TL;DR
- No loyalty to brands: AI shopping assistants recommend products based on the data they can read about them: specs, prices, stock, reviews and descriptions. A famous name with thin product data can lose to a small shop with rich data.
- Every major assistant now shops: ChatGPT, Google AI Mode and Gemini, Perplexity, Amazon and Microsoft Copilot all show products, and most can now complete a purchase for the user.
- AI traffic converts: according to Adobe, visits from AI assistants to US retail sites converted 54% better than other traffic in May 2026. A year earlier they converted worse.
- Descriptions matter more than before: shoppers describe situations (“a gift for someone who loves cooking, under €80”), while most product pages describe only technical specs.
- Answers are not always right: a consumer test by Which? found ChatGPT named the correct lowest price and retailer for only 8 of 25 products. Outdated store data produces outdated answers.
Contents
- From search results to a shortlist
- Who is shopping inside AI now
- The signals AI assistants use
- What research shows about AI product choices
- Ads, accuracy and other caveats
- What this means for stores and brands
- FAQ
A shopper types into ChatGPT: “I need a gift for my father. He loves cooking, budget up to €80.” A few seconds later there are three or four suggestions, each with a reason, a price and a link to a store. There is no page of ten blue links to compare. The assistant has already chosen. For stores and brands, the question is no longer only how to rank on Google, but how to become one of those three or four products.
From search results to a shortlist
Classic search returns a list and leaves the choice to the shopper. AI assistants return a shortlist and explain why each item is on it. That changes what a product needs in order to be seen.
| Classic search | AI shopping assistant | |
|---|---|---|
| What the shopper types | Keywords (“chef knife 20 cm”) | A situation (“gift for a father who cooks, under €80”) |
| What comes back | Ten or more links and ads | Three to five products with reasons |
| Who compares | The shopper | The assistant, before the shopper sees anything |
| What helps a product | Rankings, backlinks, ads | Clear, current, structured product data and reviews |
| Where the purchase happens | On the store’s website | Increasingly inside the chat |
This does not mean classic search is disappearing. It means a growing share of shoppers start with a question instead of a keyword, and the assistant answers it by filtering products on their behalf.
Who is shopping inside AI now
All major assistants have added shopping features since 2025, and most now support some form of checkout, often called agentic commerce.
| Assistant | What it does for shoppers | Checkout inside the chat |
|---|---|---|
| ChatGPT | Product results with prices and reviews; a “shopping research” mode that builds a buyer’s guide | Instant Checkout launched in September 2025 with the open Agentic Commerce Protocol, built with Stripe. In March 2026 OpenAI said checkout is moving to merchant apps inside ChatGPT |
| Google AI Mode and Gemini | Answers drawn from the Shopping Graph, which Google says holds over 50 billion listings, about 2 billion refreshed every hour | Agentic checkout via Google Pay, and the Universal Commerce Protocol, an open standard built with Shopify, Etsy, Wayfair, Target and Walmart |
| Perplexity | Free shopping answers for US users, personalised from memory | Instant Buy through PayPal |
| Amazon | Alexa for Shopping, which replaced the Rufus assistant in May 2026, with price history and “Buy for Me” | Yes; Buy for Me can also complete purchases on other retailers’ websites |
| Microsoft Copilot | Product comparisons inside Copilot | Copilot Checkout, launched in January 2026 with PayPal, Shopify, Stripe and Etsy |
The numbers behind this shift are still small in absolute terms but growing quickly. Adobe measured a 693% year-on-year rise in generative AI traffic to US retail sites during the 2025 holiday season. In a survey of more than 5,000 US consumers, 39% said they had used AI for online shopping. In March 2025, AI-referred visits converted 38% worse than other traffic. By May 2026 they converted 54% better, according to Adobe data reported by Digital Commerce 360.
The signals AI assistants use
No assistant publishes its full ranking formula. What the companies do say points in one direction: the decision rests on data about the product, not on the size of the brand.
