Classifieds expose the key AI fault line early

When discovery disappears, markets break.

Note from Nikhil: In case you missed it, I mailed out “Why commerce isn’t ready for AI yet” on 26th Jan, a public holiday in India. I’m probably shifting to a thrice a week cadence because I have too much to write, and twice a week doesn’t cut it for me. So Mondays, Wednesdays and Fridays.

*

When the internet shifts from search to invocation, businesses don’t get ranked lower. They disappear.

On a recent earnings conference call, Info Edge CEO Hitesh Oberoi flagged a worrying development, saying:

“This growing shift from Google search to AI chatbots along with Google’s rollout of AI summaries has however led to a decline in traffic in Shiksha over the last couple of quarters and this is something we continue to monitor, and we are working on strategies to mitigate this.”

“A lot of the traffic on Shiksha is SEO traffic. And because of the changes you’re seeing in search, platforms like Google, etc., have started answering questions directly. And there’s an AI overview, and so many other things are changing in search. Traffic on Shiksha has actually fallen. So, we are seeing a degrowth in terms of people ultimately ending up on Shiksha, right from Google and other platforms.”

For most of the internet’s history, loss of visibility was relative. You slipped from position three to position eight. Traffic declined gradually. You could measure it, contest it, buy your way back, or optimise around it.

A large part of a classifieds business is around creating content. Everything from education to real estate classifieds sites use “News” or blogs with “How To…” content to attract traffic. Oberoi mentions Shiksha as a content platform primarily, and even distinguishes from other Info Edge platforms, saying:

“Shiksha is a little different, because it’s a content-led platform. A lot of the content is static, does not change every day.”

”For example, on 99acres, Naukri, the content is very dynamic, the listings change every day, the jobs change every day, etc. All the details are there and people want to go to the details they can only apply on our platform. But Shiksha is a little different, because it’s a content led platform. A lot of the content is static, does not change every day. A lot of the people and a lot of publishers, globally have seen traffic fall, the traffic they were getting from Google, fall.”

At first, the problem is one of substitution. He says:

“But today, if you get an answer from the AI engine, or from the AI overview, you don’t necessarily sort of always end up visiting these platforms.”

There is no “lower down the page” in AI mode. There is only presence or absence. The loss of traffic is visible.

AI creates demand-side stress for classifieds platforms by substituting information discovery. Discovery is gradually evaporating even when the classifieds platforms have done nothing wrong.

Once Google Search switches to AI mode entirely, the top of the funnel will vanish for most players. The search function that helped identify and allocate intent and demand will no longer exist. Websites will become suppliers of content for AI to repurpose as answers as AI rewires the Internet, without copyright protection.

The content that enables discovery of classifieds services will no longer bring consumers to them, because their need for answers is being satisfied upstream. This is an existential threat.

Not all classifieds are created equal

Note: if you now how classifieds work, you may skip this section

Classifieds look similar on the surface, but they monetise at very different moments of user intent, depending on category. A rough snapshot, based on user behaviour and platform monetization:

1. Seeking direction (“Help me understand what I should do”): users are learning, comparing, reading FAQs, rankings and explainers. Monetization is via display advertising, sponsored content, brand campaigns. The platform monetizes understanding. Content heavy classifieds like Shiksha operate here.

2. Seeking options (“Where should I go to act?”): Users want to be pointed towards a relevant option. Monetization is via paid inquiries, and featured listings and other enhanced visibility options. This is where listings marketplaces like IndiaMART and JustDial tend to operate.

3. Evaluating and qualifying (“Is this option credible and suitable for me?”):
Users shortlist options, compare offers, and check seriousness and fit. Sellers care about lead quality, not just volume. Monetisation moves up the stack through premium listings, qualified leads, recruiter or broker tools, and workflow software. Large classifieds like 99acres and Naukri operate strongly at this stage.

4. Executing a decision (“Help me close this”):
Users apply, transact, negotiate, finance, or commit. The platform is no longer just an intermediary; it becomes part of the process. Monetisation comes from transaction commissions, financing, insurance, and execution tools. This is where execution-heavy classifieds like CarTrade and Policybazaar operate.

5. Managing what comes after purchase (“Help me manage post-purchase issues”):
Users deal with renewals, claims, upgrades, repeat transactions, or ongoing support. Monetisation shifts to renewals, cross-sell, and long-term customer value. Platforms like Policybazaar have deliberately moved here by extending beyond comparison into servicing customers post-purchase.

