Which sources AI assistants cite, and how to find yours

Why the sources matter more than the answer

When an assistant tells a shopper that three brands are worth considering, the shopper sees the answer. You, as the brand, need to see what the answer rests on. If the answer describes your product as suitable for dry skin when it is meant for oily skin, the error did not come from nowhere. Somewhere there is a page, a listing or a review that says so, or a page that fails to say the right thing clearly.

That is why sources are the useful part of AI visibility work. You cannot edit an assistant. You can edit your own product pages, correct a marketplace listing you control, and respond to material published about you elsewhere. Finding the source turns a vague worry into a specific task.

Why there is no universal list of cited sources

Articles that name the five sites AI assistants cite most are describing one assistant, one set of questions and one moment. Assistants differ in whether they search the web at answer time, whether they show links, and how they choose between pages. The same assistant can change its behaviour after an update. A source that dominated answers about protein powders last month may not appear for skincare today.

For that reason we do not publish a league table of sources, and you should be wary of any supplier who does without giving the questions, the assistant and the dates behind it. What can be stated as a fact is narrower and more useful: for your category and your questions, on a given date, these were the answers, and these were the sources shown with them.

The kinds of page to check

Instead of a ranked list, think in types of page, and check each type against the saved answers for your category. In the Indian shopping context the usual candidates are these.

Your own website: product pages, ingredient or specification pages, size charts and help pages. These are the pages you control fully, and the easiest fixes. Marketplace and quick-commerce listings: the title, bullet points, images and description on Amazon.in, Flipkart, Nykaa, Blinkit, Zepto and others. Editorial and review pages: articles, comparisons and roundups that discuss your category. Community and video content: forum threads, question sites and video descriptions where shoppers discuss products in their own words.

An assistant may lean on one type for one question and another type for the next. The point of checking each type is to find where your brand is described, and whether the description is accurate, current and specific.

• Pages you control: your site and the listings your team edits.
• Pages you influence: reviews, retailer content and partner pages.

A method you can run yourself

Write down twenty or so questions a shopper in your category would ask, in plain wording and without your brand name. Include questions about a use case, a concern, a price range and a comparison. Ask each one in the assistant, more than once. For each answer, copy the text, the date and any sources or links shown.

Then sort what you have. Which brands were named? Which sources appeared, and how often does each recur? Where your brand was named, what did the answer say about it, and does that match your product? Where it was missing, which brands took the place, and which sources supported them?

Two things to keep in mind while doing this. The answer text is a record of what was said, not of why. Saved answers do not reveal the assistant's internal reasoning, so any statement about why a source was chosen is inference, and should be labelled as such. And sources shown are the ones the assistant chose to display, which may not be everything it read.

Build a checkable answer record

For a manual review, keep the question, assistant, date, answer and any cited links together. Record how you captured the answer. Separate what the answer says from your explanation of why it appeared, and review any proposed page changes before publishing them.

Check whether crawlers can access the pages you want them to read, and whether product details are clear and structured. Accessible pages still do not guarantee a citation. The crawler readiness guide explains what these checks can establish.

The share of AI guide explains how to count how often you are named, and the zero-click scorecard shows how to review the answers month to month.

Turning a source into a fix

Once you know which pages an answer leans on, the fixes are usually plain. Make product pages state what the product is, who it is for, what it contains and what it costs, in words a shopper would use. Add structured data so machines can read price and availability. Keep marketplace titles and descriptions consistent with your own site, since contradictions between sources confuse both shoppers and assistants. Correct out-of-date pages rather than leaving them to be quoted.

Then ask the same questions again after the changes and compare with the saved answers. Expect movement to be uneven and slow, and do not promise a specific result to anyone. The evidence you keep, the before and after answers, is what lets you show what changed.

A worked example, without the numbers

Imagine a shampoo brand whose product is made for oily scalps. A shopper asks an assistant for a shampoo that will not weigh down oily hair. The assistant names three brands and yours is missing. The team saves the answer and looks at what was shown alongside it. Two of the sources are articles that compare shampoos by hair type, and one is a marketplace listing for a rival.

Reading your own listing next to the rival's, the team notices that the rival's title and bullet points say oily scalp in plain words, while yours describes the product only as a clarifying formula. Nothing on your own website says oily anywhere in the copy. That is a specific gap between how shoppers ask and how your pages describe the product.

The fix is ordinary writing. State who the product is for in the words a shopper uses, on your website and on each marketplace listing your team controls. Then check the editorial pages: is your product listed on them, and is the description right? If it is missing or wrong, that is a matter for whoever handles your press and reviews, and the answer is a polite request with the correct facts, not a demand.

Ask the same question again a few weeks later and save the new answer. Whether you are named or not, you now hold two dated records and a change made in between, which is what lets you show what happened. If the brand is still missing, look at the next source rather than assuming the first fix failed.

Common questions