In AI answers, a few names take the whole shortlist
India is now ChatGPT's second-largest market, with 100 million people using it every week. We asked ChatGPT, Google's AI Overviews and Claude 1,096 buying questions for 93 brands and read nearly 2,500 answers. Four findings decide who gets recommended.
What it means for a brand
- Being known is not being recommended. The engines know the brand, then leave it out when a buyer asks what to buy.
- AI writes from other people's pages. Listings, directories and list articles shape the answer as much as your own site does.
- Each engine is its own channel. In India, ChatGPT named the brand in 44% of buying answers and Google's AI Overviews in 23%, so winning one says little about the other.
Nearly 2,500 answers · 1,096 buying questions · 93 brands · 3,000+ companies named · April to September 2026
The evidence behind each finding
Known by name, missing from the list
Ask one of these tools about a company and it knows the company perfectly well. Ask it what to buy in that company's category and the company usually is not there. The same brands, the same scans, sorted by what the question asked for.
1,769 of the 1,847 answers where our check recorded whether the brand appeared; the rest were one-off question types. A question about the brand by name is answered with the brand in it 97 times in 100. A question about what to buy in its category, 36.
The engine is not ignorant of you. It just does not reach for you when a buyer is choosing.
The same few names fill every answer
Across 130 scans with enough answers to count, a scan's answers name 30 different companies in the middle case, and the three most-named each turn up in about 60 percent of them. Everyone else shows up once or twice, if at all. A single week of category questions shows the same shape close up: five buying questions in each of 11 categories, 9 to 32 companies named per category, and in 6 of the 11 the same three companies in all five answers.
Five questions per category, each asked once on ChatGPT on 19 September 2026. Products from the same company are counted once. The POS category mixes shop and restaurant questions, so no company fits all five; one billing question was read as software subscriptions in general, which widened that list.
A close-up: health supplements
Two of our scans asked about supplements sold in India, in September and October 2026: 14 buying questions, from shilajit and omega 3 to biotin, collagen and gut health, put to ChatGPT and Google's AI Overviews. The 37 answers show the same shape, more sharply.
Brands are counted once per answer, with a brand's product lines folded into the brand. Sources were kept for 36 of the 37 answers.
In supplements, half the answers lean on a marketplace or pharmacy page, and a handful of names fill the shortlist.
A narrower question brings a different list
Buyers rarely ask the broad question. They add a city, a use case or a price, or they ask in Hinglish. We asked those versions too.
Three customer engagement platforms appeared in 9 of 10 answers. All three were missing from the tenth, which asked for a platform for banks and insurers in one city. It named eight other companies instead, six of them described by an office or a page in that city.
A grocery wholesale platform was named in five of six answers, but not in the one that asked about its own home city. That answer named seven other suppliers, most of them local.
We expected Hinglish to be worse. It is not. Across 52 scans that asked in both English and Hindi or Hinglish, the brand was named in 47 percent of the English answers and 48 percent of the Hinglish ones. One billing app was named in two English answers and missing from both Hinglish ones, but that is one case, not a pattern.
Being named for the broad question does not carry over to the questions buyers actually type.
The tool the buyer picks changes the answer
The same brand, the same questions, asked in India. How often it was named depends on which tool the buyer used, and the tools do not read the same pages.
ChatGPT named the brand in 44 percent of 441 answers. Google's AI Overviews named it in 23 percent of 278. Claude named it in 57 percent, though only 37 of our Indian answers came from Claude. The Google surface, the one sitting on top of ordinary search results, is where a brand is hardest to find.
Google's AI Overviews used a marketplace, directory or listicle in 78 percent of its 421 answers, against 38 percent of ChatGPT's 1,125. Claude leant on review sites in 37 percent of its 92, where the other two were under 5 percent.
Winning in one tool tells you nothing about the others. Each one reads different pages.
The engine reads other people's pages about you
AI answers are written from the pages the engine reads first. Those pages were far more often a rival's than the brand's own.
1,638 answers that came with their sources, from 89 brands
A marketplace, directory or listicle was among the sources in half of all answers, so the page a buyer is sent to is often not anybody's own site. Counting each brand once instead of each answer, the middle brand had its own site cited in 20 percent of its answers. For 18 of the 89 brands, their own website was never cited once.
