Why ChatGPT recommends your competitor, not you (we asked the AI behind ChatGPT 100 Indian buying questions)

Every founder in India now has a new shop window they cannot see. A buyer opens ChatGPT, types a question, and gets three or four brand names back. If yours is not one of them, that buyer never knew you existed.

So we tested it. We asked the AI behind ChatGPT 100 real buying questions that Indians ask, and traced every brand it named back to the page it came from. The pattern was clear, and most of it is in your hands.

I have run growth for Indian brands since 2017, 25+ brands across D2C, coaches and creators. Search changed many times in those years. This change is the quiet kind. Nobody sends you a report when an AI leaves you out.

Key takeawayThe AI does not pick brands from thin air. It names brands it finds on other people’s pages, and on your page when it answers the question.

Below: how we ran the test, the 6 things we found, a 30 day plan, and a 10 question score for your own brand. One finding surprised us, and it will save you money. It is about the “AI files” everyone is selling. We get to it in finding 5.

Meet Riya, and the answer that left her out

Riya runs a small millet snack brand from Pune. She is a made up founder, but her week is real for a lot of people.

Her snacks sell well at local stores. Her reviews are kind. One evening her cousin Meera sends her a screenshot. Meera had asked ChatGPT which millet snack brand to buy. It named four brands. Riya’s was not one of them. Two of the four were brands Riya competes with every day.

“Is your brand even real?” Meera joked. Riya laughed. Then she sat with her phone for a long time.

She had a website, a good product and happy buyers. What did those four brands have that she did not? That is the exact question we set out to answer.

What did we find?

Every answer named brands. Across the 100 answers the model named 746 brands in total, 644 different ones, about 7 per answer. It cited 318 pages, about 3 per answer.

100 Indian buying questions, in numbers

100/100answers ran a fresh web search before naming any brand

56answers cited at least one ranking or comparison list

2xmore often, the top pick’s own site was cited (35 answers) than the last brand named (16)

56of 100 top picks changed when we asked again with search turned off

Now the part that matters for your brand. Where did those 318 pages come from?

Up next The pages the AI trusted. One kind showed up in more than half the answers, and it is not your website.

Where does the AI get its answer?

Think of the AI as a new kind of shop assistant. It has never used your product. So before it answers, it reads what others say. Here is what it read.

Every page the AI cited in 100 answers, by type

A brand or course’s own page37%

Ranking or comparison list27%

Government, regulator or university11%

Reddit and forums7%

Marketplace listing6%

News article4%

Research, social, Wikipedia, other8%

Two kinds lead: brands’ own pages and ranking lists. Then come the slow, trusted ones, like regulators for supplements and stock market courses. Reddit and forums are smaller, but they punch above their size, as you will see.

What it read depends on what you sell

D2C products 60 questions, 153 cited pages

Ranking and comparison lists39%

Brand’s own page21%

Marketplace listing11%

Reddit and forums8%

Courses and coaches 40 questions, 165 cited pages

Provider’s own course page52%

Ranking and comparison lists15%

Government, regulator or university8%

Reddit and forums4%

For D2C products, lists ruled. 39 of every 100 cited pages were ranking or comparison articles. For courses and coaches, the provider’s own course page carried most of the weight. If you sell a course, your own page is your shop window. If you sell a product, other people’s lists are.

Finding 1: It searches like a shopper

All 100 answers ran a web search before naming a single brand. That is 340 searches in total. We read every one.

34% of the searches asked for the top rated brands. 39% added the year, 2026. 21% added the word “review”. It typed what a careful buyer types, then read the top results.

So the question is not “does the AI know my brand?” It is “when someone searches my category with the year and the word review, am I on those pages?”

What the AI actually did, on one question

Answer

Which protein bar brand should I buy in India?

Searching: “… protein bars India 2026 review”

Brand A top pick

Brand B

Brand C

Brand D

Sources: 3 pages, all “top protein bars in India 2026” style lists. One was written by a rival brand on its own site.

