How to check whether AI assistants name your brand

What AI visibility means

More shoppers now type a question into an AI assistant before they open a shopping app. They ask which shampoo suits hair fall, which protein powder is worth buying, or which brand of ghee is safe for a child. The assistant replies with a few brand names and a reason for each. If your brand is one of them, a shopper meets you before they meet a shelf. If it is not, they may never search for you at all.

AI visibility is the plain description of that situation: whether the assistants shoppers use name your brand when they answer the questions your customers ask, and what they say about it when they do. Some people call the work AEO, short for answer engine optimisation. The label matters less than the habit behind it. You look at what the assistants said, you save it, and you compare it with what you would want a shopper to be told.

It is a different check from search rank. A search results page lists many links and lets the shopper choose. An assistant compresses the choice into a short answer, so the difference between being named and not being named is sharper.

How an assistant comes to name a brand

Nobody outside the companies that build assistants can say exactly why a particular brand appeared in a particular answer. What can be described is the material an assistant works from, and there are two kinds.

The first is what the underlying model absorbed during training: text from across the internet, up to some date. A brand that was widely written about before that date, in reviews, articles, forum threads and retailer listings, is easier for the model to recall. A newer brand, or one that is mostly discussed in Hindi or Hinglish while the model is asked in English, may be thin in that memory.

The second is what the assistant looks up when you ask. Many assistants can search the web for an answer and read the pages they find. When they do, the answer leans on whatever pages are easiest to find and read for that question: a retailer listing, a comparison article, a review thread, a brand’s own product page. Those pages can change from week to week, so the answer can change with them.

Both routes reward the same thing. Clear, consistent information about what your product is, who it is for and how it differs from the alternatives, published in more than one place that a reader would trust.¹

Named, recommended and cited are three different results

A first check usually collapses three separate outcomes into one, and it is worth keeping them apart from the start.

• Named. Your brand appears somewhere in the answer, perhaps in a list of options or as an example of what to avoid.
• Recommended. The answer points the shopper towards your brand and gives a reason. Being named in a list of ten is not the same as being the first suggestion for a shopper with a stated need.
• Cited. The answer links to or quotes a page as its source. Some assistants show sources and some do not, and a cited page may be a retailer or a review site rather than your own website.

The three can move separately. A brand can be named often and recommended rarely. A brand can be recommended without any page of its own being cited. Each one tells you something different about what to do next, so record them in separate columns.

Why one answer proves little

The easiest mistake is to ask an assistant one question, see your brand missing or present, and draw a conclusion. Several things can change the answer without anything changing about your brand.

• The wording of the question. “Best shampoo for hair fall in India” and “which shampoo should I use for hair fall, budget is tight” can produce different brands. A question that includes your brand’s name tells you almost nothing, because the assistant has been handed the answer.
• The assistant. Different assistants draw on different models and different sources. Being named in one says little about another.
• The moment. Asking the same question twice can give two different answers, and an assistant that searches the web will follow the web as it changes.
• How it is being asked. The reply a shopper sees in a consumer app can differ from the reply a developer receives through the same company’s interface, and from what a signed-in shopper with a history sees.²

So a check is a set of questions, each asked more than once, on each assistant you care about, with every answer saved with its date. A pattern across that set means something. A single reply is an anecdote.

A worked example, with an invented brand

This example is an illustration. The brands are invented, and no result in it was measured.

Suppose you sell a sulphate-free shampoo and you want to know how assistants treat your category. You pick a question a shopper would ask without knowing your name: “Which sulphate-free shampoo is recommended in India?” You ask it in two assistants, three times each, on the same day, and paste each full answer into a sheet.

Reading the six answers, you record for each one which brands were named, in what order, what reason the answer gave, and which pages, if any, it cited. Say you find that a brand called Brand A is named in most answers and described as gentle on colour-treated hair, while your brand appears in only one and is described only as “also available”.

That tells you where to look next, and it does not tell you why. You now open your own product page and compare it with the wording the answers used for Brand A. Does your page say what makes the shampoo suitable for colour-treated hair? Does it state the pack size, and does the same information appear on the retailer listings that sell you? If the answers cited a review article, is your product in it?

You form a testable idea, for instance that your page never explains who the shampoo is for. You change the page, wait, and run the same questions again. The second run may show a difference or it may not, and either result is information. You have not proven that the change caused anything, because the assistants also change. You have a record you can compare.

How to run your first check by hand

You can start today with a spreadsheet and the assistants you already have access to.

1. Write ten to fifteen questions a shopper in your category would type, without your brand name in any of them. Include the broad ones (“best”, “for beginners”), the comparison ones and the worry ones (“is it safe for…”). If your customers write in Hindi or Hinglish, add some of those.
2. Choose the assistants that matter to your shoppers. Open a fresh conversation for each question so earlier questions do not colour the answer.
3. Ask each question at least twice. Paste the whole answer, not a summary, into your sheet with the assistant’s name and the date.
4. For every answer, record whether your brand was named, whether it was recommended, the reason given, and any pages the answer cites. Do the same for the two or three competitors you care about most.
5. Look for patterns across questions before looking at any single answer. Which questions never name you? Which competitor is named for a reason that is also true of you?
6. Repeat in a few weeks with the same questions, so any change is comparable.

Keep questions that a shopper would ask, not questions that flatter you. A check built from your own marketing language will tell you how well the assistant repeats your marketing language.

What to look at on your own pages

Once you have a record, the useful move is to connect the language in the answers to the information on your own pages and listings. These are things worth checking, not steps that produce a mention.

Does your page answer the question a shopper asked?

If shoppers ask which product is suitable for a stated need, check whether your page says who the product is for and why, in plain sentences rather than only in a tagline or an image.

Is the same product described the same way everywhere?

A name, a pack size or an ingredient list that differs between your website, Amazon.in and a quick-commerce listing gives an assistant conflicting material to summarise. The discrepancy is something you can find and fix.

Can the page be read?

A page whose text only appears after scripts run, or sits inside images, may be hard for an assistant that reads pages to use. Check it yourself: open the page with scripts turned off and see whether the product text is still there. The AI crawler readiness guide lists the other checks to run on your own site.

Be cautious with anyone who claims a specific change will earn a mention. No one outside an assistant’s maker can promise how it will answer, and answers can shift with no change on your side.

Where Rewalto fits

You can run the method above with a small question set and a record of each answer. Rewalto no longer offers the AI visibility research service described in earlier versions of this guide. Current data products and their availability are listed in the data catalogue.

A careful record keeps the reasons an answer states and the sources it shows. It cannot reveal an assistant’s internal reasoning, and no one can promise a mention, a ranking or a citation. A later check can show whether a recorded answer changed after you changed a page.

Keep each saved answer beside your interpretation, and let your team decide whether it supports a change. Repeat the same questions after an edit, while recognising that assistant answers can also change for reasons outside your website.

Common questions

Notes

1. This describes how assistants behave in general and how a brand can respond. It is not a measurement by Rewalto, and it contains no figure of ours. Assistants change without notice, so read any description of how they choose brands as describing the date this guide was last updated, and no later date. Back to text
2. Many assistants are available both as a consumer app and through a developer interface, and the two can differ in the model used, in whether a web search runs and in what location the request appears to come from. A check should say which one it used. Back to text