AI shopping assistants suggest stores to customers using structured product information, third party customer ratings and reviews, and product comparison information that answers questions directly. They do not make recommendations based upon a company’s brand size or advertising spend. If you want your store to show up across multiple platforms, you need clean schema markup, complete product attribute information, and independent third-party citations, not simply a visually appealing website.
AI shopping assistants suggest stores to customers using structured product information; third party customer ratings and reviews; and product comparison information (questions that are answered directly by the system). They do not make recommendations based upon a company’s brand size or advertising spend. AI shopping assistants such as Rufus (formerly known as Alexa for shopping); ChatGPT; Perplexity; google’s AI Mode; and Microsoft Copilot access different types of information. Therefore, if you want your store to appear in searches conducted across multiple platforms, you will need to include clean schema markup on your site; provide complete product attribute information; and receive independent third-party citations referencing your store, not simply create a visually appealing website.
Over the past couple of years i have been assisting small and medium sized retailers in Japan in getting their websites indexed by “search” engines which are no longer simply “search” engines. And while i am happy to say that there are many store owners who believe that being “recommended” by an artificial intelligence (AI) means that they write better product descriptions. The truth is the opposite. Being “found” by an AI requires that you become machine readable, something that sounds quite foreign when compared to what we thought of “machine readability” a mere four years ago.
Why did all these artificial intelligence shopping agents choose what stores people will buy from?
There has been some sort of a significant change within the last 18 months. Customers began asking chatbots like Rufus; Perplexity; and ChatGPT the same type of question that they previously asked google (“what are the best running shoes?”). However instead of receiving a list of options (such as they did from google), customers were receiving an immediate response including a link to purchase those items.
This change is significant. In 2025, retail technology publication Modern Retail wrote that nearly all major retailers and AI incumbents had begun competing for what is referred to as agentic commerce. Amazon expanded the capabilities of Rufus’ automatic purchasing abilities. OpenAI embedded a checkout feature into ChatGPT. Perplexity released an AI browser that caused Amazon to take legal action against the developer because of how it accessed product information. Additionally, according to a study published by Retail Technology Innovation Hub, 58% of consumers stated that they now rely upon AI tools for making specific purchase decisions (and therefore skip searching for product information on google).
Gartner estimates that the trend toward increased reliance upon AI tools is expected to grow rapidly. According to Gartner, by 2030 approximately 20% of all monetary transactions will be programmable. This will give AI agents the actual authority to effectuate purchases without a human having clicked on the button labeled “buy.” (see Gartner’s Top Predictions for IT Organizations and Users in 2026 and Beyond, October 2025.)
What is causing this trend to grow at such a rapid pace?
There were three factors that combined at the exact same time. First, large language models developed sufficient capacity to understand purchase intent (budget, use case, etc.) as opposed to simply keyword terms. Second, retailers created the necessary infrastructure (Shopify’s Storefront MCP; OpenAI’s Agentic Commerce Protocol with Stripe; google’s Universal Commerce Protocol) to enable AI agents to complete purchases, not simply describe products. Third, and perhaps most importantly, the conversion rates for visitors who arrive at a merchant’s website as a result of being referred by an AI shopping agent are extremely high. Analytics firm Genrise reported that the amount of ecommerce traffic generated by referrals from ChatGPT and Perplexity grew by roughly 300% during Cyber Week in 2025, and that these referrals converted at an average rate of 8 times greater than traffic coming from social media, according to analysis from Opascope.
That is not an insignificant difference. That represents a viable channel for merchants to develop.
How do each of the AI shopping agents make recommendations regarding which store to shop at?
As we move forward here, it becomes significantly more complex, and much of the “just add schema markup” advice provided to businesses is inaccurate. While each agent uses different criteria in order to evaluate stores, treating them collectively as one entity is likely the largest mistake i see merchants making regarding how they approach this issue.
| Agent / Tool | Primary Data Source | What It Values Most |
|---|---|---|
| ChatGPT Shopping | Open web, retailer pages, Google Shopping fallback | Visibility beyond the retailer page, structured product schema, comparison content |
| Perplexity | Third-party citations (review sites, expert blogs, Reddit) | Independently validating claims made by the brand |
| Amazon Rufus (Alexa for Shopping) | Amazon catalog, reviews, Q&A | Listing completeness, review sentiment, category relevance |
| Google AI Mode | Google Shopping feed via Merchant Center, UCP | Real-time structured product information, freshness, Merchant Center accuracy |
| Microsoft Copilot | Bing index | Price competitiveness, completeness of structured data |
ChatGPT relies heavily on google Shopping data behind the scenes; one analysis found a 75% overlap between the top products ChatGPT recommends and the top three organic results on google Shopping. Perplexity works almost the opposite way; it prioritizes third-party citations over anything a brand says about itself, which is why a store with glowing self-written product copy and zero independent reviews will lose to a competitor with mediocre copy and forty Reddit threads recommending it.
Amazon plays by its own rules entirely. It updated its robots.txt to block OpenAI’s crawlers, removing roughly 600 million products from ChatGPT’s shopping results, and it sued Perplexity over how its Comet browser accessed Amazon’s site. Amazon has also declined to join either OpenAI’s Agentic Commerce Protocol or google’s Universal Commerce Protocol. Worth noting, as of May 13, 2026, Amazon renamed Rufus to “Alexa for Shopping” in the US, per AMALYTIX; the functionality, data sources, and recommendation logic are unchanged, it’s the same system, new name.
If you sell on Amazon, you’re optimizing for a closed system that plays by its own rules. If you sell direct-to-consumer, you’re optimizing for an open ecosystem where the same product data feed can reach four different assistants at once.