Every pick traced back to a page

1. It ran 3 searches with the words a shopper types, plus the year.
2. It cited 3 pages. All three were comparison lists. No press release, no ad.
3. It named 4 brands, the ones those lists ranked. Names hidden here on purpose, they are in the CSV.

Finding 2: Ranking lists decide who gets named

56 of 100 answers cited at least one ranking or comparison list. For product questions, it was 37 of 60.

We opened every cited list we could read, 79 of them. The AI’s top pick was named on 49, which is 62%. In plain words: the AI mostly repeats the lists.

Here is the sharp part. In 22 answers, one cited page was a comparison list written by a brand, on its own website. A protein brand’s own “top protein bars in India” page got cited, right next to the review sites. A fair comparison page is not a trick. It is useful, so it gets read.

Up next Your own website. It matters more than you think, but one kind of page matters most.

Finding 3: Your own page counts, when it answers the question

The top pick’s own website was cited in 35 of 100 answers. The last brand named in each answer had its own site cited in 16 of 99. The winner’s page was read about twice as often.

What were those pages? Mostly product and course pages that answered the question in plain words: who it is for, what it does, the price. Not home pages full of banners.

For courses and coaches this was even stronger. More than half of all their cited pages were the provider’s own course page.

The Shortlist Loop: how an AI picks who to name

1. 1A buyer asks“Which brand should I buy?”
2. 2It searches like a shopperCategory, India, 2026, review
3. 3It reads about 3 pagesRanking lists, Reddit threads, brand pages that answer the questionYou can change this step
4. 4It names brands seen in more than one placeBuyers try them, talk about them, and those words feed step 3 next time

And the loop starts again with the next buyer

Finding 4: Reddit was the most cited single website

No brand site, no news site and no marketplace was cited as often as Reddit. It was cited 18 times, in 15 answers, for questions from sunscreen to UPSC coaching.

Why? A real buyer saying “I used this for six months, here is what happened” is gold for an answer. You cannot buy that. You can earn it, by being a brand people talk about, and by showing up in those threads as yourself.

We also checked every readable third party page cited in each answer. The top pick was named on at least one of them in 66% of answers. The last brand named, in 52%. Being talked about in more places moves you up the list.

Finding 5: The “AI files” did not separate winners

This is the one that will save you money. You may have been sold llms.txt files, special schema or a Wikipedia page as the key to AI answers. We checked the public homepages of the top picks and of the last named brands.

The gaps are small, or go the wrong way. Many of those llms.txt files were made by the store platform, not the brand. They describe how to shop the store. They do not make the brand more trusted.

Google says the same thing for its own AI answers. Its guide for site owners says you do not need new machine readable files, AI text files or special schema to appear in AI Overviews or AI Mode. Google Search Central, AI features and your website (VERIFIED, read 23 Oct 2026).

Keep your schema tidy, it helps search in general. Just do not expect a file to do the job of a good reputation.

Up next The good news. The answer is not set in stone.

Finding 6: The answer moves, so you can move it

We asked all 100 questions again with web search off. The top pick changed in 56 of 100. About half of the brands stayed the same.

That tells you two things. The AI’s memory alone has a fixed list of famous names. The live web reshapes that list every time someone asks. Lists get updated. Threads get new replies. Your page gets rewritten. Each of those can change who gets named next week.

The big winners in our data were names you would guess. Minimalist was named in 7 answers. The Whole Truth, Coursera and Udemy in 5 each. When we asked about a Meta ads course, the top pick was Meta’s own free program. Big brands start ahead. But with 644 different brands named, there is plenty of room for a smaller brand to get in.

If you also want to pay for a spot next to the answer, read our ChatGPT Ads in India guide, with real costs and a Meta comparison. Ads put you beside the answer. This article is about getting into the answer itself.

What happened to Riya?

Riya stopped guessing. She asked ChatGPT her five buying questions and wrote down every page it cited. Three were “top millet snacks” lists. One was a Reddit thread. One was a rival’s product page.

