Twenty Indian brands.Six hundred AI answers.Who gets named.

For each brand we wrote fifteen questions a buyer in its category asks, most of them without any brand name in the question and a few with the brand's own name, such as is it good or what does it cost. Each question was sent once to ChatGPT and once to Perplexity through their search-grounded APIs on 18 August 2026, from India, with the full answer text and every cited URL saved.

A mention is the brand's name, or a known alias, appearing as a whole word in the answer text. Share is answers naming the brand divided by answers, thirty per brand. The same match was run for each brand's tracked competitors, so an answer that named none of them is counted as naming no brand.

Two engines were measured. Gemini and Google AI Overviews were not part of this sweep and are not estimated here; our separate payroll and HRMS study covers all four engines for one category, 40 questions and 159 answers, by hand in the consumer apps.

Answers naming the brand, out of thirty, then out of fifteen on each engine, with the competitor named most often in the same answers.

6 of the 20 brands were named at least four answers more often on one engine than the other, out of fifteen. HROne was named in 13 Perplexity answers and 6 ChatGPT answers. Keka 14 on ChatGPT against 9 on Perplexity; BlissClub 12 on ChatGPT against 7 on Perplexity; Nutrabay 3 on ChatGPT against 8 on Perplexity; StayVista 14 on ChatGPT against 10 on Perplexity; Vyapar 10 on ChatGPT against 14 on Perplexity.

The direction is not fixed. In HRMS, ChatGPT cited vendor pages and Perplexity cited listicles and LinkedIn posts, so a vendor with a strong site and a thin third-party record did better on ChatGPT. Elsewhere it ran the other way. The strong engine is the one that reads the page type the brand's ecosystem publishes.

Vaaree's own domain was the second most cited source in its category's answers, ahead of ikea.com, and the brand was named in 16 of 30 answers while Pepperfry was named in 23. Supertails's site was the most cited source in pet care and the brand was absent from 10 answers, several of which cited the site and recommended Royal Canin.

Crawlability is necessary. It is not what puts a brand in the shortlist. The engines rank by how often a name recurs across everything they read, and a brand's own site is one source among many.

In 5 categories a large share of answers named no tracked brand at all: men's health 17 of 30, menopause nutrition 17 of 30, real estate brokerage 14 of 30, insurance advisory 8 of 30, study abroad counselling 7 of 30. These are answers with an empty shortlist, where the engine described the category and recommended nobody.

An empty shortlist is the cheapest position to take. Whichever brand becomes quotable first on the pages these engines read will hold it.

Zimyo was named in 10 of 30 answers. All of those came from questions that contained its name. On the questions that did not, the ones a buyer asks before they have a shortlist, it was named in 2 of 22. Branded and unbranded questions have to be scored apart, and the unbranded share is the one that describes discovery.

In study-abroad counselling, idp.com was the single most cited domain across Leverage Edu's fifteen questions, and IDP was named in 21 answers to Leverage Edu's 7. In HRMS, one vendor's blog was cited in nearly half of the pooled answers to three competitors' buyer questions. When a rival's content is what the engine reads to answer your category, its name comes with it.

• Best free billing software went to Zoho Invoice on the strength of one sentence on its site, 100% free for Indian businesses forever, while a competitor's real free tier went unnamed.
• A restaurant POS vendor with no published rupee price had its price quoted from a US dollar listing on G2 by one engine and from a third-party estimate by the other, and lost the cheapest-POS question to an unknown vendor with a published price of one hundred rupees a month.
• A SaaS brand rated above four on G2 from more than a hundred reviews had a thirteen-review Trustpilot profile at 1.9 out of 5 quoted into the answer to whether it is any good, on both engines.
• A premium skincare brand was named in every answer to questions phrased in its own vocabulary and in 3 of 26 answers to questions phrased the way buyers phrase them.

• Two engines. Gemini and Google AI Overviews were not measured in this sweep.
• One run per question per engine. Answers vary between runs; the shares are a reading, not a constant.
• API answers. The consumer apps add personalisation and memory that the API does not, so what a given buyer sees can differ.
• Fifteen questions per brand, written by us. A different question set would move individual numbers, though we have not seen it move the pattern.
• 20 brands chosen because we were looking at them, not sampled from a population. The median is the median of this set.