When Shoppers Ask AI What to Buy, Is Your Brand in the Answer?

A shopper who asks an assistant what to buy gets three or four brand names, and your brand is either one of them or absent. There is no second page to scroll to, no sponsored slot to buy your way into after the fact, and nothing in your ad console that tells you which side you landed on.

In 2026 the front door moved. Amazon renamed Rufus and folded it into Alexa for Shopping, Google put a cart inside Search AI Mode and Gemini, and OpenAI started selling ads through Amazon's DSP. The volume running through those doors is still small next to a normal results page, which is exactly why this is the cheap moment to find out where you stand.

Quick Answer

• What changed: Discovery is shifting from a results page with dozens of slots to an assistant answer that names a handful of brands. Amazon, Google and OpenAI each shipped a version of this in 2026.
• The opportunity: Adobe reports AI-referred retail visitors converting at higher rates than other traffic, and knowing where you stand today costs little.
• The risk: Adobe also describes AI traffic as modest in overall volume, and a Feedvisor and Zogby survey found only 5% of respondents would buy straight away on an AI recommendation. Moving budget on this evidence would be premature.
• The distinction that matters: A brand can appear in an assistant's answer without being the one it recommends. Measure both.
• Our read: Measure now, spend cautiously, and treat any claim about what drives recommendations as unproven unless it comes from a source that can show its work.

There is no page two in an AI answer

A search results page gives a brand dozens of chances to appear: organic positions, sponsored placements, a second page for the patient shopper. An assistant answer compresses all of that into a short list with a reason attached to each name. If you are not on the list, the shopper never sees you, and the shopper never sees that you were missing.

That changes what a brand lead needs to know. In search, the question was where you ranked for a term. In an assistant answer, the question is whether you are one of the brands it names, what it says about you, and which sources it leans on to say it. Those are different measurements from the ones in an ad account, and none of them show up in impression share. For the ad-side version of that metric, see our guide to Amazon share of voice.

Amazon, Google and OpenAI each changed the front door in 2026

Amazon: Rufus became Alexa for Shopping, and ads came with it

Amazon renamed Rufus to Alexa for Shopping on May 13, 2026, merging it with Alexa+ and adding price history and Buy for Me. On earnings calls, Andy Jassy put Rufus at more than 350 million users over the trailing twelve months and said those users are about 60% more likely to purchase. Both figures are Amazon's own.

On the ads side, Amazon Ads announced on June 11 that existing sponsored campaigns are automatically eligible in Alexa for Shopping. Sponsored Prompts reached general availability in the US on March 25 at existing CPCs. In September Amazon reported that Sponsored Products and Sponsored Brands prompt clicks convert to a sale 48% more often and spend 21% more on average. That is a self-reported figure, and it describes clicks on paid prompts, not what the assistant recommends organically. Our walkthrough of sponsored products in Amazon's conversational shopping covers the mechanics, and the unBoxed and Accelerate 2026 recap covers how the Amazon Ads Agent fits around it.

Google: a cart inside AI Mode, with Amazon missing

Google's Universal Cart launched in the US on October 15 across Search AI Mode and Gemini. Named retailers include Walmart, Target, Sephora, Nike and Wayfair, plus Shopify merchants. Amazon is not on the list. A retailer also gets no signal when a shopper saves an item to the cart, so a brand selling through those retailers cannot see this demand building.

OpenAI: ads through Amazon's DSP, managed

A pilot announced September 10 lets select advertisers buy ChatGPT ads through Amazon DSP, on a managed basis. It is a pilot, not a self-serve channel, and it says more about where Amazon expects budgets to go than about where they are today.

The channel is still small, and the evidence on both sides says so

Growth rate and conversion quality of assistant-led traffic are well documented. Share of purchases is not, and the sources that quantify the growth are careful to say the base is small.

Adobe's own forecast coverage is summarized by Digital Commerce 360. The pattern across all four rows is the same: the direction is clear and the size is not. That argues for measuring now and spending cautiously, not for shifting budget.

A brand can be mentioned in an answer without being the one recommended

Most early tracking counts mentions: did the assistant name us. That number flatters. An answer might list five brands, call one of them the best fit for the shopper's stated need, and mention yours as an alternative with a caveat. A mention count scores that as a win.

The useful measurement separates three states for each shopper question you test: absent, named, and recommended. It also records which sites the assistant cited when it answered, because those are the pages shaping the answer about you. The gap between named and recommended is where the work is.

What moves a recommendation is partly unknown, so treat claims with suspicion

Listing content, product attributes, reviews and off-Amazon content such as editorial reviews and forum threads all plausibly feed an assistant's answer. Nobody outside the AI companies knows the full recipe, and anyone selling a formula for it is guessing. The honest way to learn what matters for your category is to read the reasons the assistant gives and the sources it cites, across enough questions that patterns repeat.

That gives you a short list of things to check on your own listings: whether the attributes a shopper would name in a question are filled in and consistent, whether your review themes match the use cases in the questions, and whether the third-party pages being cited about you say what you would want them to say. Treat each as a hypothesis to test against the answers, not a rule.

Measure now, spend later: a baseline you can build this week

Step 1: write the question set

Draft 15 to 25 questions a shopper in your category would ask in their own words: use-case questions, comparison questions, constraint questions ("under a certain price", "for a specific need"). Pull them from search-term reports, customer service threads and reviews instead of inventing them.

Step 2: run each one across the assistants that matter

Ask each question in ChatGPT, Claude, Gemini and Perplexity, and in Alexa for Shopping if you sell on Amazon. Record the five fields from the table above for every answer.

Step 3: repeat it, because answers move

One pass tells you almost nothing, since the same question can return different brands on different days. Run the set again and compare. Here the manual version stops being sustainable: 20 questions across four or five assistants, recorded in five fields and repeated, is several hundred data points in a spreadsheet nobody wants to maintain.

Where Qore's AI Share of Voice fits

Step three is where our product Qore comes in. Qore is where you turn a review you do by hand into a workflow that runs itself, and AI Share of Voice is the workflow for this particular review. It runs a scan of your shopper questions across ChatGPT, Claude, Gemini and Perplexity, and it reports which brands win the answers and which sites the assistants cite. A Why AI picks you breakdown shows the reasons given, and a Cited vs Chosen view compares your standing against up to 10 competitors you choose, so the named-versus-recommended gap is visible instead of inferred. Results can be shared as a read-only link or exported as PDF or Markdown, which is useful when the person who needs to see the baseline is not the person running it. You can see where it sits among the rest of the platform on the Meet Qore page, and our post on Qore as agentic commerce for Amazon and Walmart covers the mechanics of how Qore workflows run.

The limits are specific. Amazon's own assistant, Alexa for Shopping, is not among the four assistants it scans today. The scan reports what assistants say, not why their models say it, so the Why AI picks you view is a reading of the stated reasons and cited sources, not a window into the model. And a scan tells you where you stand; it does not change your listings, reviews or third-party coverage. That work still belongs to you, and the scan is how you decide where to start.

Conclusion: find out where you stand before the channel gets crowded

The evidence supports two statements at once. Assistant-led shopping is growing quickly and converting well, and it is still a small share of how people buy. Both are reasons to act, and neither is a reason to move budget. Build the baseline, separate named from recommended, and revisit it as the volume numbers firm up. For the ad-side companion metric, read share of voice versus share of shelf on Amazon. If you want the AI-answer scan running without the spreadsheet, you can book a walkthrough of Qore.