Metro Privilege: Does an MSME Need to Be in Mumbai or Delhi to Be an AI Answer?

The single-answer format raises a geography question search never had to answer

Classic search returned a list. A business ranked forty-seventh for its category could still be found by someone willing to scroll, and a business with no ranking at all could still show up on the map pack for a nearby search. Generative answer engines collapse that list into a handful of names, sometimes one. Whatever used to correlate loosely with visibility, review volume, backlink profile, a complete Google Business Profile, a mention in the local press, now decides something closer to a binary: named in the answer, or not mentioned at all.

That compression raises an obvious question for a country where most small business owners do not operate out of Mumbai, Delhi, Bengaluru, or the other cities popularly treated as India's metros. If an engine is asked for a chartered accountant, a packaging vendor, or a boutique hotel in a Tier-1 city, does it name a real, checkable business the way local search used to? And if the same category question is asked with a Tier-2 or Tier-3 town in place of the metro, does the engine still name someone, or does it retreat to generic category advice, a national platform, or nothing at all?

This article does not claim to have run that test. It sets out how the test would work, states plainly that no public dataset yet runs it for India, and reads what the surrounding published evidence, on India's internet base, its MSME geography, and the one relevant precedent study that exists elsewhere, already implies about the answer.

What "metro" and "Tier-2/Tier-3" mean, and why the terms are looser than they sound

India does not have one official line that separates a metro from everything else. The government's formal city-classification scheme, descended from the Sixth Central Pay Commission and used to set House Rent Allowance for public servants, ranks cities by population and recognizes eight cities as Tier-X, the closest thing to an official "metro" designation. The "Tier-1, Tier-2, Tier-3" shorthand used constantly in retail, real estate, and marketing is a separate, looser convention, roughly population- and market-size-based, applied inconsistently across industries and rarely defined the same way twice.

That looseness matters for a proposed test, not just as a caveat. A study that wants to compare "metro" against "non-metro" AI-answer behavior has to pick its own working definition before it can produce a number, and it has to be explicit that the definition is a design choice, not a settled category. The method proposed later in this piece anchors "metro" to the eight-city Tier-X set and "Tier-2/Tier-3" to towns matched on population and on documented MSME density within the same business category, specifically to avoid the trap of comparing categories or market sizes instead of geography.

Where India's MSMEs and internet users actually are

Before asking whether AI answers favor metros, it helps to know how small a share of the addressable population and business base the metros actually represent. On population, DataReportal's Digital 2025: India report put 62.9% of the country's population in rural areas as of January 2025, against 37.1% urban, a split that by itself excludes most Indians from anything resembling a Tier-1 metro.

On internet access specifically, the IAMAI-Kantar ICUBE 2025 report counted 958 million active internet users nationally, of whom 548 million, 57%, were rural, with rural adoption growing at roughly four times the pace of urban growth. The finding is drawn from a sample of close to 100,000 respondents surveyed across more than 400 towns and over 1,000 villages, which is itself a useful marker of how far outside metro geography the country's actual internet population now sits.

The MSME base looks similar. The Ministry of MSME's own account of the sector, published via the Press Information Bureau in May 2026 under the title "Empowering the Grassroots Economy: A Comprehensive Push for Rural and Semi-Urban MSMEs," states that MSMEs contribute about 31.1% of GDP, 48.58% of exports, and employ close to 32.8 crore people, explicitly framing "a large share of these enterprises" as operating in rural and semi-urban areas, with over 7.9 crore enterprises registered on Udyam and Udyam Assist as of March 2026. A second PIB release from June 2026 puts registrations at over 8.7 crore by that month and employment at 38.9 crore people, the second-largest source of employment after agriculture. The Udyam dashboard itself, checked live, shows roughly 9.25 crore total registrations, of which 99.3% are classified as micro enterprises, the segment with the thinnest marketing budget and least capacity to invest in how it is described online.

State-level data sharpens the point. In a written reply to the Lok Sabha on 23 July 2026, the Minister of State for MSMEs reported that Uttar Pradesh led all states in Udyam-linked employment for FY2025-26 with 1.18 crore jobs, followed by Maharashtra, Tamil Nadu, Telangana, and Bihar. Four of those five leading states are large, geographically dispersed states whose MSME employment is spread across hundreds of Tier-2 and Tier-3 towns, not concentrated inside one or two metro city limits. India's MSME employment base is not a metro story with a rural footnote. It is closer to the reverse.

