Thirty Percent of GDP, Absent From the Answer Layer: The Economic Cost of MSME AI Invisibility

The weight is not in dispute

Start with what the government itself reports, because it is unambiguous and repeated across releases. In July 2025, the Union Minister for MSME told Parliament that the sector contributes 30.1 percent of India's GDP, 35.4 percent of manufacturing output, and 45.73 percent of exports. By January 2026, a subsequent Ministry release put the figures at 31.1 percent of GDP and 48.58 percent of exports, with manufacturing holding at 35.4 percent, now attributed to more than 7.47 crore enterprises employing 32.82 crore people. The exact decimal moves release to release, as it should for a sector still being counted in near-real time, but the order of magnitude does not: MSMEs are not a niche of the Indian economy. They are close to a third of it, and closer to half of what the country sells abroad.

The count itself has a paper trail. The Udyam Registration Portal, launched in July 2020, and the Udyam Assist Platform, added in January 2023 to bring informal micro-units into the same registry, together carried more than 7.83 crore registered enterprises by the end of February 2026, up from 6.19 crore a year earlier. That growth rate is partly real expansion and partly formalization, informal businesses that already existed being counted for the first time, but either reading supports the same conclusion: the base is large, growing, and increasingly on the record. Separately, the sector is described in Ministry and industry material as India's second-largest employer after agriculture, a claim consistent across multiple annual releases even where the specific employment figure varies.

None of this is contested. It is government data, reported in Parliament, reproduced consistently across ministry releases and industry bodies. The question this article asks is not whether the weight is real. It is whether the sector carrying that weight is visible to the layer of the internet that is starting to decide who a buyer, domestic or foreign, ever hears about.

What "the answer layer" means, and what this piece is not claiming

When a buyer today asks an AI system, ChatGPT, Google's AI Overviews, Gemini, Perplexity, to recommend a supplier, a manufacturer, or a service provider, the system does not return ten blue links for the buyer to evaluate. It reads a large set of sources, decides which ones it trusts, and synthesizes a short answer that names a handful of businesses inside it. The 2024 peer-reviewed paper that coined the term "Generative Engine Optimization" tested this directly and found that specific, measurable properties of a source, cited statistics, direct quotations, authoritative corroboration, changed whether that source got cited inside the generated answer. Being present on the web and being cited inside an AI answer are different achievements, and the second is the one now standing between a business and a growing share of buyer attention.

This piece does not report a proprietary measurement of Indian MSME AI visibility. No RavenEye test, sample, or multi-engine audit sits behind any number here. What follows is a synthesis of what is publicly known, correctly labeled by how solid it is: government data on the sector's economic footprint, which is established and repeated; India's scale of AI adoption among consumers, which is established and reported by OpenAI directly; and a single vendor-commissioned assessment of MSME-level AI visibility, which is the only public attempt at this specific measurement and should be read as an early signal rather than a verified fact. Where the answer is that no one has measured this rigorously yet, that is what this piece says, along with what a rigorous measurement would need to look like.

The one public attempt at measuring the gap, and its limits

Between February and May 2026, an AI-visibility research firm called Zaillor assessed 700 Indian SMEs across eight sectors, seeded from a cohort at the AI Impact Summit in New Delhi, evaluating publicly available, machine-readable signals across ChatGPT, Gemini, and Claude. The headline figures were an average AI Visibility Score of 50.4 out of 100, and a finding that 95 percent of the businesses assessed fell short of what the firm classified as strong, consistent visibility. Those numbers have circulated widely in Indian marketing press under framings like "9 in 10 Indian SMEs cannot be found by ChatGPT or Gemini."

Read that finding for what it is. It is the only public dataset that attempts to score MSME-level AI visibility at any scale, and directionally it lines up with everything else known about how generative engines work, they cite a small, curated set of sources rather than ranking a large one, which structurally favors firms with more resources to spend on structured, corroborated web presence. But it is a single vendor's proprietary methodology, run once, not peer-reviewed, not replicated by an independent body, and published by a firm with a commercial interest in the finding. That does not make the number false. It makes it a first data point, not a settled measurement, and it should be cited with that qualifier every time.

