector is enormous: as of early 2026, about 7.86 crore enterprises were registered on the Udyam Registration Portal and the Udyam Assist Platform, and the sector accounts for roughly 31 percent of GDP and close to 49 percent of exports, according to India Brand Equity Foundation data. If the layer that now mediates buyer discovery is one that these firms are structurally absent from, that is a distribution problem sitting on top of nearly a third of the economy.

Why India is a live case, not a hypothetical

The reason to ask this about India now, rather than in a few years, is adoption. OpenAI has said India passed 100 million weekly active ChatGPT users, making it the company's second-largest market after the United States. The usage skews young and practical: users under 30 account for roughly 80 percent of ChatGPT activity in India, users aged 18 to 24 alone account for nearly half of the messages sent, and about 35 percent of Indian messages relate to professional tasks, slightly above the global share.

That matters because the cohort forming its search habits around answer engines is the same cohort that will spend the next two decades as buyers, procurement leads, and founders. A behavior that is already default for a hundred million people is not an edge case. It is the emerging center of how discovery works.

The second signal is that classic search itself is being reshaped in India, not only through standalone chatbots. Google's AI Overviews now appear across a meaningful share of Indian search results. Industry trackers such as SE Ranking have reported AI Overviews surfacing in roughly one in six searches in India, a figure that should be read as directional industry monitoring rather than an audited standards-body measurement. The precise number moves month to month and varies by query type, but the direction, more answers synthesized above the links, is not seriously disputed.

A framework: how a share-of-answer index would be built

If the goal is to move from anecdote to measurement, the index needs three axes and a disciplined method. None of this requires speculation; it is a sampling design that anyone could run, and that stating it publicly makes falsifiable.

What the published evidence already implies

A framework with no data is just a diagram. But we are not starting from nothing: several established findings constrain what a first reading would probably show, even before anyone runs the full grid.

The first is the citation mechanism. The peer-reviewed paper that introduced Generative Engine Optimization tested which content levers change whether a source is cited inside a generated answer, and found that adding cited statistics, quotations, and authoritative sources could raise a source's visibility by up to 40 percent in the engines it tested. The lever, in other words, is legibility and corroboration, precisely the properties that thinly documented small businesses tend to lack and that large, heavily cited brands tend to have in abundance.

The second is the behavioral shift. A 2025 Pew Research Center browsing-panel study found that users clicked a traditional search result in about 8 percent of searches when an AI summary was present, against 15 percent without one, and clicked a link inside the summary itself only about 1 percent of the time. Being ranked is no longer the same as being visited; being named inside the answer increasingly is. For a business that has optimized only for classic rank, that is a quiet erosion of the visits it used to earn.

The third is the structure of Indian discovery. For most of the last decade, a small number of aggregator platforms have sat between Indian buyers and Indian small businesses across food, travel, local services, and retail. Answer engines tend to lean on exactly the kinds of large, consistent, heavily linked sources that these aggregators represent, which means the most likely default is that an AI answer names the platform, or the businesses the platform ranks highest, rather than the independent MSME directly. This is an inference from how the engines are known to weight sources, not a measured Indian result, and it is flagged that way.

The counter-current: open networks and agent-ready commerce

There is a genuine reason the pessimistic reading is not the whole story, and it is worth stating fairly. India has spent years building public digital infrastructure designed to reduce exactly the platform concentration described above. The Open Network for Digital Commerce, a state-backed initiative of the Department for Promotion of Industry and Internal Trade incorporated as a not-for-profit in December 2021, was created to decentralize e-commerce by moving discovery and transactions onto an open, interoperable protocol rather than a single company's app.

The scale is no longer trivial. ONDC has reported on the order of 14 million transactions in a single month in late 2024, with roughly 200 percent year-on-year growth, and by early 2024 counted more than 370,000 sellers and service providers fulfilling orders across 800-plus cities, according to figures compiled on the network. These are network-reported numbers rather than independently audited ones, but the trajectory is real, and it puts a large, structured, machine-readable catalog of Indian small businesses into an open system rather than a walled one.

The reason this connects to share of answer is the next layer: machine-readable, protocol-based commerce is the substrate that agent-driven buying will run on. ONDC is built on the open Beckn Protocol, and a small business whose identity, catalog, and reviews live in structured, interoperable form is far easier for a future buying agent to discover and act on than one whose only presence is an unstructured web page. Open networks do not guarantee an MSME shows up in a ChatGPT answer today. But they are the most credible mechanism by which India's smallest firms could become legible to the machines at scale, which is why the pessimistic baseline should be read as a current condition, not a fixed fate.

The baseline, read carefully

Put the pieces together and a careful, non-overstated position emerges. High and young AI adoption means the answer layer already mediates a large and growing share of Indian buyer discovery. The citation mechanism rewards structure and corroboration, which most small firms lack. The click evidence shows that being named in the answer matters more than ranking beneath it. And the discovery structure has historically favored aggregators over independent MSMEs. Every one of those points is established or well-supported. Their combination strongly suggests that the current share of answer for independent Indian MSMEs is low.

What none of those points do is license a specific number. Anyone who tells an Indian small business that it appears in "only X percent" of AI answers, without publishing the grid, the prompts, the engines, and the dates behind that figure, is inventing precision. The responsible claim is narrower and more useful: the baseline is probably low, it is measurable, and it has not yet been measured at population scale for Indian MSMEs. That is an invitation to measure, not a finished finding.

The practical unit of measurement is the individual business. A national index is a research project; a single firm's share of answer is something that can be read today, across the specific engines and buyer questions that matter to it, and tracked as it changes. That per-business read is where the abstract question becomes an operational one, and it is what a Machine-Readiness Score is built to produce.