MSME & Global Commerce · emerging evidence

The Zero-Click Economy Arrives in India: Inside the 83% of AI-Answered Searches That Never Reach an MSME's Page

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

A zero-click search is one that ends inside the results page, with no visit to any website. Figures widely attributed to Similarweb clickstream analysis put the zero-click rate near 80% median and about 83% average for queries that trigger a Google AI Overview, against roughly 60% for queries without one. That gap is why the answer layer matters: when a machine reads the web and writes the reply, being present online stops being the same as being found. India is where this lands hardest and fastest. The country has about 958 million active internet users, Google has extended its AI Mode to Hindi and six more Indian languages, and Indian audiences are among the heaviest users of AI assistants in the world. For the tens of millions of micro, small, and medium enterprises registered on Udyam, the practical question is no longer where they rank. It is whether the systems now writing the answers can read, trust, and name them at all. This piece reads the public evidence and marks clearly where no audited India-specific dataset yet exists.

The 83% number, and what it actually measures

The headline figure driving the zero-click conversation is a comparison, not a single number. Queries that surface a Google AI Overview show a zero-click rate near 80% at the median and about 83% on average, while queries without one sit closer to 60%. Those percentages are widely attributed to Similarweb clickstream analysis and repeated across industry reporting through 2025 and 2026. They describe the same behavior from two angles: when the page answers the question, most people stop there.

A zero-click search is not a failure of search. For a decade it has been the quiet majority of what search does, because many queries are questions with short factual answers that never needed a website. Similarweb's own reporting traces the no-click share of Google searches rising from roughly 56% to 69% over the period AI Overviews rolled out, and separately puts the broad zero-click rate at about 60% in 2024 climbing toward 68% by 2026, with AI Overviews now appearing on more than 20% of all searches. The direction is not in dispute. The web is being read on the searcher's behalf more often than before.

The counterweight matters. Semrush, analysing about 10 million keywords from January to November 2025, found that when it tracked the same keywords before and after an AI Overview appeared, the zero-click rate actually eased from 33.75% to 31.53%. Its reading is that Google tends to place AI Overviews on informational queries that already produced few clicks, so the high zero-click rate travels with the query type rather than being purely caused by the summary. Both things are true at once: AI-answered queries are overwhelmingly zero-click, and much of that was zero-click already. The correct conclusion is not that AI destroyed traffic overnight. It is that the answer layer concentrates attention on whatever it decides to cite, and everything it does not cite becomes harder to reach.

Why India is the sharp end of this shift

The zero-click economy is a global pattern, but its edge is sharpest where AI adoption is fastest, and that is India. The IAMAI and Kantar Internet in India 2025 report counts about 958 million active internet users, with rural India now home to roughly 548 million of them and growing faster than urban India. The same report found that 44% of users had already used AI-enabled features such as voice search, image-based search, and chatbots. This is not an early-adopter niche. It is close to half of the largest connected population on earth.

Google has moved its answer surfaces onto Indian ground directly. Through 2025 it expanded AI Mode, its fully generative search experience, to Hindi and then to Bengali, Kannada, Malayalam, Marathi, Tamil, Telugu, and Urdu, and launched Search Live in India. Chief executive Sundar Pichai singled out the United States and India as the two markets where AI Mode first went live. When the answer engine speaks the language a buyer searches in, the summary stops being an English-web novelty and becomes the default way a query in Chennai or Lucknow gets resolved.

The assistant apps tell the same story. Industry usage reports through late 2025 and early 2026 describe India as ChatGPT's single largest market by monthly active users and one of the largest for Gemini, with India accounting for a very large share of global AI-assistant usage. Treat the specific app-by-app counts as vendor and analyst estimates rather than audited figures, because they move month to month and rarely disclose method. The pattern beneath them is consistent: Indian buyers are asking machines the questions they used to type into a search box, and a growing share of those questions get answered without a link ever being clicked.

The MSME exposure: found by a machine, not a buyer

India has formalised its small-business base at remarkable speed. More than 7 crore enterprises have registered across the Udyam Registration Portal and the Udyam Assist Platform since mid-2020, according to the Ministry of MSME, and about a quarter of proprietary MSMEs are now women-led. Registration is a legal and credit milestone, not a visibility one. A firm can hold a Udyam number, a GST identity, and a physical shopfront and still be effectively unreadable to the systems that now compose answers.

That is the exposure. For most of the search era, a small business competed to rank: to appear as one blue link among ten, where a determined buyer would scroll and click. In the answer era the machine reads many sources, decides which few to trust, and names a short list inside a single reply. If the firm's identity, category, location, hours, and credibility are not structured, consistent, and corroborated across the open web, it does not lose a rank. It is simply absent from the sentence the buyer reads.