What the closest published evidence says, and does not say, about geography

No India-specific study of AI-answer geographic bias exists in the public domain as of this writing. The closest available precedent is a 2025 whitepaper from Local Falcon, a local-search analytics vendor, which ran a geo-grid study of Google AI Overviews across 4,423 businesses in 20 countries using 60,000 queries, with a deeper United States analysis using a 7-by-7 latitude-longitude grid around individual businesses. Two findings are relevant here. The study found effectively no correlation between a business's literal distance from the simulated searcher and its ranking position inside the AI Overview, a correlation coefficient of 0.001, though businesses at the center of a search grid appeared somewhat more often overall (72.0%) than those at the edge (68.5%). And queries that named a specific location triggered an AI Overview less often, 35.0% of the time, than location-agnostic queries at 46.1%.

Read carefully, that evidence cuts against the simplest version of a metro-bias story. It is not that the model has learned to prefer a Mumbai address as such; raw proximity barely moved the needle in the one rigorous geo-grid test that exists. The study's own read of a saturated market, dense competition for frozen desserts in Philadelphia showing near-uniform visibility across an entire grid, suggests the engine defaults to whichever business has the most corroborating signal available, treating that as a proxy for authority rather than rewarding a location for being a location.

The caveat is that this is a single-market study, testing distance within a few miles, not a cross-country comparison between a Tier-1 metro and a Tier-2 or Tier-3 town with a different competitive density, review volume, and baseline of press coverage. It tells us proximity alone is probably not the mechanism. It does not tell us whether India's metro-versus-non-metro gap in review volume and citation density produces the same practical outcome as a bias would, even without the model "knowing" anything about geography at all.

A proposed method: the paired query

The direct test has a straightforward design and requires nothing beyond access to the major engines and a documented list of business categories. Select categories with clear MSME representation, drawn from the categories already visible in the Ministry of MSME's own data, for example a chartered accountant, a packaging manufacturer, or a boutique hotel. For each category, write one category-defining query with no proper noun attached, the kind a real buyer would type, then run it twice per engine: once with a Tier-X metro named as the location, once with a matched Tier-2 or Tier-3 town named in its place, chosen for having a documented MSME cluster in that same category so the comparison is not confounded by category or by a town with no relevant business at all.

Run each pairing across more than one engine, ChatGPT, Perplexity, Google AI Overviews or AI Mode, and Gemini at minimum, and repeat each query per engine to average out the model's own response variance, the same reason the Local Falcon methodology split its query set across desktop, mobile, and seven distinct intent types. Score each response on one axis first: does it name a real, verifiable local business with a checkable address, or does it fall back to generic advice, a national aggregator, or a refusal. A second axis, whether the named business is well corroborated (reviews, a complete profile, press mentions) or thinly documented, would test the corroboration-density hypothesis directly rather than assuming it.

None of this has been run yet for India, and this piece does not claim otherwise. The method is buildable from data that already exists, a comparably careful precedent already exists for a single market, and the state of the evidence is an open, testable question rather than a settled finding in either direction.

What the surrounding evidence already implies

Two things are true at once. The population and business base an answer engine would need to represent fairly are now majority non-metro by any reasonable count, concentrated in large states whose MSME employment is spread across hundreds of Tier-2 and Tier-3 towns rather than clustered inside eight city limits. At the same time, the one rigorous precedent that exists suggests engines reward whichever business is best corroborated, wherever it sits, rather than literal proximity to a metro address, a mechanism that could still produce a metro-favoring outcome in practice if metro businesses are, on average, more likely to have the review volume and press mentions that make a business legible to a model, without the model ever "knowing" it prefers a city.

That reframing matters for what a Tier-2 or Tier-3 MSME owner can do about it. If the mechanism were literal geography, there would be no fix short of relocating. If the mechanism is corroboration density, closing the gap is a fundable, sequenced piece of work: a complete and accurate business profile, a real base of reviews, and consistent listings across the directories a category-specific buyer might check. That is a testable claim, not a guaranteed outcome. The base rates make the concern worth testing; nothing published yet proves Tier-2 and Tier-3 firms are underrepresented in AI answers because of where they are rather than how well documented they are.