How to Get Your Brand Cited in AI Answers

Originally published on depra.ai on 14 Aug 2026, updated 15 Sept 2026.

What decides which websites AI engines cite, what the public research says works, and where 4,640 citations in Indian shopping answers pointed.

To get cited by AI, get your brand named on the pages ChatGPT, Gemini, Perplexity and Google AI Overviews already use as sources for your category: best-of lists, marketplace listings, review videos and forum threads. Then make your own pages easy to quote, and check each engine's sources every week, because they change fast.

A citation is a link an AI engine shows as a source for its answer. Google AI Overviews are the AI-written summaries that appear above the normal results on some Google searches. A prompt is one question typed into an AI engine, and tracking tools count the questions they monitor as prompts. This guide uses public research we opened and checked on 15 September 2026, plus 4,640 citations from our own study of 480 AI shopping answers asked from India.

What makes AI cite a website?

An AI engine with web search runs searches behind your question, reads a small set of results and writes its answer from the text inside them. The pages it links are the pages it drew from. So a page gets cited when two things are true: it shows up in those searches, and it contains a clear answer to the question.

Take a buying question such as "best face wash for oily skin under ₹500". The pages that answer it are lists of brands, product listings with prices, review videos and threads where buyers compare options. A brand homepage rarely answers it. That is why most of the work to get cited happens on other people's websites.

If AI never names your brand at all, start with why ChatGPT doesn't mention your brand before you spend on outreach.

Do backlinks or brand mentions matter more for AI citations?

Brand mentions. Ahrefs studied 75,000 brands and found that brand web mentions correlated with visibility in Google AI Overviews at 0.664, against 0.218 for backlinks (published May 2025). Ahrefs used a Spearman correlation: a score where 1 means brands that rank higher on one measure always rank higher on the other, and 0 means no relationship. Mentions sat much closer to 1 than links did. A mention counts without a link.

Two limits apply. The study covered Google AI Overviews only. And a correlation does not prove cause, since brands with many mentions also tend to be bigger brands. Even so, the result points toward PR, reviews and community presence over link building.

A mention is any page that names your brand in text: a review, a news story, a best-of list, a marketplace listing or a Reddit comment. Keep a list of them the way you would keep a list of links: which websites, how recent, and whether each one describes your product, price and category correctly.

Does each AI engine cite different sources?

Yes, by a wide margin. Kevin Indig took a random sample of 20,000 prompts from Omnia, a platform that tracks AI search answers, and found that only 2.37% of cited URLs appeared in all three of ChatGPT, Perplexity and Google AI Overviews for the same prompt. 91.07% appeared in exactly one engine. The sample ran mostly in European markets.

Our India data shows the same split. The three engines in our study cited 600 distinct pages. 6 of them (1.0%) were cited by all three engines, and 539 (89.8%) by only one. At website level, 20 of 247 websites (8.1%) were cited by all three.

The engines also link sources at very different rates:

Engine	Answers	Citations	Per answer	Most cited website
ChatGPT with web search	160	408	2.55	beminimalist.co (29)
Gemini	160	1,164	7.28	indisangam.com (99)
Perplexity	160	3,068	19.18	youtube.com (518)

Gemini attached no source at all to 55 of its 160 answers. Most of those were Hinglish answers: Gemini cited sources in 90.0% of English answers and 41.3% of Hinglish answers. A plan built on one engine's sources will miss most of the others, so decide which engines your buyers use and work through each one's list.

Where does AI get its sources in India?

In our study, YouTube, Reddit and Nykaa were the three most cited websites. We asked 10 buying questions (5 skincare, 5 fashion) in English and Hinglish, 8 times each, to ChatGPT, Gemini and Perplexity from inside India on 14 August 2026. Hinglish is Hindi and English typed in English letters, the way many Indian buyers write. The 480 answers carried 4,640 citations:

• youtube.com: 534 citations, 518 of them from Perplexity. ChatGPT linked YouTube twice.
• reddit.com: 191 citations, 82 of them to r/IndianSkincareAddicts. Reddit appeared in 106 of the 480 answers, more than any other website.
• nykaa.com: 176 citations, and one of the few websites all three engines cited often: 14 times on ChatGPT, 83 on Gemini and 79 on Perplexity.

ChatGPT leans on different websites. India is ChatGPT's second-largest user base after the United States, with 100 million weekly users as of February 2026. Its most cited websites in our study were beminimalist.co, hindustantimes.com and ndtvshopping.com: a brand's own site and two news publishers. YouTube and Reddit barely register on ChatGPT: 2 and 3 of its 408 citations.