A separate and more solidly established data point sits alongside it. SIDBI's May 2025 survey of more than 2,000 MSMEs across 19 industries, drawn from primary field research plus secondary data, found that 90 percent of MSMEs surveyed accept digital payments, while roughly 70 percent still rely on traditional marketing channels rather than digital marketing or e-commerce to reach customers. That is not an AI-visibility statistic. It is evidence of a much more basic gap, transactional digitization running far ahead of marketing digitization, and it matters here because AI visibility is downstream of exactly the signals, a structured website, consistent listings, third-party corroboration, that a firm relying on traditional marketing has usually never built. A business that has not digitized its marketing presence has, almost by definition, not built the raw material an answer engine would need to cite it.

Why scale does not protect a business from invisibility

There is an intuitive but wrong assumption buried in headlines that lead with "30 percent of GDP": that a sector this large must already be well represented online, because scale usually correlates with visibility. Generative answer engines break that correlation in a specific way. They do not aggregate a market and show it to the buyer; they select a small number of names and present those as the answer. A sector's aggregate economic weight, spread across 7.8 crore individually small and mostly under-resourced enterprises, produces almost no advantage at the level where an AI system decides which two or three businesses to name in a single generated answer. The GDP share is a property of the sector. Machine legibility is a property of the individual firm, and most individual MSMEs, per both the SIDBI marketing-adoption data and the early Zaillor visibility data, have not yet built it.

This matters most acutely for the export side of the sector's weight. A buyer sourcing from a supplier abroad increasingly starts that search inside an AI system rather than a browser, and the properties that earn citation, structured product data, third-party verification, consistent claims across sources, are exactly the properties a firm optimized for domestic word-of-mouth or a marketplace listing typically lacks. The nearly half of India's exports that MSMEs generate is not a guarantee that the firms behind those exports will keep being found the way global sourcing search itself changes.

The domestic amplifier: 100 million weekly users

The export argument would matter even if AI-mediated search were a foreign phenomenon happening to Indian sellers from the outside. It is not. In February 2026, OpenAI's Sam Altman said India had reached 100 million weekly active ChatGPT users, making it the platform's second-largest national market after the United States, with the country's user base reported to have quadrupled over the prior year and India holding the largest student user base of any country on the platform. That is not a niche export-facing audience. It is a domestic consumer and business base large enough that AI answers are becoming a mainstream channel through which Indian buyers, not just overseas ones, decide which local business to call.

Put the two facts next to each other. A sector that is close to a third of GDP sits behind enterprises that, per the best available survey data, are mostly still marketing themselves through traditional channels. The population most likely to be asking those enterprises' questions through an AI system, rather than a search engine, is now one of the largest in the world for that behavior. The gap between economic weight and machine legibility is not a slow-moving structural condition MSMEs can address at their own pace. It is a gap opening under a buyer population that has already moved.

What a rigorous measurement of this gap would require

No public dataset currently ties firm-level AI-visibility scores to sector-level economic contribution, at any statistically meaningful sample size, in India or elsewhere. That is worth stating plainly rather than papering over with a single vendor's 700-firm snapshot. A rigorous version of that measurement would need, at minimum, a sample stratified across the trading, services, and manufacturing splits the Udyam data actually shows (roughly 43, 36, and 21 percent of registrations respectively), tested across multiple AI engines with a documented, replicable prompt methodology, run at more than one point in time to separate a firm's baseline visibility from noise, and ideally cross-referenced against firm-level export or turnover data to see whether visibility and economic weight actually correlate or diverge.

That study does not yet exist in the public record. Until it does, the responsible claim is narrower than "MSMEs are invisible to AI": it is that the one public attempt at measuring this found a low average score and a large share of weakly visible firms, that the finding is directionally consistent with everything else known about how generative engines select sources, and that no individual business should assume its own position from a sector-wide vendor average. The only way to know where a specific firm stands is to measure that firm specifically, which is a different exercise from citing a national percentage.

How to read this

Two things are true at once, and this article holds both. The government figures on MSME economic weight, 30 to 31 percent of GDP, 35.4 percent of manufacturing, 45 to 49 percent of exports, more than 7.8 crore registered enterprises, are established, repeated, and not in question. What is not yet established is a rigorous, independent, India-specific measurement of how much of that weight is legible to the AI systems now mediating an increasing share of buyer decisions, domestically and for export. The single public attempt at that measurement points toward a real gap; it does not close the question of its exact size.

The practical conclusion does not depend on resolving that open question at the national level, because no national average tells an individual business where it stands. It tells a policymaker or a researcher where to look next. For an individual MSME, the only way to answer "am I part of the visible minority or the invisible majority" is to measure that specific business against the surfaces buyers now actually use, not to extrapolate from a sector-wide percentage that was never built to describe any single firm.