The Answer Era: AI, ONDC and Machine Discovery, 2023 onward

The newest surface: an answer, not a list

The previous era in this series, Social, Commerce and the UPI Rails, 2016 to 2024, left India's businesses reaching people through a feed, a marketplace app or a payments rail, and left people finding businesses inside those same three surfaces. Those surfaces now reach nearly the whole country: India's active internet user base crossed 958 million in 2025, more than half of it rural, a scale that traces back to the Digital India program the government launched in July 2015 to connect the country's broadband and government services in the first place. From 2023 a fourth surface began to sit on top of the feed, the marketplace and the rail: not a new place to look, but a machine that reads across every place at once and hands back one composed answer. OpenAI's ChatGPT is the clearest evidence of how fast that surface grew in India specifically. By February 2026, CEO Sam Altman said India had passed 100 million weekly active ChatGPT users, OpenAI's second-largest market after the United States and, by the company's own account, the country with the largest number of student users of the product anywhere in the world.

OpenAI backed that position with money and headcount rather than treating India as a market it could win from abroad. A New Delhi office opened in August 2025, the company's first in the country, and the same month brought ChatGPT Go, a subscription tier priced under $5 a month. From 4 November 2025, OpenAI made ChatGPT Go free for a year to new and existing Indian subscribers alike, a promotional window built to convert the country's scale of free usage into a habit around the paid product before a competitor could do the same.

Google took a different route to the same surface: instead of a separate chatbot, it folded the synthesized answer into the search bar most of the country already used every day. AI Overviews launched in India on 15 August 2024, in English and Hindi from the first day, alongside five other countries; Google's own release on the rollout volunteered that Indian users engaged with the AI-written responses more than users in the other markets that received the feature that same week. AI Mode, Google's more conversational search product, followed inside Search Labs in India from 24 June 2025, before the Labs sign-up requirement was dropped for all Indian searchers. Hindi joined AI Mode on 8 September 2025 alongside Indonesian, Japanese, Korean and Brazilian Portuguese, and seven more Indian languages, Bengali, Kannada, Malayalam, Marathi, Tamil, Telugu and Urdu, followed on 8 October 2025.

The carriers become AI distribution rails, again

This series has already named this pattern once, in the era the Jio shock defined: a carrier underprices something valuable to build volume fast, then owns the distribution layer everyone else has to compete for. From 2025, two of India's largest telecom operators ran a version of the same play for an AI subscription instead of a gigabyte of data. On 30 October 2025, Google and Reliance announced that eligible Jio users would receive Google's Gemini 2.5 Pro AI subscription free for 18 months, a package Google valued at Rs 35,100 (about $396), against a Rs 1,950-a-month list price. The offer rolled out first to users aged 18 to 25 before extending to Jio's wider base, and Reliance's AI arm separately became a go-to-market partner for Google Cloud's enterprise Gemini offering across Indian organizations.

Bharti Airtel had moved first. From 17 July 2025, all of Airtel's 360 million mobile, broadband and DTH subscribers could claim a free 12-month Perplexity Pro plan, a subscription Airtel valued at Rs 17,000 (roughly $195), under an arrangement exclusive enough to bar any other Indian carrier from offering the same deal. Perplexity's own India downloads rose sharply the same quarter, part of a broader push in which the company has described India as central to a strategy of partnering with more than two dozen telecom operators worldwide to win distribution it could not buy through app-store marketing alone.

Read the two deals side by side and the economics are not new, only the product is. In the smartphone era, Jio underpriced mobile data to build a subscriber base fast enough to reshape what an ordinary Indian phone could do. A decade later, two carriers are underpricing an AI subscription to build the habit of asking a machine first, at the exact moment three companies are racing to become the answer a person reaches for by default.

The zero-click shift and the machine-readiness thesis

Two separate questions sit inside the phrase "AI search visibility," and they matter in a specific order. The first is whether a person even clicks through to a business's own website once an engine has answered the question for them. The second, and the one that decides whether the first ever becomes relevant, is whether the engine names that business inside the answer at all.

Vernacular AI and the sovereign-model push

An answer composed only in English is not an answer most of India can use, and the country's language infrastructure predates the current AI wave by two years. Bhashini, launched in 2022 under the National Language Translation Mission, covers all 22 scheduled languages and is reported to process around 100 million API inferences a month, drawing on more than 300 AI models and over 1,200 domain-specific datasets; it sits underneath government services from PM-Kisan to the eSanjeevani telemedicine platform. A newer grassroots tool built on the same infrastructure, SabhaSaar, records and translates local-government proceedings and logged more than 15.6 crore, over 156 million, interactions in its first year of national rollout, averaging more than 5 lakh, 500,000, uses a day.

