MSME & Global Commerce · emerging evidence

The Udyam-to-Visible Gap: Inside the 7.8 Crore Registrations AI Engines Cannot See

Last reviewed 2026-08-09. Written by Chandranshu Kumar, Founder, Raveneye Global. · 12 min read

India has formalized its small-business economy at a pace with no precedent. As of February 28, 2026, more than 7.83 crore, roughly 78 million, enterprises were registered on the Udyam Registration Portal and Udyam Assist Platform, up from 0.79 crore at the end of 2021-22. That record measures one thing well: how many firms exist on paper. It measures nothing about the AI visibility of Udyam registered businesses, that is, whether those firms can be found when a buyer asks a question. And the buyer has changed. India is now the second-largest market for ChatGPT, with about 100 million weekly active users, and Google has brought AI Overviews and AI Mode to Indian search. When a machine reads the web and names a few businesses in an answer, being registered is not the same as being readable. No public dataset yet measures how many Udyam-registered firms surface inside AI answers, but the surrounding evidence on web presence, citation mechanics, and answer-funnel width all point the same way: a large formalization gain has not yet become a discoverability gain. This piece defines that gap and shows what the public record can and cannot tell us about its size.

A formalization record the data can see

Udyam is the Government of India's free online registration system for micro, small, and medium enterprises, launched in July 2020 alongside a revised definition of an MSME based on investment and turnover. Its growth since then is one of the clearest formalization stories any economy has produced this decade. According to the Ministry of Micro, Small and Medium Enterprises, the cumulative count of registered enterprises rose from 0.79 crore at the end of 2021-22 to 1.64 crore, 4.12 crore, 6.19 crore, and then 7.83 crore by February 28, 2026, a figure that combines the Udyam Registration Portal and the Udyam Assist Platform built to bring informal micro units into the fold.

The scale matters because the sector it counts is the spine of the economy. The Ministry reports that MSMEs account for about 30.1 percent of India's GDP, 35.4 percent of manufacturing output, and 45.73 percent of exports, and that the enterprises registered on Udyam are linked to roughly 34.50 crore jobs. Deregistration has been minimal, around 0.17 percent since the portal opened, which suggests the registrations reflect durable businesses rather than paperwork that lapses.

So the paper record is strong, current, and primary-sourced. The question this article asks is narrower and, so far, unmeasured in public: of these tens of millions of formally registered firms, how many are discoverable at the moment a buyer, or a buyer's AI assistant, actually goes looking?

What a registry entry records, and what it does not

A Udyam registration is a government fact about a business. It certifies that the firm exists, states its category, and unlocks access to schemes, priority-sector lending, and procurement benefits. It is genuinely useful, and it is a real signal of intent to operate formally. What it is not is a presence on the open web that a search engine or an answer engine can read.

The two live in different systems. The Udyam database sits behind a government portal; it is not a public catalog of shopfronts, menus, service areas, hours, or reviews. When a person types a buying question into Google, or asks ChatGPT or Gemini for a recommendation, none of those engines query the Udyam registry. They read what the business has published on the open web and what others have said about it. A firm can hold a valid Udyam number and have no website, no complete business profile, no structured description of what it sells, and no third-party corroboration. In that case the registry knows the firm exists and the answer engines do not.

This is the core of the Udyam-to-visible gap. Formalization moved a firm from invisible-to-the-state to visible-to-the-state. It did not, by itself, move the firm from invisible-to-buyers to visible-to-buyers. Those are separate thresholds, cleared by separate work, and the second one is the threshold that now decides whether the firm gets found.

The discoverability floor: most firms lack a readable web surface

If the registry cannot identify who is discoverable, what can? The most direct public evidence is on the supply side: whether registered firms have any readable web presence at all. Here the picture is discouraging, though it comes with caveats worth stating plainly.