Three more findings decide where to start.

Your own website can be cited on every engine. Of the 6 pages all three engines cited, 3 sit on a brand's own website: a Cetaphil article on top body moisturisers, a Deconstruct article on sunscreens without a white cast, and Levi's men's jeans page.

List pages carry the most weight. About 4 in 10 citations went to addresses with "best" or "top" in them, and Gemini leaned on them most. The full top 20 websites, split by engine and by language, with the page types behind each count, are in which websites AI cites for shopping in India.

Hinglish questions pull different sources on Perplexity and Gemini. Overlap here means the websites both answers cited, as a share of every website either answer cited. On Perplexity, two runs of the same question in the same language overlapped by 85.2%, and an English run against a Hinglish run by 52.8%. On Gemini the figures were 33.1% and 10.8%. ChatGPT changed little. If Perplexity or Gemini matter to your buyers, keep your Hinglish questions as a separate list and compare its sources with your English list.

How we counted, and the limits. We read the stored source links for all 480 answers on 15 September 2026. We counted a website by its domain name, such as nykaa.com, and a page by its web address with everything after the "?" removed. YouTube links keep their video ID, so each video counts once. The cross-engine page overlap pools all 10 questions, so it is looser than a same-question comparison. The data covers 2 categories, 1 country, 1 day and 3 engines, with no Google AI Overviews and no signed-in account. Citations show what an engine linked, and they do not prove why a brand was named. Depra ran the study and sells AI visibility tracking. You can recheck every count from the raw answers in the IndicGEO repository on GitHub.

What on-page changes help AI cite your pages?

On-page work is cheap, so do it once. The GEO paper by Aggarwal et al. found that adding citations, quotations and statistics raised a page's visibility in AI answers by up to about 40% in a GPT-3.5 test setup (KDD 2024).

Later research is more cautious. C-SEO Bench (NeurIPS 2025) tested published conversational SEO methods, which rewrite a page to make it more persuasive to an AI model. Most were largely ineffective and often lowered the page's rank inside the answer. Traditional SEO, which gets the page into the results the model reads, worked significantly better.

Put the two findings together: get the page into search results first, then make it easy to quote. In practice:

• Answer the buying question in the first 2 sentences under a question-shaped heading.
• State your category, price band and who the product is for in plain text.
• Put specs, prices and ingredients in text or tables, as well as in images.
• Give numbers with dates and link your sources.
• Publish an honest comparison or best-of page for your category that lists rivals next to your product. Our data shows engines cite these pages on brand websites.

Here is the difference on a product page. A line such as "our best-selling serum, loved by thousands" gives an engine nothing to quote. A line such as "vitamin C serum, 30 ml, ₹599, for oily skin, 10% vitamin C" answers a size, a price, a skin type and an ingredient question at once. The second line is the one that matches a buyer's question.

What is not worth your time?

llms.txt is the clearest case. An llms.txt file is a plain-text guide to your site written for AI tools. Ahrefs checked 137,210 domains and found that 97% of llms.txt files received zero requests in May 2026. AI retrieval bots made 1.1% of the requests that did arrive. Publishing one is harmless, but it does not count as progress. We looked at the evidence in more detail in does llms.txt work.

The second waste is trusting one screenshot. SparkToro had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI tools 2,961 times in November and December 2025. The chance of two answers giving the same brand list was under 1 in 100. The same list in the same order came up about once in 1,000 runs. One answer that names you is a sample of one. We cover the reasons in does ChatGPT give the same answer to everyone.

How long does an AI citation last?

In ChatGPT, often less than a week. SISTRIX tracked 82,619 prompts over 17 weeks, from 17 December 2025 to 8 April 2026 in six countries, and found that ChatGPT Search replaced 74% of its cited domains each week. In Google AI Mode, the chat-style search mode inside Google, the figure was 56%.

A one-off PR push therefore gives a short spike. What holds up is a steady flow of new mentions on the websites each engine keeps returning to, checked every week.

