Who Does ChatGPT Cite? A Monthly Brand Check

The short answer: It depends on the question. When someone asks an AI tool a factual question about a business (hours, price, location, what it sells), the answer comes mostly from the brand's own website and listings. When someone asks which brand is best, the answer comes mostly from other people: reviews, comparison articles, YouTube and Reddit. You need to be right in the first group and present in the second, and the only way to know where you stand is to ask the tools yourself, every month, and write down what they say.

This guide gives you a map of which sources matter for which questions, a monthly check with prompts you can copy, a scoring sheet and a 30-day plan.

Why the studies seem to disagree

Three pieces of research get quoted a lot, and at first they look like they cancel each other out.

Study	What it looked at	Headline finding
Yext, October 2025	6.8 million citations from ChatGPT, Gemini and Perplexity, for location-based questions in retail, finance, healthcare and food service	86% of citations came from sources the brand controls: 44% its own website, 42% business listings. Reviews and social were 8%, forums 2%
AirOps with Kevin Indig, 2026	Brand mentions across ChatGPT, Gemini, Perplexity and Google AI Mode	85% of brand mentions came from external domains, not the brand's own site
Ranqo preprint on arXiv, June 2026	Over 100,000 prompt responses across 100+ brands	Global brands appeared in 73% of relevant answers, mid-market brands in 44%, small brands in 11%. "Best of" list articles were about 21% of all citations

They are measuring different things. Yext asked about specific businesses and places, where the facts live on the brand's site and its listings. AirOps counted where a brand's name turns up when the AI is recommending or comparing, where the AI wants a view from outside. Both are true at once.

A caution: all three come from companies that sell AI visibility software, and the arXiv paper has not been peer reviewed. Treat the numbers as direction, not law.

Which sources the AI leans on, by question

This is our reading of those studies, put in the order a customer actually asks.

The customer asks	Example	What the AI leans on	Your job
A fact about you	"Is [brand] open on Sunday?" "Does [brand] deliver to Pune?"	Your website, Google Business Profile, directories	Make the facts identical everywhere
What you are	"What is [brand]?" "Is [brand] genuine?"	Your About page, Wikipedia if you have one, press coverage, reviews	A clear, plain description of the company, repeated consistently
Which is best	"Best protein powder under ₹2,000"	Comparison and "best of" articles, YouTube reviews, Reddit threads	Be included in the lists and reviews that already rank
You against a rival	"[brand] vs [competitor]"	Third-party comparisons, review sites, your own comparison page	Publish an honest comparison page, and earn independent ones
How to do something	"How do I choose a protein powder?"	Detailed guides, wherever they are	Write the best guide, with something original in it

The pattern: the more a question asks for an opinion, the less the AI trusts what you say about yourself.

Two more findings worth knowing. In the Yext data, each tool had its own habits: Gemini leaned on websites (about 52% of its citations) and ChatGPT leaned on listings (about 49%). And an earlier AirOps analysis, as reported by M&C Saatchi Performance, found that 68% of brand mentions appeared in only one AI model. Being visible in ChatGPT tells you nothing about Gemini.

The monthly AI visibility check

This takes about an hour. Do it on the same day each month.

Step 1. Write 12 prompts and never change them. Three from each of four groups. Here they are for a made-up plant protein brand in Mumbai, which we will call Brand K.

Facts: "Where can I buy Brand K protein?" / "How much is Brand K plant protein per kg?" / "Does Brand K deliver across India?"
Identity: "What is Brand K?" / "Is Brand K protein safe and lab tested?" / "Who owns Brand K?"
Category: "Best plant protein powder in India" / "Best protein powder under ₹2,000 for beginners" / "Which vegan protein has no added sugar?"
Comparison and how-to: "Brand K vs [largest competitor]" / "Plant protein or whey for weight loss?" / "How do I choose a protein powder?"

Step 2. Run each prompt in four places: ChatGPT, Gemini, Perplexity and Google's AI Mode. Use a fresh chat each time, logged out or in a private window, so your own history does not colour the answer.

Step 3. Record five things for every answer: Were you named? In what position? Was what it said about you correct? Which sources did it cite? Which competitors were named instead?

Step 4. Save the cited links. This is the most valuable column. Those URLs are the pages the AI trusts for your category. If the same "best of" article is cited in three tools and you are not in it, you have found this month's task.

Answers vary from run to run, so do not react to one result. The Ranqo paper found that the tone of what is said about a brand changes far more often than whether the brand is mentioned at all. Look at the trend over three months.

A simple scoring sheet

One row per prompt per tool. With 12 prompts and 4 tools, that is 48 rows.

Result	Points
Not mentioned	0
Mentioned, but with a wrong fact	1
Mentioned correctly	2
Mentioned correctly and among the first three brands, or your site is cited	3

The maximum is 144. Track the total, but also track each of the four groups separately, because they have different fixes.

A made-up example: Brand K scores 30 out of 36 on facts, 20 on identity, 4 on category and 10 on comparison. Total 64 out of 144. The reading is clear. The AI knows Brand K exists and gets its details right, but never recommends it. The work is off the website, not on it.

Any score of 1 is urgent whichever group it sits in. A wrong price or a wrong claim, repeated by an AI, costs you more than not being mentioned.

A 30-day plan

Week 1: baseline. Run the check. Fill in the sheet. List every wrong fact and every cited URL.
Week 2: fix the facts. Correct name, address, phone, hours, prices and product details on your site, Google Business Profile and every directory that appeared in the citations. Write one clear paragraph that says what the company is, and use the same wording everywhere.
Week 3: make your own pages quotable. For each how-to and comparison prompt, publish or rewrite one page that answers it in the first three lines, with a table and at least one thing nobody else has: your own test, your own numbers. This is the same discipline as answer engine optimization.
Week 4: go where the AI is already looking. Take the five most-cited third-party pages from your sheet. Ask the authors of list articles to consider your product, and send a sample. Get genuine reviews onto the platforms that showed up. Brief creators for honest YouTube reviews, which is where influencer marketing earns its place in search.

Then run the check again on day 30 and compare.

Three ways brands get this wrong

Buying mentions. Paid placements in low-quality lists and planted forum posts are easy to spot. Google's own guidance on AI search says that seeking inauthentic mentions across the web is less helpful than it seems, and that its spam systems apply here too.

Checking one tool. Most teams look at ChatGPT and stop. The overlap between tools is small.

Publishing more instead of publishing better. Twenty thin blog posts do not make you more quotable. One page with original data does.

Where this fits

Getting named in AI answers is what we call GEO. It rests on ordinary SEO: Google says its AI answers run on the same ranking systems as normal search, and the other tools search the web too. If the three terms blur together, our plain-English explainer sorts them out.