A 2025 study by SERP Forge examined 40,000 MSME-registered businesses across seven Indian states. It found that only about 1.6 percent had a website listed, only about 1 percent had a website that actually worked, and of those working sites, only around 10 percent met a basic set of digital-hygiene checks such as HTTPS, mobile-friendliness, an SEO title tag, and visible contact information. This is a single private study rather than an audited government or academic census, and its sampling frame is not nationally representative, so the exact percentages should be read as directional. But the direction is stark and consistent with everything else known about the sector.

That web-presence floor is the precondition for the entire conversation about AI answers. Answer engines synthesize responses from readable, corroborated sources. A business with no working website, or a working site missing the basics, gives those engines almost nothing to read and almost nothing to cite. Before one can even ask whether a firm surfaces in an AI answer, it has to clear the more basic bar of publishing a legible surface at all. On the current evidence, most registered firms have not.

The reader changed: AI answers now sit between buyers and firms in India

While the supply side stayed largely offline, the demand side moved fast. India is now central to the generative-answer shift, not peripheral to it. In February 2026, OpenAI's chief executive said India had reached about 100 million weekly active ChatGPT users, making it the platform's second-largest market after the United States, with Indian students the largest single segment of student users worldwide.

Google has moved in parallel. It brought AI Overviews to India in 2024 in English and Hindi, and in June 2025 launched AI Mode, its fuller conversational search experience, to Indian users among the first countries outside the United States, with Hindi support. The effect is that a growing share of buying questions in India now returns a synthesized answer that names a few businesses, rather than a plain list of ten blue links a buyer scrolls through.

That change reshapes the stakes of the discoverability floor. When search returned a long list, a firm ranked tenth still had a chance of a click. When a machine reads the web and names two or three providers inside a single answer, the firms it does not name are not lower on the page; they are absent from the answer. The narrowing of the funnel is what turns a web-presence problem into a being-chosen problem.

What the click evidence already establishes

The behavioral evidence that answers displace clicks is no longer speculative. A 2025 Pew Research Center browsing-panel study in the United States found that users clicked a traditional search result in about 8 percent of searches when an AI summary was present, against about 15 percent when it was not, and clicked a link inside the summary itself only about 1 percent of the time. The market studied is not India, so the magnitudes do not transfer directly, but the mechanism, an answer capturing attention that used to flow to ranked links, is the same mechanism now operating in Indian search.

The lever that decides inclusion is also documented. The peer-reviewed paper that introduced Generative Engine Optimization, presented at KDD 2024, tested which content properties change whether a source is cited inside a generated answer and found that adding cited statistics, quotations, and authoritative sources measurably raised a source's visibility in the engines tested. The thing being optimized is no longer a rank; it is a citation, and it is earned by how legible and corroborated a business is.

The measurement that does not yet exist, and how it could be built

Here is the boundary of what anyone can currently claim. Put the two records side by side, the 7.83 crore Udyam registrations and the arrival of AI answers as a primary discovery layer in India, and the natural question is: what fraction of registered firms actually surface when a real buying question is asked of ChatGPT, Gemini, or AI Overviews? As of this writing, no public dataset answers that question. There is no government census, no audited academic study, and no standards-body measurement that samples Udyam-registered firms and tests their appearance inside answers produced by AI engines. Any specific percentage presented as that measurement would be invented, and this article does not offer one.

What can be said is how such a study could be built, because the method is straightforward. Draw a representative sample of Udyam-registered firms stratified by category, sector, and geography. For each firm, construct the buying questions a real customer in its market would ask, in English and in the relevant Indian language. Put those questions to the major answer engines under controlled conditions and record whether the firm, or a page that clearly represents it, appears or is cited in the response. Repeat across engines and over time to separate signal from volatility. The output would be a discoverability rate: the share of registered firms that a buyer's AI assistant can actually surface. Until that work is done and published, the rate is unknown.