The private sector's answer to the same language problem is a model, not a translation layer sitting in front of an English-trained one. On 26 April 2025, the government selected Sarvam, one of twelve organizations chosen under the IndiaAI Mission's Innovation Centre pillar, to build an indigenous foundation model with dedicated compute and Rs 246.72 crore (about $28 million) in financial support, a selection explicitly framed around strategic autonomy rather than a commercial contract. Sarvam unveiled the results, a 30-billion and a 105-billion parameter model, at the India AI Impact Summit in February 2026, the same event around which Altman's 100-million-user figure for ChatGPT became public. Four months later, Sarvam raised $234 million in the first close of a Series B round led by HCLTech, reaching unicorn status at a $1.5 billion valuation.

Whether a sovereign, India-trained model changes which businesses an engine names in an answer is not something any published study has measured yet, and this piece does not claim it does. What is measurable is the scale of the intent behind the investment: a government-run language layer processing on the order of a hundred million requests a month, and a government-selected foundation model built for reasoning in Indian languages, both aimed at making the answer era's discovery surface legible in the languages most of the country actually searches in, not only in English.

Agentic commerce: ONDC, Beckn and the AP2 frontier

Underneath discovery, India also began opening the transaction itself to a machine, years ahead of most of the world. The idea is not new to this series: it is the same open-protocol logic that built UPI, applied this time to the exchange of goods and services rather than money alone.

Formalization, the digital economy and the answer economy

The answer era did not arrive on an empty economic floor. It arrived on top of a formalization drive that had already brought a large share of India's small businesses into a government record. The Udyam Registration Portal, the self-declared filing that classifies an enterprise as micro, small or medium, had recorded 7.83 crore, about 78.3 million, cumulative registrations by 28 February 2026, a scale that did not exist five years earlier. India's digital economy, measured with an OECD-aligned methodology by the Ministry of Electronics and Information Technology, accounted for close to 11.7 percent of national income in 2022-23, about Rs 31.64 lakh crore or roughly $402 billion, a share the ministry projects, as a forward estimate rather than a measured result, will near one-fifth of GDP by the end of the decade.

A formally registered enterprise and a machine-legible one are not the same achievement, and this is the layer a companion piece on this site calls the answer economy: the state of affairs in which a synthesized reply, not a ranked list a person scrolls through themselves, decides which businesses a buyer even learns exist. Whether India's roughly 78 million registered MSMEs are proportionately represented inside that answer is an open empirical question, not a settled fact in either direction. No public, ongoing dataset yet tracks an Indian share of answer the way rank trackers have tracked Google position for two decades, a gap Raveneye's own India MSME research has named and proposed a method to close.

The political scaffolding: privacy, governance and the sovereignty debate

Law caught up with the surface on the same kind of delay this series has tracked at every prior turn. India's Digital Personal Data Protection Act received presidential assent on 11 August 2023, the country's first full personal-data statute, but its implementing Rules were not gazetted until 14 November 2025, more than two years later, and even then only in phases. The Data Protection Board began operating from that date; obligations on consent managers, cross-border data transfer and large platforms follow on 14 November 2026; and the regime does not reach full force, including penalties that can climb to Rs 250 crore (roughly $29 million) for the most serious violations, until 14 May 2027.

Government guidance on AI itself moved on a similarly staggered timeline. The Cabinet approved the Rs 10,371.92 crore (about $1.2 billion) IndiaAI Mission on 7 March 2024, built around seven pillars that include subsidized compute, now 38,000 GPUs offered to developers at rates as low as Rs 65 an hour, an indigenous foundation-model program, and an explicit Safe and Trusted AI pillar. Detailed, sector-agnostic guidance followed only in November 2025, when the Ministry of Electronics and Information Technology published India's AI Governance Guidelines, built around principles that include a stated "do no harm" standard, after a public consultation that drew more than 2,500 submissions.

Behind both the sovereign-model funding and the governance guidelines sits a live argument this piece can describe without resolving: whether a country's AI layer, the model, the compute it runs on and the data it trains on, needs to sit inside its own borders to count as genuinely sovereign, or whether that framing is running ahead of any settled technical or economic case for it. That debate remains open, and any confident claim about which side is right belongs to a different kind of piece than this one.

The culmination of the arc

This is the sixth and final part of this series for a reason that has nothing to do with running out of eras and everything to do with what discovery has become. Controlled broadcast gave way to satellite choice; satellite choice gave way to telecom and search; telecom and search gave way to the smartphone and the Jio shock; and the Jio shock gave way to the social feed, the marketplace app and the UPI rail. Each of those eras added a new channel or rail alongside the ones that came before it. The answer era does something structurally different: it does not add a new place to look. It inserts a machine between the buyer and every place that came before, list, feed, marketplace and rail alike, and lets that machine decide what surfaces first.

What is documented in this piece is documented: user counts the companies themselves reported, launch dates confirmed by the platforms and by Indian government sources, an Act and its Rules with gazetted dates, a registration count published by a ministry. What is not documented is how that machine treats one specific, individual business, an accountant, a boutique, a small manufacturer, rather than the aggregator or the household name sitting next to it in the answer. That is a measurement problem, not a rhetorical one, and it is the problem a Machine-Readiness Score is built to answer for a single business at a time, rather than assert for the market as a whole.