How to get cited by AI: a 7-step plan

1. Freeze a question set. Write 20 to 40 buying questions your customers ask, none containing your brand name. If you sell in India, add Hinglish versions. Keep the set fixed so before-and-after comparisons hold.
2. Measure a baseline per engine. Record how often each engine names you and which pages it cites. Run every question more than once, because the same question gets a different answer each time.
3. List the websites each engine cites for your category. Rank them by how often they appear.
4. Find the gaps. Mark the websites and pages that cite your competitors and never mention you. That is your outreach list.
5. Get into the list pages. Send the authors of cited best-of articles samples, prices and facts they can check, and say which list you want to appear in. Start with lists that more than one engine cites.
6. Work the marketplaces, videos and threads. Grow recent reviews on Nykaa, Amazon and Flipkart, get honest YouTube reviews, and answer real Reddit threads with your affiliation disclosed.
7. Do the on-page work once, then keep measuring. Check each engine's sources every week and compare your visibility month to month. Act on moves larger than normal rerun noise.

For the recommendation side of the same work, read how to get ChatGPT to recommend your brand.

How to see which sites AI cites for your category

You can do it by hand:

1. Ask each engine your questions from a logged-out browser, so your own chat history does not shape the answers.
2. Run each question at least 3 times.
3. Copy every source link into a spreadsheet, with the engine, the question and the date.
4. Cut each link down to its domain name and count how often each domain appears.
5. Mark which domains name your competitors and which name you.

For 20 questions on 4 engines with 3 runs each, that is 240 answers to read, and SISTRIX's figures mean you would repeat it every week.

Depra's citation gap analysis does the counting. It lists the websites and pages AI cites when it recommends your competitors but never when it names you, at domain, subdomain and page level, with competitors' own websites filtered out. Each cited page carries a type label such as listicle, comparison, how-to or product page, so you can see which formats win in your category. If a competitor keeps getting named and you do not, the guide on why ChatGPT doesn't mention your brand has test prompts to find the cause.

How do you track AI citations?

Citation tracking means asking the same questions on a schedule and recording which websites each answer links. Depra asks your buyers' questions to ChatGPT with web search, Gemini and Google AI Overviews every day and to Perplexity every week, from inside your chosen country, in English and Hinglish. It stores every full answer with its sources, and you can export them as CSV.

On top of that record:

• Each cited website carries a type label: editorial, company site, forum, reference, marketplace, directory, government or your own site.
• A daily "what changed" page lists new and lost cited sources since the previous scan day.
• The AI action plan ranks up to 12 places to get mentioned and up to 12 page formats to publish. After you mark an action done, it shows your visibility before and after. That view is a comparison over time, and it does not prove cause.
• Depra suggests 5 topics with 8 buying questions each, none naming your brand, and it does not count questions that already contain your name.
• Your report shows a confidence range, so you can tell a real move from rerun noise.

Starter costs ₹1,999 a month plus GST and includes 20 prompts, meaning 20 tracked questions. See how Depra stacks up against other trackers on the AI visibility tool comparison page.

Start 7 days free on any plan, no card. Your first scan starts at signup.

Frequently asked questions

How do I get cited by ChatGPT?

Get your brand named on the pages ChatGPT with web search pulls for your category questions. In our 160 ChatGPT answers asked from India on 14 Aug 2026, ChatGPT showed 2.55 source links per answer on average, and its most cited websites were beminimalist.co, hindustantimes.com and ndtvshopping.com. With that few links per answer, find the pages ChatGPT cites for your own buying questions and work on those first.

Which pages should I target to get cited instead of a competitor?

Start with the pages that already cite your competitor for your buying questions. In our India study, 39.5% of ChatGPT's citations pointed at URLs with best or top in the address, so best-of lists usually come first, then marketplace listings and brand blogs. Pitch the authors of those lists, or publish an honest comparison page on your own site that names your rivals.

Is llms.txt worth publishing to get cited by AI?

There is no public evidence that it helps. Ahrefs checked 137,210 domains and found that 97% of llms.txt files received zero requests in May 2026, and AI retrieval bots made 1.1% of the requests that did arrive. Publishing one is harmless. Spend the real effort on mentions in the pages AI engines already cite.

How long do AI citations last?

Often less than a week. SISTRIX tracked 82,619 prompts for 17 weeks and found that ChatGPT Search replaced 74% of its cited domains each week. Google AI Mode, the chat-style search mode inside Google, replaced 56%. Treat citations as ongoing work: keep earning new mentions and check your sources every week.

What is AI citation tracking?

AI citation tracking means asking AI engines the same buying questions on a schedule and recording which websites each answer links as a source. Depra does this every day for ChatGPT with web search, Gemini and Google AI Overviews and every week for Perplexity, stores every answer, and shows which websites cite your competitors but never mention you.