How much narrower the answer funnel is for local businesses specifically is not yet settled by audited research. Marketing-industry monitoring has reported that generative assistants recommend a far smaller set of local businesses than classic local search does, but those specific rates come from vendor analysis, not a standards-body or peer-reviewed study, and should be read as directional. What is established is the mechanism, not the magnitude: the peer-reviewed Generative Engine Optimization study published at KDD 2024 showed that adding cited statistics, quotations, and authoritative sources measurably raised whether a source was cited inside a generated answer. The lever is no longer position on a page. It is how legible and corroborated a business is to the reader that is now a model.

ONDC and the open-network counter-move

India did not wait passively for platforms to decide who gets discovered. The Open Network for Digital Commerce, built on the open-source Beckn protocol, was designed to unbundle discovery, ordering, and fulfilment so that a small seller on one app can be found and bought by a buyer on any other compatible app. As of a December 2025 government statement, more than 1.16 lakh retail sellers from 630 cities and towns were live on the network across food, grocery, fashion, and electronics, joined by a wider ecosystem of buyer and seller applications. Because ONDC charges no central network commission and asks primarily for a valid tax identity, it lowers the cost of being present for exactly the kirana stores and artisans the platform economy priced out.

The ambition and the reality both deserve accuracy. ONDC's leadership had projected 30 to 40 million monthly transactions by March 2025, and independent reporting through late 2024 put actual volumes well below that early target as the network matured. Treat the growth as real and the pace as contested. The point for this article is structural rather than promotional: ONDC changes who can list, but it does not by itself change who gets read. A product record on an open network still has to carry clean, structured, machine-parseable attributes to be discoverable, whether the reader is a buyer app's ranking logic or an AI assistant summarising options.

This is where the open-network story and the answer-engine story converge. ONDC solves the right to be present at low cost. Machine readability decides whether that presence is legible enough to be surfaced. A seller who joins the network but publishes thin, inconsistent, uncorroborated data has cleared the entry gate and stopped at the threshold, present but not found. The open network raises the floor. It does not remove the requirement to be readable.

The agent is the next reader: AP2 and autonomous commerce

The zero-click search removes the click. The next shift removes the human from the transaction. In September 2025 Google announced the Agent Payments Protocol, an open standard developed with more than 60 payments and technology partners, among them Mastercard, American Express, PayPal, Coinbase, Worldpay, and Etsy, to let AI agents transact on a person's behalf. It works through cryptographically signed mandates: an Intent Mandate that captures what the user asked for, a Cart Mandate that fixes the exact items and price, and a payment step that produces what Google calls a non-repudiable audit trail. The protocol exists to answer three questions an agent-led purchase raises: did the user authorise it, does it reflect their true intent, and who is accountable if it goes wrong.

Read that against ONDC and the answer layer together and the trajectory is clear. A buyer asks an assistant for a product. The assistant reads the web and the open network, selects options, and, increasingly, can carry the purchase through to payment without the buyer visiting a single storefront. In that flow there is no page view to optimise, no ad to click, and no session to retarget. There is only the question of whether the agent could read the business, trust its data, and include it in the cart it assembled.

For an Indian MSME this reframes the entire objective of being online. The old goal was traffic. The emerging goal is to be legible and trustworthy enough that both the human-facing answer engine and the machine-facing purchasing agent can name the firm and act on its behalf. Nothing about agentic commerce is fully settled, and the standards are young. But the direction of every major move, from AI Mode in seven Indian languages to open payment protocols for agents, points the same way: the reader of a business is becoming a machine, and readability is becoming the precondition for revenue.

What no public dataset measures yet, and how it could be

It would be easy to state a precise figure for how much AI answers have cut traffic to Indian small businesses. No audited, India-specific, MSME-level dataset of that kind is public. The zero-click percentages come from global clickstream panels weighted heavily toward Western markets. The local-recommendation funnel figures come from vendor monitoring. The assistant usage counts come from analyst estimates. Each is useful as a signal. None is a measured account of what happens to a sweet shop in Indore or a boutique in Kochi when a buyer asks an AI for a recommendation.

The gap is measurable in principle, and naming the method matters more than inventing a result. A credible study would sample real Indian buyer queries across verticals and cities, in English and in Indian languages, run them through the answer surfaces buyers actually use, and record three things for each business: whether it appears in the generated answer at all, whether the details are accurate, and how its machine-readable footprint, its structured data, its consistency across directories, and its third-party corroboration, correlates with being cited. That study does not yet exist in the public literature. Until it does, the claim that holds is bounded: the mechanism is established, the India-specific magnitude is not, and anyone quoting a single clean percentage for Indian MSME traffic loss is quoting a global proxy, not a domestic measurement.

The practical response does not depend on that missing number. Whether the answer funnel for a given business is slightly narrower or dramatically narrower, the corrective is the same. Find out where the business currently stands across every surface a buyer now uses, fix the readability and corroboration gaps in priority order, and re-measure. The value is in the diagnosis, not in a frightening headline.

