Has AI search changed how Indians buy, and what is actually measured about it

The honest answer for India splits in two, and the split is the useful part. On the supply side, meaning what an assistant says back when somebody asks an Indian buying question, there is measured change and it has now been read three times on three different question banks. On the demand side, meaning what share of Indian buyers ask an assistant before they contact a supplier, the number that matters is the one for your own category, and it takes about an afternoon to collect. Anybody quoting a single national percentage for Indian AI buying behaviour is quoting a target, not a reading.

This page sets out what is measured, with each dataset's own date and size, and then gives the forty reading method that produces your own number.

What the supply side already shows, three readings deep

Three runs, three question banks, three dates. They are separate datasets and the counts below are never added together or averaged, because they asked different questions.

Every one of those runs used ChatGPT and Gemini, two engines rather than five, with Perplexity, Claude and Google AI Overviews asked nothing. The three agree on the mechanism. When an assistant answers a commercial question, the great majority of what it reads is a page somebody published on their own domain to sell something. Review directories took 0.8 per cent, 0.7 per cent and 0.7 per cent of the sources across the three runs, which is the same answer three times.

The India finding that changes what a seller should do first

In the 50 blind readings of the India audit, with the country named in the wording of every question because the search provider accepts a country parameter and answers from elsewhere, the answers split cleanly into two groups.

• In 19 of the 50, a business was named in the answer text and it was not us.
• In 31 of the 50, no business was named at all. The answer explained the subject, listed the considerations and recommended nobody.

That second number is the opening. On three fifths of these India buying questions the engines currently have no firm to put forward, which is a different competitive position from a market where an incumbent already owns the answer. Where a name did appear, the ones that appeared most were Indian agencies, in a set of readings where no question named any firm at all.

What is not measured, and the number to stop looking for

There is no verified Indian figure for the share of buyers who consult an assistant before contacting a supplier, and a percentage of that shape is the most quoted and least sourced claim in this category. The adoption numbers that do exist and are properly sourced come from American survey panels, and our page on what the measured data shows about trust in AI recommendations sets those out with the panel, the sample size and the publication date attached. Those are United States figures and they describe American adults. Read them as the shape of the trend rather than as an Indian number.

The forty reading method, which gives you your own number

This is the measurement that answers the question for your business, and it is cheaper than reading another study.

• Write twenty questions your buyers actually ask before they buy. Ten that name a place and your service, ten that name a problem, a price or a comparison.
• Name no company anywhere in the wording, including your own. A question that names a brand primes the answer and the result is worthless.
• Fix the market in the question text. Write India, or your city, into the wording. Setting a country parameter on a tool is not the same thing and does not do it.
• Ask all twenty on ChatGPT and twenty on Gemini, which is forty readings, and save the full answer and every source.
• Record three things per reading. Was any business named. Was it you. Which websites were cited.

Forty readings is one afternoon and it produces three numbers nobody can give you from a national average, the share of your own buying questions that name any firm, the share that name you, and the list of sites your category's answers are built from.

What to change once you have the numbers

• If most of your questions named nobody, the work is to publish the page that answers one of them properly, with the rupee figure, the date and the method. That is the gap.
• If a competitor was named repeatedly, read the pages the answer cited and look at what those pages carry that yours does not. In almost every case it is a specific number with a date attached.
• If your category's citations were dominated by platform documentation, the buying question behind them was really a how does this work question, and a vendor page was never going to win it. Pick a different question.
• Repeat the same forty questions in sixty days, unchanged, so the second reading is a comparison rather than a new study.

Questions about AI search and Indian buying behaviour

Is there a reliable figure for how many Indian buyers use ChatGPT before buying

Not one that stands up with a named sample and a date, which is why this page gives you the method for your own category instead. A number collected on your own thirty questions is worth more to your decision than any national share, because it is about your buyers.

Why does the India audit show 91.1 per cent when the earlier run showed 85.6 per cent

Different question banks, different dates and different markets. The India audit asked 198 India specific questions on 22 October 2026; the 85.6 per cent came from 597 readings across three markets on 15 October 2026. Both are measurements of the same mechanism on different samples, and the three runs agreeing within that range is the finding rather than the gap between them.

Do these readings cover Google AI Overviews

No. All three runs covered ChatGPT and Gemini. Google AI Overviews, Perplexity and Claude were not asked anything in any of them.

How often should an Indian business retake this reading

Every sixty days on the same frozen question set, so the comparison is clean. Changing the questions and the date in the same pass produces two variables and no finding.

Does any of this apply to a purely local business

Yes, and the local case is the sharper one, because a question that names a city and a service is exactly the shape an assistant answers with a short list of firms. Write ten of those, with your city in the wording, and the answer will tell you whether you are on the list.

The reading that is not in anybody's analytics

What makes this question hard to answer from your own data is that an assistant's answer leaves no trace in it. A buyer who reads a short list of three firms and calls one of them arrives as a direct visit or a phone call, and the sentence that put the three names in front of them is not recorded anywhere you can see. Our page on why Google Analytics has no row for a citation takes that apart. AI Knows Us exists to take the reading that fills the gap, by asking the buying questions directly and recording what comes back, and the three market benchmark publishes the whole method so the forty reading version above is something you can run without us.