Most large classifieds operate across multiple levels.

IndiaMART shows how AI can erase a marketplace

IndiaMART operates at the discovery layer (seeking options), not conversion. Just enough of its database entries is exposed, in order to enable enquiries and leads for millions of SMEs.

Traffic on website, once a KPI for IndiaMART, has now become undecipherable for the company. In its Q3-FY26 earnings conference call, when an analyst queried the company about traffic data, Founder Dinesh Agarwal pointed that the company is dropping traffic as a KPI:

“…the bot traffic is coming from everywhere. Whether it is ChatGPT bot. And there are hundreds of LLM bots now, search engine bots, new browser bots, agentic bots. So in the traffic, when trying to identify the actual human traffic versus bot traffic, it was years of Google being monopoly, that systems have become stable enough to identify what is a bot traffic and what is not. Nowadays, there is so many new crawlers are coming like this Parallel Web, Exa, those kind of things.”

What is not clear here is whether it is dropping traffic as a KPI because of just bots, or because traffic is also declining. Agarwal acknowledges that (other) publishers are facing traffic issues because of stagnation from Google, but highlights that “Where there is a transactional unique content discovery, people continue to go to search”, while chat is where people go for research, not necessarily search.

Agarwal tries to put a positive spin on the switch from search to chat by saying that whenever new technology emerges, it expands the Total Addressable Market, because, the total number of people “that will use either Google or Gemini or ChatGPT” would be higher than just Google. This optimism only holds if invocation remains neutral and comprehensive, and if AI Mode doesn’t replace search, which it is likely to.

Importantly, IndiaMART isn’t using Cloudflare to block bot traffic because it probably has to manage the transition from search to chat invocation for discovery. They’re getting invoked within Google and Grok, but ChatGPT wasn’t invoking them.

If you’re not being invoked, then for the user, you don’t exist.

That is why omission matters differently from not being placed in search discovery. It’s not about one business versus another, not about whether users are clicking on links in the invocation or not. Lack of invocation via AI chat is creating supply-side stress for IndiaMART.

This is why IndiaMART went to court to get ChatGPT to invoke them. We don’t know how invocation decisions are made, but their effects are market-shaping regardless of intent. IndiaMART has argued, as per the petition that the company shared with me, that OpenAI has:

“specifically and consciously excluded [IndiaMART] from being shown or surfaced and made available.” “ChatGPT, including its search interface, has been configured in a manner whereby when a user asks a product listing related query or asks for specific product listings, it surfaces product listings from true e-commerce marketplaces like Amazon and Flipkart, but excludes the sellers’ listings on/from [IndiaMART’s] website entirely. In fact, this exclusion is evident and continues to operate even when the user specifically requests for results only “from IndiaMART” or “on IndiaMART”

The same queries also resulted in responses with listings from TradeIndia, a long time IndiaMART competitor. OpenAI, according to IndiaMART’s petition:

expressly stated and relied upon the Office of the United States Trade Representative (USTR) Reviews, in which [IndiaMART] has been featured, to justify and defend their exclusion of [IndiaMART] from ChatGPT.

The USTR essentially listed IndiaMART in the “Review of Notorious Markets for Counterfeiting and Piracy”, which the EU’s “Counterfeit and Piracy Watch List” has not.

This makes discovery by an AI platform a geopolitical trade issue.

IndiaMART’s argument:

“Where a service is not reasonably discoverable online, users predictably gravitate to a substitute that surface more prominently through search results, recommendation systems or conversational interfaces, irrespective of intrinsic merit of the underlying service.”

“These AI-based discovery layers influence which services are surfaced to users, at what stage of the decision-making journey, and with what accompanying contextual framing, thereby acquiring increasing economic and strategic importance for both providers of internet-based services seeking growth, and for users who increasingly rely on such systems to navigate, compare and discover digital offerings in a complex online ecosystem.”

Most importantly:

“Exclusion from such new discovery channels therefore has a catastrophic effect on such a service.”

When everything moves to chat, not being invoked doesn’t just reduce traffic. In AI-driven markets, omission is extinction.

Welcome to all the new subscribers at Reasoned. If you’re a subscriber, I’d appreciate your responses to a few short questions.

Invocation optimises for answers, not markets

Invocation turns what was once a discovery problem into a market access problem.