What the surrounding evidence implies, without pretending to be that measurement, is a low ceiling. If only about 1 percent of sampled MSMEs have a working website, if answer engines cite corroborated and legible sources, and if the answer funnel names only a handful of businesses per query, then the fraction of registered firms that surface in AI answers is bounded from above by their web presence and pushed lower by the narrowness of the funnel. The gap between 7.83 crore registered and the number an engine can name is almost certainly large. Its exact size is a real, answerable, and still-open research question, and it is more useful to name it as open than to fill it with a number no one has measured.

India built the rails; being on them is not being legible

India's response to the discoverability problem is unusually ambitious, and it is worth understanding precisely because it is easy to mistake for a solution to the gap this article describes. The Open Network for Digital Commerce, or ONDC, is a government-backed effort to unbundle e-commerce into an open, interoperable network so that any buyer app can discover and transact with any seller app. It is built on the Beckn Protocol, whose specifications are published openly on GitHub and define domain-agnostic APIs for discovery, catalog browsing, ordering, and fulfillment. As of its May 2025 figures, ONDC reported more than 7.64 lakh sellers and service providers, over 300 network participants, and more than 16 million total orders across 616-plus cities and 26 live domains.

ONDC matters because it makes a seller's catalog machine-readable in a structured, standardized form, which is exactly the kind of surface that agents and answer engines can parse. The frontier is moving further in that direction. In September 2025, Google published the Agent Payments Protocol, or AP2, an open specification developed with more than 60 payments and technology organizations that lets an AI agent prove a real user authorized a specific purchase, designed to extend the agent-to-agent and Model Context Protocol standards. The pieces of an agent-buyable commerce layer are being assembled, and India is building more of that public infrastructure than almost anywhere else.

But infrastructure is a rail, not a destination. Being eligible to appear on ONDC is not the same as having a complete, accurate, well-described catalog that an agent will actually select. Being reachable through an open protocol is not the same as being the business an answer engine names when a buyer asks. The same distinction that separates a Udyam entry from a web presence separates network membership from network legibility. The rails lower the cost of being found; they do not, on their own, make a specific firm the one that gets chosen. That last step is still the firm's own work, and it is the work most registered firms have not yet done.

Closing the gap

Two things are true at once, and holding both is the point. India has achieved a genuine, primary-sourced formalization success: 7.83 crore firms on Udyam is a real number describing real businesses that now exist in the official record. And that success has not yet translated into discoverability, because a registry entry, a web presence, and a citation inside an AI answer are three separate thresholds, and most firms have cleared only the first.

The productive response is neither triumphalism about the registration count nor alarm that AI has erased small firms overnight. It is measurement. The Udyam-to-visible gap is not a slogan; it is a specific, testable distance between how many firms exist on paper and how many a buyer's tools can actually find, and it can be closed one firm at a time by clearing the thresholds in order: publish a readable surface, structure it so machines can parse it, corroborate it so answer engines will trust it, and stay present on the open networks where agents will increasingly shop.

What is missing at the level of the whole economy, a published discoverability rate for registered firms, is also missing at the level of the individual business: most owners have no read on where they actually stand across search, maps, AI answers, and reputation. That is the gap a diagnostic exists to close: a measured starting point.

The evidence

Key findings, with their sources

More than 7.83 crore (about 78 million) enterprises were registered on the Udyam Registration Portal and Udyam Assist Platform as of February 28, 2026.

established Ministry of MSME, via Press Information Bureau and IBEF, 2026.
Cumulative Udyam registrations rose from 0.79 crore (end of FY2021-22) to 1.64, 4.12, 6.19, and 7.83 crore by February 28, 2026.

established Ministry of MSME / PIB, 2026.
MSMEs account for about 30.1% of India's GDP, 35.4% of manufacturing output, and 45.73% of exports; Udyam-registered firms are linked to roughly 34.50 crore jobs.

established Ministry of MSME / PIB, 2025-2026.
In a 2025 study of 40,000 MSME-registered businesses across seven states, only about 1% had a working website, and only ~10% of those met basic digital-hygiene checks.