How to read this

Two conclusions hold, and both resist the extremes. The first is that the zero-click economy is real and arriving in India faster than almost anywhere, driven by near-universal internet reach, multilingual AI search, and unusually high assistant adoption. When about 83% of AI-answered queries end without a click, the businesses named inside the answer capture the attention, and the businesses left out of it lose a channel they may not even know they had.

The second is that this is a legibility problem before it is a marketing problem. ONDC lowers the cost of being present, AP2 hints at a future where agents buy on a person's behalf, and AI Mode already reads and answers in the languages Indian buyers search in. The common thread is a machine doing the reading. A small business that is structured, consistent, and corroborated can be found and chosen in that world. One that is present but unreadable cannot, no matter how good its product is.

So the useful question is not whether AI is taking traffic. It is where this specific business actually stands, right now, across classic search, the local map pack, AI answers, and its own reputation, and which correction moves it first. That is a measurement discipline, and it is the one that turns an alarming statistic into a plan.

The evidence

Key findings, with their sources

Queries that trigger a Google AI Overview show a zero-click rate near 80% at the median and about 83% on average, versus roughly 60% for queries without one.

contested Figures widely attributed to Similarweb clickstream analysis, reported across industry coverage, 2025-2026.
The no-click share of Google searches rose from about 56% to 69% over the period AI Overviews rolled out, with AI Overviews now appearing on more than 20% of all searches.

emerging Similarweb clickstream reporting, 2025-2026.
Analysing about 10 million keywords, when the same keywords were tracked before and after an AI Overview appeared, the zero-click rate eased from 33.75% to 31.53%; AI Overviews peaked at just under 25% of queries in July 2025 and settled below 16% by November.

established Semrush study, via Search Engine Land, 2025.
India has about 958 million active internet users, roughly 548 million of them rural, and 44% of users have used AI-enabled features such as voice search, image search, and chatbots.

established IAMAI and Kantar, Internet in India 2025 report.
Google extended its generative AI Mode to Hindi and then to Bengali, Kannada, Malayalam, Marathi, Tamil, Telugu, and Urdu, and launched Search Live in India during 2025.

established Google / Search Engine Land / TechCrunch, 2025.
More than 7 crore enterprises have registered across the Udyam and Udyam Assist platforms since mid-2020, and about a quarter of proprietary MSMEs are women-led.

established Ministry of MSME, Government of India, Year End Review 2025 (PIB).
More than 1.16 lakh retail sellers from 630 cities and towns were live on ONDC as of December 2025; ONDC is built on the open-source Beckn protocol and charges no central network commission.

established Government statement via IANS, December 2025; Nasscom on Beckn.
ONDC leadership projected 30 to 40 million monthly transactions by March 2025; independent reporting put actual volumes well below that early target as the network matured.

emerging Business Standard, ONDC CEO interview, July 2024, and subsequent reporting.
Google announced the Agent Payments Protocol (AP2) in September 2025 with more than 60 partners, using signed Intent, Cart, and payment mandates to let AI agents transact on a user's behalf.

established Google Cloud Blog, "Announcing Agent Payments Protocol (AP2)", September 2025.
Adding cited statistics, quotations, and authoritative sources measurably raised whether a source was cited inside generated answers in the engines tested.

established Aggarwal et al., "GEO: Generative Engine Optimization", KDD 2024, arXiv:2311.09735 (peer-reviewed).

Calibration

What is proven, what is promising, what is unproven

Evidence tier	Tactics	What the evidence says
established	The mechanism of the answer layer: AI-answered queries are overwhelmingly zero-click; India's scale, multilingual AI search, and high AI-feature adoption; the GEO levers that earn a citation; ONDC's open-network reach; AP2's agent-payment standard.	Semrush via Search Engine Land; IAMAI-Kantar 2025; Google / TechCrunch language rollouts; Aggarwal et al. KDD 2024; IANS on ONDC; Google Cloud on AP2.
emerging	The precise size of the traffic shift; ONDC transaction growth versus its early targets; assistant market-share counts in India; machine legibility as the decisive discovery factor for MSMEs.	Similarweb clickstream aggregates; ONDC volume reporting; analyst app-usage estimates; extension of the GEO mechanism to local commerce.
contested	The exact 83%-versus-60% zero-click gap, and the claim that answer engines recommend an order of magnitude fewer local businesses than classic local search.	Industry attribution to Similarweb and vendor monitoring; no audited, India-specific, MSME-level study yet corroborates the magnitude.