The world allowed Google to become a search monopoly. It became a single point of failure for businesses dependent on discovery.

It is now everybody’s burning platform.

IndiaMART’s lawsuit indicates - for the first time, to my knowledge - that lack of invocation can be chosen by an AI platform. Info Edge has a market cap of almost $9 Billion while IndiaMART is about $1.4 Billion: this is a key market risk for their businesses.

An analyst on CarTrade’s Q2-FY26 earnings conference call pointed to a behavioural shift that IndiaMART also alluded to. He argued that AI platforms will simply scrape the data, and users won’t need CarTrade at all: “they will directly visit the dealers, say, Hyundai or Maruti and see the products live.”

CarTrade Founder Vinay Sanghi first claimed that their traffic has increased since ChatGPT and AI Mode have become pervasive. Sanghi’s response reframes the threat. Instead of defending discovery, he defends execution:

“Buying a new car is a very involved purchase… eventually, when you go down to buy a car, you’re going to go into deep involvement, in terms of understanding the quality of the car, what other people think about it, finding the car, what price to pay, how do you get a loan approved.”

“Today, on CarWale, we have 25 banks giving approvals for loans or getting a discount on the car or connecting to a dealer or buying it online, etcetera. We call that the journey from discovery to purchase.”

CarTrade is arguing that once a purchase reaches a certain threshold of seriousness, discovery no longer matters as much as coordination, trust, financing, and throughput.

Yesterday, Sanjeev Bichchandani, Founder of Info Edge said something similar to me when I reminded him about a conversation we had about LinkedIn vs Naukri about a decade ago. He said:

“Everyone on Naukri is looking for a job. That is why they are there. That is not true of LinkedIn. Maybe one in five or one in ten people on LinkedIn are looking for a job. This impacts recruiter productivity in favour of Naukri.”

This is fundamentally different from IndiaMART’s position. IndiaMART monetises being named at the moment of option-seeking. CarTrade and Naukri monetise what happens after intent is clear.

What invocation gets wrong, and why classifieds still have something to cling to

Invocation optimises for answers, not for markets.

It narrows choice to the point of almost choosing for you. That might feel efficient, but it changes how markets function in categories where choice, comparison and scale matter.

About a decade ago, I asked Bikhchandani about the threat of LinkedIn for Naukri, which I’m sharing with his consent because it was a private conversation. My argument then: LinkedIn is an open database, while Naukri is closed. As a recruiter, you need to pay Naukri for access. Why would anyone do that when profiles are freely available on LinkedIn?

His response was simple: You can’t use LinkedIn to hire a 150 people for IT or BFSI in a short period of time.

The problem wasn’t access to profiles. It was scale, throughput, and the ease of operating when you want many options to choose from. That distinction matters even more in the context of AI: invocation surfaces a good option, not “the market for something” when you need it. Classifieds exist to enable, expose and help navigate this mess at scale.

Invocation also struggles when it’s trying to give you the “right” answers. Context comes with its own challenges:

First, users routinely give incomplete context: there’s a huge gap between what they want and what they tell an AI model. Model fill these gaps with assumptions, and those assumptions shape outcomes.

Second, as context windows grow, models must decide what to retain and what to discard. Context selection becomes an invisible act of prioritisation, and in that process, nuance is often lost. Longer windows also trigger compression, where specificity gives way to generalisation.

Third, there is some element of what I can only describe as context pollution. Users often shift topics mid-conversation, instead of opening a fresh chat for refreshed context. Someone could be discussing automobile stocks and cars to buy in the same window, and models could mix these signals when invoking results.

Memory thus introduces a new kind of problem: not whether the system recalls, but how it forgets, reinterprets, prioritises or downgrades context over time. These are not edge cases, and they directly impact what gets surfaced and what gets excluded.

At the same time, the larger risk is still in the loss of discovery.

Once invocation becomes the dominant interface of the Internet, discovery stops being a growth lever.

Optimising content for AI summaries does not guarantee traffic, and optimising for invocation does not guarantee choice. At best, AIO/AEO/GEO makes you an option. It does not restore discovery as a market function. Classifieds businesses, like content, are the canaries in the coal mine.

How many users will you move towards the market or execution if the top of your acquisition funnel, which has long been driven by discovery, shrinks to the point of disappearance?

What happens to millions of small sellers when discovery becomes discretionary?