contested SERP Forge study, via SMEStreet, December 2025 (single private study, directional).
India reached about 100 million weekly active ChatGPT users by February 2026, the platform's second-largest market after the United States.

established OpenAI CEO Sam Altman, reported by Storyboard18 / TechCrunch, February 2026.
Google brought AI Overviews to India in 2024 (English and Hindi) and launched AI Mode to Indian users in June 2025, among the first countries outside the US.

established Search Engine Land; NewsOnAir, 2024-2025.
ONDC reported over 7.64 lakh sellers and service providers, 300-plus network participants, and 16-plus million total orders across 616-plus cities (May 2025 figures).

established Open Network for Digital Commerce, ondc.org, 2025.
The Beckn Protocol that underpins ONDC is published as an open specification defining discovery, catalog, ordering, and fulfillment APIs.

established Beckn / ONDC protocol repositories, GitHub.
Google published the Agent Payments Protocol (AP2) in September 2025 with 60-plus organizations, letting an AI agent prove a user authorized a specific purchase.

established Google Cloud Blog, September 16, 2025.
In a 2025 US browsing-panel study, users clicked a traditional result in about 8% of searches with an AI summary present versus 15% without, and clicked links inside the summary only about 1% of the time.

established Pew Research Center, July 2025.
Adding cited statistics, quotations, and authoritative sources measurably raised a source's visibility inside generated answers in the engines tested.

established Aggarwal et al., "GEO: Generative Engine Optimization", KDD 2024, arXiv:2311.09735.

Calibration

What is proven, what is promising, what is unproven

Evidence tier	Tactics	What the evidence says
established	The Udyam registration counts and growth trajectory; MSME macro contribution; ONDC and Beckn open infrastructure; AP2; ChatGPT India scale; AI Overviews and AI Mode in India; the AI-summary click effect; the GEO citation levers.	Ministry of MSME / PIB (2026); ONDC.org and Beckn GitHub; Google Cloud Blog (2025); OpenAI CEO statement (2026); Search Engine Land / NewsOnAir; Pew Research Center (2025); Aggarwal et al., KDD 2024.
emerging	The framing of a distinct Udyam-to-visible gap, and a proposed method for measuring a discoverability rate for registered firms across answer engines.	A synthesis across primary registration data, web-presence evidence, and answer-funnel mechanics; the specific cross-measurement has not yet been published.
contested	The precise share of registered MSMEs with a working, standards-passing website, and any specific fraction that surfaces in AI answers.	The web-presence figure is a single private study (SERP Forge, 2025), not a representative census; the AI-answer discoverability rate is unmeasured in public and stated here as an open question.

Reference

Glossary

Udyam registration: The Government of India's free online registration for micro, small, and medium enterprises, launched in July 2020, which certifies a firm's existence and category and unlocks scheme, lending, and procurement benefits. It is a government record, not a public web presence.
The Udyam-to-visible gap: The distance between how many firms are formally registered on paper and how many a buyer, or a buyer's AI assistant, can actually find and choose. Formalization clears the registry threshold; discoverability is a separate threshold that most firms have not cleared.
ONDC (Open Network for Digital Commerce): A government-backed open network that unbundles e-commerce so any buyer app can discover and transact with any seller app, built on the open Beckn Protocol. It makes seller catalogs machine-readable but does not by itself make a given firm the one an engine chooses.
AP2 (Agent Payments Protocol): An open specification published by Google in September 2025 with 60-plus organizations that lets an AI agent prove a real user authorized a specific purchase, extending the agent-to-agent and Model Context Protocol standards. Part of the emerging agent-buyable commerce layer.
Machine legibility: The degree to which a business's identity, offering, and credibility are structured, consistent, and corroborated in forms that answer engines and agents can parse and cite. Being registered, or even reachable on an open network, is not the same as being legible.
Discoverability rate: The still-unmeasured share of registered firms that a buyer's AI assistant actually surfaces when asked a real buying question. A study could measure it by sampling Udyam-registered firms and testing their appearance across answer engines.