OpenAI states that product results in ChatGPT “are organic and unsponsored, ranked purely on relevance to the user”. When several sellers offer the same item, it looks at availability, price, quality, whether the merchant is the primary seller and whether checkout is enabled. Its shopping research mode reads what it calls trusted sites, avoids low-quality ones, and can use the shopper’s memory and past chats.
Google’s AI answers draw on Merchant Center feeds and the Shopping Graph. In January 2026, Google announced dozens of new Merchant Center attributes for AI shopping, such as answers to common product questions, compatible accessories and substitutes.
| Signal | What the assistant reads | Why it matters |
|---|---|---|
| Structured product data | Feeds and markup: title, price, availability, brand, specs, images | It is the fastest and most reliable way for a machine to understand a product |
| Freshness | Current price and stock | OpenAI lists availability and price among its ranking factors |
| Reviews and ratings | Volume, average score, what reviewers say | Used to judge quality and to explain the recommendation |
| Use-case language | Who the product is for, in which situation, what problem it solves | Matches the way shoppers phrase questions |
| Third-party mentions | Buyer’s guides, comparisons, forums, editorial reviews | Helps the assistant decide which sources and products to trust |
| Personal context | The shopper’s memory, past chats, budget, location | The same question can return different products for different people |
The last row explains a common experience: the same request can give different products in ChatGPT and in Google, and even for two people using the same assistant. Each one draws on different data sources and different context.
What research shows about AI product choices
Independent research on AI shopping is new, but a few findings repeat.
A 2025 study by researchers from Columbia, Yale and MyCustomAI (Allouah et al.) tested AI shopping agents in a simulated store. The agents rewarded higher ratings and more reviews and penalised higher prices. They were also strongly influenced by where a product appeared on the page, and their choices changed a lot between model versions: one fitness tracker’s share fell from about 25% to 6% after a GPT update, while a Claude update raised it from 45% to 77%. Product descriptions rewritten with AI gained 3 to 15 percentage points of share in five of the six models tested. The study also found that some categories were dominated by one brand, so brand is not irrelevant.
An earlier Harvard study showed that hidden text added to a product page could push a language model to recommend that product first. The finding is usually cited as a warning about manipulation, not as a tactic.
The original generative engine optimization paper from Princeton and partners found that adding statistics, quotations and sources could raise a page’s visibility in AI answers by up to 40%. It tested general questions, not shopping specifically.
| Finding | Source | Caveat |
|---|---|---|
| Ratings and review volume raise the chance of selection | Allouah et al., 2025 | Simulated store, specific model versions |
| Higher prices lower the chance of selection | Allouah et al., 2025 | Same |
| Choices shift sharply between model versions | Allouah et al., 2025 | Results today may differ tomorrow |
| Clear, evidence-rich content gets cited more | Aggarwal et al., 2024 | General queries, not product-specific |
| Hidden instructions can manipulate rankings | Kumar and Lakkaraju, 2024 | Tested on fictitious products; platforms work to block it |
The research and the companies’ own statements point the same way: brand alone does not decide. Data quality, reviews and price weigh heavily, and that gives small stores with well-described products a real opening against bigger competitors.
Ads, accuracy and other caveats
Ads. Sponsored results are entering AI shopping. Google shows sponsored listings and “Sponsored deal” offers inside AI Mode. ChatGPT began testing ads in February 2026 on its Free and Go plans. OpenAI says ads are labelled, visually separated from the answer and “do not influence the answers ChatGPT gives you”. That is the company’s own claim and has not been independently audited.
Accuracy. In October 2026 the UK consumer group Which? found that ChatGPT gave the correct lowest price and retailer for only 8 of 25 products, and 68% of results had at least one problem. OpenAI’s own documentation warns that shopping research “might make mistakes about product details like price and availability”. For stores, the risk runs both ways: an assistant can show an old price, the wrong stock status or outdated specs.
Machine readability. Adobe scored retail product pages at 66% for how easily AI systems can read them, lower than category pages and homepages. A product that a human understands from photos and a video may be almost invisible to an assistant that reads text and structured data.
