Formalize First, Get Found Never: Which Udyam Sectors Are Most Invisible to AI Search

A formalization wave, sorted by activity type

The Udyam Registration Portal and the Udyam Assist Platform have pulled more Indian enterprises into the formal economy than any single scheme before them. As of February 28, 2026, more than 7.83 crore enterprises had registered since the portal launched on July 1, 2020, and by mid-July 2026 the government put the combined URP and UAP count above 8.9 crore, with employment attributed to more than 38 crore people. Every one of those registrations is filed under a National Industrial Classification code, which sorts the enterprise into one of three broad activity types: manufacturing, trading, or services.

That sorting is not incidental. It is the closest thing India has to a sector-level census of its small business base, and it produces a distribution that most casual observers would not predict. Trading enterprises, meaning wholesalers, retailers, and resellers, are the single largest category, not manufacturing and not services.

The split: trade leads registration, manufacturing leads output

As of February 27, 2026, trading enterprises accounted for 42.89% of all registered MSMEs, with services at 36.22% and manufacturing at 20.89%, per Ministry of MSME data compiled by the India Brand Equity Foundation, a trust of the Department of Commerce. The same pattern shows up in employment. In a written reply to the Lok Sabha, the Ministry reported that trading generated the most MSME-linked jobs in FY2025-26 at 2.98 crore, ahead of services at 2.57 crore and manufacturing at 2.50 crore.

Set that against the sector's output profile and the mismatch becomes visible. The Ministry's own release dated January 31, 2026 put the MSME sector's contribution at 35.4% of India's manufacturing output, 48.58% of exports, and 31.1% of GDP, drawing on a base of 7.47 crore enterprises employing 32.82 crore people. Manufacturing MSMEs are a fifth of registrations by count but carry a disproportionate share of what the sector actually produces and sells abroad. Trading enterprises are the largest registrant class but the smallest contributor to manufacturing output and exports, by definition, since resale is not production.

This is the setup for the visibility question. A buyer researching a supplier, a distributor, or a service provider through an AI answer engine is not querying the registration count. They are querying a category of intent, "textile exporters in Tiruppur," "packaging machinery suppliers," "GST-compliant wholesale distributors in Delhi", and the engine returns whichever firms it can read, trust, and cite. Registration count indicates how many firms exist inside a category. It says nothing about how many of them are legible to the system doing the answering.

Two different kinds of "machine-readable," and only one of them faces AI engines

India has spent the past several years building real machine-readability into its MSME economy, but it is worth being precise about which machine is doing the reading. GST e-invoicing, the Government e-Marketplace, and the Open Network for Digital Commerce all require enterprises to publish structured data. ONDC in particular runs on the Beckn Protocol, an open specification whose seller-app schema defines exactly how a catalog, an offer, and a fulfillment status must be encoded so that any compliant buyer app on the network can parse it. That specification is public on GitHub, and it is a genuinely strong form of structured data.

It is also a closed loop. Beckn/ONDC catalog schema makes a firm's inventory legible to other applications on the ONDC network, buyer apps, logistics partners, payment gateways. It does not make that firm's identity, credentials, or offering legible to a general-purpose web crawler or an AI answer engine, because that schema was never designed to be crawled by Googlebot or ingested by a large language model's retrieval pipeline. The structured data that answer engines read is a different vocabulary entirely: schema.org markup embedded in a public webpage, corroborating mentions across independent sources, and content structured well enough to be lifted into a generated answer.

The one indicator that is actually measured: website ownership

The clearest available proxy for whether a formalized MSME can be found by an answer engine at all is not sector-specific AI citation data, because that does not yet exist publicly. It is website ownership, and a large 2025 primary survey has measured it directly. The India SME Forum's META Report Card, based on 7,835 MSMEs surveyed in 2024-25 across manufacturing, services, and trade, found that even among the digitally active cohort, the enterprises that had already adopted at least one digital tool, only 26.9% owned a dedicated business website. The national average across all registered MSMEs was estimated lower, at 20-25%, against 80% or higher among the leading fifth of digitally mature MSMEs.

The same survey found that 53.8% of respondents had integrated the internet, digitization, or e-commerce in some form, and that email, CRM software, and e-commerce platforms were the most common digital tools in use. Website ownership lagged all three. That gap matters specifically for AI visibility because a marketplace listing, a WhatsApp Business profile, or an ONDC catalog entry is not a crawlable, citable web asset in the way a webpage with schema markup is. An enterprise can be fully digital by the survey's definition and still have nothing on the open web for an answer engine to find.

What is not yet measured, and how it could be

Here is the boundary. No public, audited dataset currently reports AI-answer citation or recommendation rates broken out by Udyam sector or by NIC code. The claim this piece's title implies, that a specific sector is measurably "most invisible" to AI search, cannot be made responsibly without that dataset, and nobody, including large industry-monitoring firms, has published one for the Indian MSME base specifically.

What does exist is a directional signal from an adjacent, better-resourced population. Industry monitoring reported via a Q2 2026 AI Citation Benchmark found that 51% of B2B technology brands had zero citations across ChatGPT, Perplexity, and Gemini, even as a cited 2026 Forrester study found 72% of B2B software buyers now consult ChatGPT during vendor evaluation. Read carefully, not as an Indian MSME statistic but as an upper bound: if roughly half of resourced, English-language, website-owning B2B software companies earn zero AI citations, a trading or manufacturing MSME with no website, no case studies, and no earned coverage starts from a considerably weaker position. That inference is reasonable. It is not a measurement, and this piece does not claim to have run one.

A real answer would require a defined method: sample a fixed number of Udyam-registered firms per NIC 2-digit code across manufacturing, trading, and services; construct category-intent buyer queries for each (a supplier search, not a brand-name search); run those queries against two or three major answer engines on a fixed schedule; and record citation, mention, and recommendation rates per sector, corrected for firm size and geography. That study does not exist yet for India. Building it, sector by sector, is the open work.

How to read this

Two things can be true at once. The registration data is unambiguous: trade formalized fastest and dominates enrollment counts, manufacturing carries the sector's output and export weight, and website ownership lags across all three categories, with the digitally active cohort itself only reaching 26.9%. And the specific, sector-by-sector AI-visibility gap that the framing of this piece points toward remains an open empirical question, not a settled finding.

The practical response does not require waiting for that study to exist at the national level. It requires a firm-level answer. A manufacturer, a trader, or a services provider registered on Udyam can each be measured individually, on their own web presence, local listings, and current AI-answer standing, using the same diagnostic method a sector-wide study would eventually scale. The sector question is a research question. The firm question is answerable today.