Straight answers

Frequently asked questions

How many businesses are registered on Udyam?

More than 7.83 crore, roughly 78 million, enterprises were registered on the Udyam Registration Portal and Udyam Assist Platform as of February 28, 2026, according to the Ministry of MSME, up from 0.79 crore at the end of 2021-22. It is one of the fastest formalization records any economy has produced.

Does a Udyam registration make a business show up in Google or ChatGPT?

No. Udyam is a government registry that certifies a firm exists; search and answer engines do not query it. Those engines read what a business publishes on the open web and what others say about it. A firm can hold a valid Udyam number and still have no website and no presence an engine can find or cite.

How many registered MSMEs are actually findable in AI answers?

No public dataset yet measures this. There is no government, academic, or standards-body study that samples Udyam-registered firms and tests their appearance inside answers produced by AI engines. Anyone quoting a specific percentage is inventing it. The surrounding evidence, very low web-presence rates and a narrow answer funnel, implies the fraction is small, but the exact number is an open research question.

Does ONDC solve the discoverability problem for small sellers?

It helps by making catalogs machine-readable on an open, interoperable network, which is the kind of structured surface agents can parse. But being eligible to appear on ONDC is not the same as having a complete, well-described catalog that an agent selects, or being the business an answer engine names. The rails lower the cost of being found; the firm still has to do the work of being chosen.

Why does AI search matter so much in India specifically?

India is now the second-largest ChatGPT market with about 100 million weekly active users, and Google has brought AI Overviews and AI Mode to Indian search. A large and growing share of buying questions now returns a synthesized answer that names a few businesses rather than a long list, so firms the engine does not name are absent from the answer, not merely lower on the page.

What should a registered business do first?

Clear the thresholds in order: publish a working, standards-passing web surface; structure it so machines can parse what the firm sells and where; corroborate it so answer engines will trust it; and stay present on the open networks where agents will increasingly shop. Starting with a measured read of where a firm currently stands keeps the effort aimed at the threshold it is actually stuck at.

Provenance

Sources

Ministry of MSME, "Over 7.83 crore enterprises registered on Udyam Registration Portal", Press Information Bureau, 2026 (established)
IBEF, "Over 7.83 crore enterprises registered on Udyam platforms", 2026 (established, mirrors the PIB release)ibef.org
Ministry of MSME, "MSME sector accounts for 30.1% of India's GDP, 35.4% of manufacturing and 45.73% of exports", Press Information Bureau (established)pib.gov.in
SME Futures, "7.83 crore MSMEs register on Udyam portal, create 34.50 crore jobs: Minister", 2026 (established)smefutures.com
SERP Forge study of 40,000 MSMEs, via SMEStreet, "MSME Digital Presence Remains Low Despite India's Online Push", December 2025 (contested, single private study)smestreet.in
Storyboard18, "India now has 100 million weekly ChatGPT users, says Sam Altman", February 2026 (established)storyboard18.com
Search Engine Land, "Google expands AI Mode beyond English", 2025 (established)searchengineland.com
NewsOnAir, "Google expands AI Overviews feature to India and five other countries", 2024 (established)newsonair.gov.in
Open Network for Digital Commerce, network statistics (May 2025 figures), ondc.org (established)ondc.org
Beckn Protocol core specifications, GitHub (open specification) (established)github.com
ONDC Protocol Specifications, GitHub (established)github.com
Google Cloud Blog, "Announcing Agent Payments Protocol (AP2)", September 16, 2025 (established)cloud.google.com
Pew Research Center, "Do people click on links in Google AI summaries?", July 2025 (established)pewresearch.org
Aggarwal et al., "GEO: Generative Engine Optimization", KDD 2024, arXiv:2311.09735 (peer-reviewed, established)arxiv.org

Every figure above is attributed to a real, dated source and tagged with its evidence tier. Where a claim could not be verified to a primary source, it is not stated as fact.