Regulation. EU consumer law was written for human buyers. A July 2026 report by the think tank CERRE notes that it is still unclear how rules on consent and information duties apply when an AI agent completes the purchase. The Digital Fairness Act, which the European Commission has been preparing, is expected to address some of these questions.
What this means for stores and brands
The shift does not require a new kind of marketing, but it does change what a product page has to do. A short checklist of what to review:
| Area | What to check |
|---|---|
| Product feed | Is there a feed in Google Merchant Center, and is the store eligible for ChatGPT’s merchant program? Are price and stock updated daily or faster? |
| Descriptions | Does each key product say who it is for and in which situation, not only its specs? |
| Reviews | Do key products have enough recent reviews, on the store and on third-party sites? |
| Structured data | Does the product page carry product markup with price, availability and ratings? |
| Outside mentions | Does the product appear in independent guides, comparisons and forums? |
| Testing | What do ChatGPT, Gemini and Perplexity say when asked for the product category with a realistic budget and situation? |
The last row is the simplest place to start. Ask the assistants the questions your customers ask, in their words, and note which products come back and why. For a related look at how assistants choose service businesses rather than products, see How AI Assistants Decide Which Businesses to Recommend, Explained, and for the mechanics behind citations, How AI Search Chooses Its Sources.
FAQ
How does ChatGPT decide which products to show?
OpenAI says product results are organic and ranked by relevance to the user. It uses structured product data, prices, availability, reviews and, when memory is on, the user’s past preferences. When several sellers offer the same item, availability, price, quality and the seller’s role also count.
Can a store pay to appear in AI shopping answers?
Partly. Google shows sponsored listings and sponsored deals in AI Mode. ChatGPT tests labelled ads on its Free and Go plans. OpenAI says organic product results are unsponsored and that ads do not affect them.
Do big brands get an advantage in AI recommendations?
Not automatically. Research and company statements show that data quality, reviews and price weigh heavily, although some categories are still dominated by well-known brands. A large brand with thin or outdated product data can lose to a smaller store with clear, current information.
Why does the same question give different products in ChatGPT and Google?
Each assistant draws on different data: Google on its Shopping Graph and Merchant Center, ChatGPT on structured product data from merchants and providers and on the sites it trusts. Personal context, such as memory, location and budget, also changes the result.
Are AI shopping prices reliable?
Not always. A 2026 Which? test found ChatGPT named the correct lowest price and retailer for only 8 of 25 products. Shoppers should check the final price on the store’s page before buying.
What is agentic commerce?
It is shopping in which an AI assistant does more than recommend: it adds items to a cart and completes the purchase for the user, usually through protocols such as OpenAI’s Agentic Commerce Protocol or Google’s Universal Commerce Protocol.
Vyara’s take
I shop through AI myself. When I looked for a gift with a budget and a short description of the person, I bought from the shortlist, and one of the suggestions was a small shop I had never heard of. When I ran the same request in Google, I got different products. With clients I see the other side: strong products that AI simply doesn’t mention, and competitors with more reviews and more detailed descriptions recommended instead. Most product pages are written for people who already know what they want, while AI is answering people who describe a situation. AI has no loyalty to brands, only to data, and that is good news for smaller stores. Classic SEO alone is no longer enough, but the fix is mostly better information, not a bigger budget.
Sources: OpenAI Buy it in ChatGPT · OpenAI Shopping research · OpenAI Testing ads in ChatGPT · Digital Commerce 360 on OpenAI checkout plans · TechCrunch on Google AI Mode shopping · Google Universal Commerce Protocol · Engadget on Perplexity shopping · eMarketer on Alexa for Shopping · Axios on Copilot Checkout · Adobe holiday 2025 · Adobe Q1 2026 · Digital Commerce 360 on Adobe May 2026 · Allouah et al. 2025 · Kumar and Lakkaraju 2024 · Aggarwal et al. GEO · Which? AI deals test · CERRE on agentic AI
About the Author
👉Want guest posting?🪙Subscribe to the @egorithms Telegram