Reference

Glossary

Zero-click search: A search that ends on the results page itself, with no click through to any website, because the answer is delivered in place, increasingly by a summary the search engine writes with AI.
AI Overview: Google's AI-written summary that appears above the traditional results for a query, synthesising an answer from multiple sources and citing a small selection of them.
Machine legibility: The degree to which a business's identity, offering, location, and credibility are structured, consistent, and corroborated in the forms that answer engines and purchasing agents can parse and cite.
ONDC and the Beckn protocol: India's Open Network for Digital Commerce, built on the open-source Beckn protocol, which unbundles discovery, ordering, and fulfilment so a seller on one app can be found and bought through any compatible app.
AP2 (Agent Payments Protocol): An open standard announced by Google in 2025 that lets AI agents transact on a user's behalf using cryptographically signed intent, cart, and payment mandates.
Generative Engine Optimization (GEO): The emerging discipline of structuring and corroborating content so that generative answer engines are more likely to cite a source inside the answer they compose.

Straight answers

Frequently asked questions

What is a zero-click search?

It is a search that ends inside the results page, with no visit to any website. The answer is shown in place, so the searcher never clicks through. Figures widely attributed to Similarweb put the zero-click rate near 80% median and about 83% average on queries that trigger an AI Overview, against roughly 60% on queries without one.

Is the 83% figure a measured fact?

It is a reported figure attributed to Similarweb clickstream analysis and repeated across industry coverage, not a government or peer-reviewed statistic. A separate Semrush study of about 10 million keywords found that AI Overviews tend to appear on queries that were already low-click, so the summary is not the sole cause. Treat the direction as solid and the exact number as an industry estimate.

Why does this matter more in India than elsewhere?

India has about 958 million active internet users, Google has brought its generative AI Mode to Hindi and six other Indian languages, and Indian audiences are among the heaviest users of AI assistants in the world. When the answer engine reads and replies in the language a buyer searches in, the summary becomes the default way a query is resolved, so being absent from it costs more.

Does joining ONDC solve the discovery problem?

ONDC lowers the cost of being present by letting a small seller be found across many buyer apps without a central commission, and more than 1.16 lakh sellers were live on it by December 2025. But an open network still surfaces businesses whose product data is clean, structured, and machine-readable. Being listed is not the same as being read, so machine legibility still decides whether a seller gets surfaced.

What does agentic commerce and AP2 mean for a small business?

Google's Agent Payments Protocol, announced in 2025 with more than 60 partners, lets AI agents select and buy on a person's behalf. In that flow there is no storefront visit to optimise. The only question is whether the agent could read the business, trust its data, and include it in the cart, which again comes down to how legible and corroborated the firm is.

How could the impact on Indian MSMEs actually be measured?

No audited, India-specific, MSME-level dataset exists yet. A credible study would sample real Indian buyer queries across verticals, cities, and languages, run them through the answer surfaces buyers use, and record whether each business appears, whether its details are accurate, and how its machine-readable footprint correlates with being cited. Until that exists, any single percentage for Indian MSME traffic loss is a global proxy, not a domestic measurement.

Provenance

Sources

Search Engine Land, "Google AI Overviews surged in 2025, then pulled back: Data" (Semrush 10M-keyword study) (established)searchengineland.com
Similarweb, "Zero-Click Marketing: What the 2026 Data Means" (clickstream analysis) (emerging)similarweb.com
Search Engine Roundtable, "Similarweb Says No Clicks From Google Grew From 56% to 69% Since AI Overviews" (emerging)seroundtable.com
Pew Research Center, "Google users are less likely to click on links when an AI summary appears", July 2025 (established)pewresearch.org
Aggarwal et al., "GEO: Generative Engine Optimization", KDD 2024, arXiv:2311.09735 (peer-reviewed, established)arxiv.org
Search Engine Land, "Google expands AI Mode beyond English" (established)searchengineland.com
TechCrunch, "Google's AI Mode adds 5 new languages including Hindi, Japanese, and Korean", September 2025 (established)techcrunch.com
The Tech Portal, "Google launches Search Live in India, expands AI Mode to seven Indian languages", October 2025 (established)thetechportal.com
IAMAI and Kantar, "Internet in India 2025" report, via BestMediaInfo (established)bestmediainfo.com
Ministry of MSME, Government of India, "Year End Review 2025" (Udyam registrations) (established)pib.gov.in
IANS, "Over 1.16 lakh retail sellers from 630 cities and towns live on ONDC: Govt", December 2025 (established)ianslive.in
Nasscom Community, "What is the Beckn protocol, the backbone of ONDC?" (established)community.nasscom.in
Business Standard, "Expecting 30-40 mn monthly transactions by March 2025: ONDC CEO Koshy", July 2024 (emerging)business-standard.com
Google Cloud Blog, "Announcing Agent Payments Protocol (AP2)", September 2025 (established, primary)cloud.google.com
Digit, "ChatGPT is king of AI in India: Gemini, Perplexity lag far behind in daily users", December 2025 (contested, analyst estimates)digit.in

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.