When the shopper is an AI agent: how Indian brands get on the shortlist

A new kind of customer in the funnel

For twenty years digital marketing has assumed a person at the other end of the screen: someone who scrolls, sees a banner, clicks, compares and buys. That assumption is starting to crack. More Indian shoppers now ask an assistant what to buy, and some are beginning to let software handle part of the purchase.

Two studies this year put numbers on it. In late September Boston Consulting Group published Future of Influence, based on a survey of nearly 5,500 consumers in China, India, Japan and South Korea across 16 product and service categories. In India, LLM search, meaning people asking tools such as ChatGPT or Gemini for product advice, featured in 22% of purchase journeys. That is the highest of the four markets and almost four times Japan’s 5%. Social media still reached more journeys, at 29%. But among people who used each channel, LLM search shaped the final brand choice in 65% of journeys against 64% for social media. India was the only market where the chatbot came out ahead.

BCG also found that more than half of LLM-assisted journeys led people to consider brands they might not otherwise have found. That cuts both ways. A smaller brand can be discovered. A well-known one can be quietly left out.

Accenture’s Consumer Pulse Survey 2026, covering 25,590 consumers in 16 countries, points the same way. In India, nearly 30% of consumers said large language models are now their first place to discover products. Fifty-nine per cent said they would tell their AI agent which brands to consider, yet 36% said they would switch their preferred brand if the agent said another was a better fit.

What an agent actually looks at

A human shopper can be won over by a film or a clever line. An agent, at least for now, mostly can’t. It builds an answer from what it can read and cross-check: product details, prices, availability, reviews, comparisons, expert write-ups, and whatever is consistent across many sources.

Ask a few assistants about almost any category and the same weak spots show up in how they describe brands:

• Missing facts. If your site doesn’t state dimensions, ingredients, warranty or compatibility in plain text, the agent fills the gap from someone else’s page, or recommends a competitor whose page is complete.
• Inconsistency. A different price or pack size on your site, on a marketplace and in an old press release looks like unreliability.
• Claims without support. “Best in class” means nothing to a system weighing evidence. “Tested to 50,000 cycles by an accredited lab” does.
• Little outside presence. Agents lean on reviews, forums, comparison articles and press coverage. A brand that only exists in its own ads gives them little to quote.

Why quick commerce feels it first

An October analysis in Storyboard18 made a pointed argument about Indian e-commerce. Quick-commerce platforms such as Blinkit and Instamart earn much of their profit from ads shown to people while they browse. Agents don’t browse. Elara Capital’s Karan Taurani estimated that a 15% fall in Blinkit’s grocery advertising would reduce parent company Eternal’s FY29 EBITDA by roughly 5%, while noting that the overall effect looks modest and uneven across platforms.

The most exposed categories are habitual ones: staples, groceries, refills. Nobody needs a film to reorder detergent, and an agent will happily pick the best-rated pack at the right price. Discovery-led categories such as beauty and fashion are less exposed, because people still want to look around.

For brands, the lesson is that paying for visibility inside a storefront may buy less when the shopper never sees the storefront. As ReBid founder Rajiv Dingra put it in the same piece, advertising will increasingly need to exist “inside the decision layer rather than only inside the interface layer”.

What brands can do now

There is no secret trick for being chosen by an AI agent, and anyone selling one deserves suspicion. Google’s own documentation says there are no additional requirements or special schema needed to appear in its AI Overviews or AI Mode, and that ordinary SEO fundamentals still apply. What works is unglamorous: be easy to understand, be consistent everywhere, and be worth recommending.

1. Put the facts on your own site, in words

Every product page should answer the questions a careful buyer asks, in text a machine can read: what it is, who it is for, sizes and variants, materials or ingredients, price, warranty, delivery and returns. Not inside an image, and not buried in a PDF. Structured data such as schema.org Product markup helps systems parse a page, as long as it matches what the page visibly says, but the content has to be there first. Much of what our Tech team builds is exactly this groundwork.

2. Say the same thing everywhere

Audit your listings across your website, marketplaces and quick-commerce apps. Same product names, same pack sizes, same key specifications, consistent pricing logic. Contradictions read as risk.

3. Earn mentions you don’t control

Genuine reviews, honest comparisons, coverage in trade press, creators who actually use the product. These are what agents quote when they explain a recommendation. PR and creator work has always built reputation; it now also builds the evidence that machines read. Our PR & Creator practice works on both.

4. Answer real questions, plainly

Publish guides that answer what buyers really ask, written by people who know the product. “Which water purifier suits borewell water in Gurugram?” is a question an agent can match to a good answer. “Pure life, pure you” is not.

5. Check what the assistants say about you

Once a month, ask three or four assistants about your category the way a customer would. Note whether you appear, how you are described and which sources are cited. It is a rough test, but it quickly shows where the facts about you are missing or wrong.

Brand still decides who gets named

It would be easy to conclude that brand building matters less when software does the shopping. We think the opposite. Accenture’s 59% describes people telling their agent which brands to consider. Someone has to be known well enough, and liked enough, to be named in that instruction.

So the work splits in two. Brand, film and experiences make people want you, so they ask for you by name. Clean product information, consistent claims and independent proof make sure that when an agent checks, it finds good reasons to agree. Neglect the first and you become a commodity compared on price. Neglect the second and you are a famous name the agent can’t verify.

Both halves can be built on purpose. If you want help working out which one your brand is weaker on, start a conversation with us.