Even a company's own writing can work against it. One company's blog post comparing shipping platforms was the source an answer cited to recommend a competitor, and the company itself was left out of that answer.
Small site problems hide a company completely
Some gaps have nothing to do with the market. They sit on the company's own website.
One software company's entire website asks search engines not to index it, while an older copy of the site stays open. When ChatGPT described the company's case studies, it read the old copy.
A company that renamed itself in 2025 is named only by its old name. Its own article on the category came up in the search, and the answer still did not use the new name.
One answer quotes a platform's case study with a bank. The answer for banks and insurers in a city leaves that platform out.
What a place on that list is worth
Our scans show who gets named. They do not show what it earns. Two larger studies published this year put a price on it. Neither is about India alone, which is the gap our own reading fills.
more often, in March 2026, a visit that came from an AI tool ended in a purchase, against visits from other channels such as paid search and email. A year earlier those visits bought 38 percent less often. They also lasted 48 percent longer. Adobe scored the sector's pages too: a quarter of an average retail homepage cannot be read by AI models.
Adobe Digital Insights, April 2026, from more than a trillion visits to United States retail sites. Read itgrowth in shopping use of these tools in nine months, and more than 60 percent of people now say they trust what the tools tell them. India is one of the nine countries in the survey, and the report's advice to brands is to move before the habit hardens.
BCG, January 2026, more than 9,000 consumers in nine countries. Read itIf you sell to businesses rather than shoppers, the number to know is 6sense's: 95 percent of won deals went to a company that was already on the buyer's first shortlist, written down before the buyer spoke to anyone in sales, and 94 percent of those buyers used AI chat tools while buying (6sense, about 4,000 buyers).
Buyers are already there, the first list decides the deal, and the visits that arrive from AI are now the ones most likely to buy.
What to do about it
- Write down the questions your buyers ask. The broad one, and the versions with a city, a use case, a price and Hinglish.
- Check each one. Are you named, in what place, and which pages did the engine read to decide?
- Fix what it reads. Make your own pages say plainly who you serve, where and at what price. Make sure search engines can index them under your current name. Then get onto the comparison and review pages the engine already trusts.
Where AnswerTrace comes in
AnswerTrace runs this check on your buyers' questions, across the AI engines, and shows which pages to change.
Your SEO or content team runs it: track the questions, see who gets named, and fix the pages.
No credit card required.
How we measured
- Every answer stored from a scan between 24 April and 18 September 2026: 1,096 different questions, put to ChatGPT, Google's AI Overviews and Claude, 2,485 answers in all, from 173 scans of 93 brands. Demo brands, our own two brands and plan test runs are left out, and so are the 42 rows where the engine gave no answer. 3,061 different company names appear across those answers.
- "Named" is the check each scan records for the company it is scanning. It was recorded for 1,847 of the 2,485 answers, and every share of answers named is out of those.
- Sources are the pages an engine listed with its answer, kept for 1,638 answers from 89 brands. "The brand's own website" means the source domain is the scanned company's own. "A rival's website" means the domain matches another company named in the same answer; marketplaces, directories and review sites are not counted as rivals.
- 756 answers carry an India market tag, from 54 scans of 40 brands. Older scans carry no tag. Where a number is India only, it says so.
- Separately, on 19 September 2026: 65 ChatGPT answers, with web search on and the question asked as a buyer in India. Five questions in each of 11 categories, plus ten narrower questions (a city, a use case, Hinglish) in two of them. These were read by hand, and "named" was checked against the answer's own text.
- Supplements close-up: two scans, 13 September and 5 October 2026, 14 questions, 37 answers (ChatGPT 20, Google's AI Overviews 17), asked as a buyer in India. Brands come from each answer's stored list of names, with a brand's product lines counted once; marketplaces and online pharmacies are matched by their web addresses among each answer's sources.
- No company is named in this study.
- The figures under "What a place on that list is worth" are other people's studies, each linked where it appears. They cover the United States and global samples, not India, and we have not re-checked their data.
- The ChatGPT usage figure is OpenAI's, from February 2026, as reported by TechRepublic.
AnswerTrace · September 2026