She rewrote her main product page to answer the question in two lines. She made a fair comparison page, including when a buyer should pick someone else. She wrote to the three list writers with samples and plain facts. She answered two Reddit questions as herself, with no links.

Six weeks later Meera asked the same question again. This time, Riya’s brand was the fourth name in the answer. Not the top pick. But in the room, at last.

Riya and her results are made up, to show the method. They are not from a real account, and the same steps can give different results for you.

Your 30 day GEO plan

GEO means generative engine optimisation: making sure AI answers can find you, trust you and name you. Here is the whole plan in 8 steps, based on what we saw.

1. Week 1Ask and write downAsk ChatGPT your 5 biggest buying questions. Note every brand it names and every page it cites. That list is your map.
2. Week 1Fix your answer pagePick your top question. Make one page that answers it in the opening two lines, with price and who it is for as plain text.
3. Week 2Build a fair comparisonPublish your product vs 3 common options. Facts in a table. Say who should pick someone else.
4. Week 2Get on the listsFrom your map, pick the 10 ranking articles that rank your category. Send each writer facts, photos and a sample.
5. Week 3Show up in threadsAnswer real questions on Reddit India and niche forums as yourself. Help comes before any link, no link drops, no fake reviews.
6. Week 3Turn WhatsApp into a FAQCopy the 10 questions buyers ask most into a public FAQ, answered in two lines each.
7. Week 4Collect reviewsAsk every buyer for an honest review 7 days after delivery, with one link.
8. Week 4Ask againRe run your 5 questions. Compare who gets named and which pages get cited. Repeat monthly.

Your AI visibility score

Answer 10 quick questions about your brand. Each answer shows a tip. Your score, out of 100, shows which three moves to make next.

AI visibility score

10 questions. Be strict: “sometimes” is a Partly.

01A page on your site answers the exact question a buyer types, in its opening two lines.

YesPartlyNo

02You are named in 3 or more ranking or comparison articles for your category in India.

YesPartlyNo

03Real customers talk about you on Reddit or forums, in their own words.

YesPartlyNo

04You have an honest comparison page: your product vs other options, with when to pick someone else.

YesPartlyNo

05Your product or course pages show price, who it is for and what is included as plain text, not inside images.

YesPartlyNo

06You have 50 or more public reviews on a marketplace or your Google profile.

YesPartlyNo

07A news site, blog or creator has written about you in the last 12 months.

YesPartlyNo

08Your FAQ page answers the 10 questions buyers ask you on WhatsApp, in public.

YesPartlyNo

09Once a month, you ask ChatGPT your top 5 buying questions and write down who it names.

YesPartlyNo

10Your pages open fast and read fully without a login, pop up or wall.

YesPartlyNo

0/100

Answer the questions above

Your score and your next three moves show here. Nothing leaves your phone.

The numbers in one place

Download the data

Use it, quote it, build on it. Please link back to this page if you do.

• All 100 questions and answers (CSV)Question, category, top pick, every brand named, cited domains, the searches the AI ran, and the answer with search off.
• All 318 cited pages (CSV)Page title, link, website and page type for every citation.

What should you remember?

• The AI searched the web in all 100 answers. It reads before it recommends.
• Ranking lists, Reddit threads and your own answer page are where the picks come from.
• The top pick’s own page was cited twice as often as the last brand’s. Answer the question on your page.
• llms.txt and special schema did not separate winners. Reputation did.
• The answer changes. Ask your 5 questions every month and track who gets named.

Sources (checked 23 October 2026)

• Our data: 100 questions asked to the AI behind ChatGPT, through its developer API, with web search on, location India, 23 Oct 2026, 3:09 PM to 3:12 PM IST, then again with search off, 3:12 PM to 3:14 PM IST. Brands and page types were sorted by a second model and spot checked by hand. Both CSV files are above.
• Google Search Central: AI features and your website (no additional requirements, no AI text files or special schema needed for AI Overviews or AI Mode)
• Homepage checks: we read each brand’s public homepage and its /llms.txt on 23 Oct 2026. Some sites blocked our reader, so the totals in the table are the pages we could read.