What does a store actually need to get recommended?
This is the part that separates stores getting cited from stores getting skipped. And i’ll be honest, when i first started advising smaller retailers on this, i assumed a nicer product page would move the needle. It doesn’t, not on its own. What moves the needle is whether a machine can confidently extract facts from your page without guessing.
Complete, structured product data matters most. Research from SE Ranking, cited by Alhena, found that 65% of pages cited by google’s AI Mode and 71% of pages cited by ChatGPT include structured data. If your product pages don’t have Product, Offer, and Review schema, in JSON-LD, validated, matching what’s visibly on the page, you are functionally invisible to a huge share of AI shopping queries, no matter how good your actual products are.
Independent validation matters more than self-promotion. Since Perplexity and increasingly the other assistants weight third-party sentiment heavily, a store’s own marketing copy carries less weight than what strangers say about it elsewhere.
Answer-shaped content matters too. The stores winning citations write pages structured around the actual question a shopper is asking (“is this good for a small apartment,” “will this fit a wide toe box”) rather than generic marketing descriptions. Here’s a nuance worth flagging, because a lot of GEO advice online is now slightly out of date: Google formally deprecated the FAQ rich result feature in Search as of May 7, 2026, so the expandable Q&A dropdown under a listing no longer appears, and Search Console reporting for it phases out through August 2026. But per google’s own documentation, this is a search-appearance change, not a content-value change. FAQPage remains a valid schema.org type, and the underlying content, clear questions with direct answers, still gets crawled and cited by AI systems whether or not it produces a visual snippet in google. So keep writing genuine FAQ content; just stop expecting it to buy you a fancy dropdown.
(And on the subject of AI shopping getting a little unpredictable, Amazon suing another AI company for how it accessed a shopping website is either irony or foreshadowing, and i genuinely can’t decide which.)
Trust in closed-loop assistants beats trust in open ones, for now. Bain & Company’s 2026 research found that consumers trust retailers’ own on-site ai agents roughly three times more than third-party agents like ChatGPT or Perplexity. If you run a Shopify store with your own AI-assisted product finder, that in-house trust advantage is real and worth building on.
What should a store owner actually do this month?
Skip the theory. Here’s the sequence i actually walk clients through, in order of impact.
- Audit your product schema first, not your copy. Run your top ten product pages through google’s Rich Results Test and fix errors before touching anything else.
- Fill every attribute field, not just the required ones. Color, material, dimensions, compatibility, certifications; vague attributes lose to specific ones.
- Get your google Merchant Center feed current. This is the backbone that both google’s AI Mode and (indirectly) ChatGPT rely on.
- Actively pursue third-party mentions. Reach out to relevant review sites and respond to every review you get, including the negative ones.
- Write content in question form, aimed at real buyer language, not “Product Features.”
- Decide your Amazon posture deliberately. If you’re Amazon-first, optimize for Rufus specifically; if you’re DTC-first, you can largely ignore Amazon-specific optimization and focus resources on the open protocols instead.
What mistakes are stores making right now?
The most common one i see is treating this like a single campaign instead of an operating model; a store fixes its ChatGPT visibility and never touches Perplexity or google AI Mode, not realizing each one reads product data through a completely different lens.
The second mistake is assuming AI shopping assistants reward polish. They don’t. They reward completeness and verifiability.
The third, and this one is almost universal among smaller retailers, is ignoring Amazon’s walled garden entirely rather than making a conscious decision about it. A product that only exists on Amazon is effectively invisible to ChatGPT, Perplexity, and google AI Mode.
Frequently Asked Questions
Question: Which AI shopping assistant matters most for a small retailer?
Answer: It depends on where your customers already are. If you sell direct-to-consumer, google AI Mode and ChatGPT combined reach the widest audience since they share underlying data sources. If you sell primarily on Amazon, Rufus (now Alexa for Shopping) is the one that actually controls your visibility, and it operates by its own separate rules.
Question: Do i need to pay for advertising to get recommended by AI shopping assistants?
Answer: No. Recommendations from ChatGPT, Perplexity, and google’s AI Mode are generated from structured product data, review signals, and content quality, not paid placement.
Question: Is FAQ schema still worth using after google’s May 2026 deprecation?
Answer: Yes. google removed the visible FAQ dropdown from search results, but FAQPage remains a valid schema type that ai crawlers and retrieval systems still process.
Question: What’s the single biggest technical fix for a store with no AI visibility?
Answer: Complete, validated Product and Offer schema in JSON-LD, matched exactly to what’s visible on the page.
Question: Does having good reviews actually affect AI recommendations?
Answer: Yes, particularly for Perplexity and Amazon’s Rufus, both of which weigh review sentiment and third-party validation heavily.
Question: Should a small store try to optimize for all five major AI shopping assistants at once?
Answer: Generally, no. Build one strong foundation, complete structured data, a current Merchant Center feed, genuine third-party validation, since it reaches ChatGPT, google AI Mode, Perplexity, and Copilot simultaneously. Amazon requires a separate, deliberate decision.
The bottom line
Being recommended by an AI shopping assistant isn’t about gaming an algorithm the way SEO sometimes felt like gaming google in 2015. It’s about giving a machine enough verifiable, structured, honestly-sourced information that it can confidently vouch for you to a stranger. The store that wins is usually the one that actually has its data in order, not the one with the biggest ad budget.
If your product pages are still built for a human browsing on a Tuesday afternoon and nothing else, that’s usually the first thing worth fixing, and it’s exactly the kind of audit Ongito runs for retailers who want to know where their data is failing before an AI agent ever gets the chance to skip them. Get in touch with Ongito if you want a second set of eyes on whether your store is actually machine-readable, not just good-looking.