How AI shopping assistants choose products, and how to make yours the pick

AI shopping assistants such as ChatGPT, Google’s AI Mode, Perplexity and Claude choose products by matching the shopper’s request against product data they can access: merchant feeds, structured product information from data providers, crawled product pages and reviews. What they have documented points the same way: complete, accurate product data and pages their crawlers can reach make you easier to recommend. Paying for placement is not how these product results work, according to OpenAI and Perplexity.

There are two sides to this. Shoppers ask external assistants what to buy, so your catalog needs to be easy for those assistants to read and trust. And shoppers who reach your store, WhatsApp or Instagram want the same guided help from you. This guide covers both: what each platform publicly says, a practical checklist, and how to run your own AI shopping assistant in chat.

What is an AI shopping assistant?

The term covers two different things:

	External AI assistant	Your own AI shopping assistant
Examples	ChatGPT, Google AI Mode and Gemini, Perplexity, Claude	An assistant on your WhatsApp, Instagram DMs or website chat
Recommends from	Many stores and sources	Only your catalog
Who controls it	The AI company	You
Your job	Make your products easy to find, read and trust	Design the questions, data, guardrails and handover
Ends in	A link to a store, or checkout inside the assistant where supported	Your checkout, a payment link or a person on your team

How do ChatGPT, Google, Perplexity and Claude choose products?

None of these companies publishes a full ranking formula. Here is what each one states on its own pages as of October 2026.

ChatGPT

OpenAI’s help center article on shopping in ChatGPT search says:

• Product results are chosen for relevance to the shopper’s request, including context such as memory and custom instructions, and are “selected independently by ChatGPT and are not ads”.
• Inputs include structured metadata from first-party and third-party providers (such as price and description) and other third-party content.
• Labels like “budget-friendly” and review summaries are generated by the model, and OpenAI does not verify reviews and ratings.
• Merchants are ranked using factors such as availability, price, quality and whether they are the maker or primary seller.
• Shopify merchants are included through Shopify Catalog; other merchants can apply to send a product feed directly.

OpenAI’s developer documentation for agentic commerce describes that product feed as a regularly refreshed CSV or JSON file with identifiers, pricing, inventory, media and fulfillment details, where optional signals such as reviews can help ranking. Instant Checkout, which lets a shopper buy inside ChatGPT, is available to approved partners, and checkout and payment still run on the merchant’s systems.

On the crawling side, OpenAI says its OAI-SearchBot is used to surface websites in ChatGPT’s search features, and sites that block it are not shown in ChatGPT search answers, apart from navigational links. GPTBot, the training crawler, is controlled separately.

Google AI Mode, AI Overviews and Gemini

Google’s Search Central guidance on AI features is clear: there are no additional requirements, special schema.org markup or AI text files needed to appear in AI Overviews or AI Mode. A page needs to be indexed and eligible to show with a snippet, and normal SEO fundamentals apply. Google lists keeping your Merchant Center and Business Profile information up to date as a good practice.

In May 2026 Google announced AI performance insights for Merchant Center, to show how products are discovered on AI Mode, AI Overviews and the Gemini app. Among the planned reports is product attribute insights, with an attribute completeness score that shows which products are missing structured attributes such as color, style and material. Google said the rollout would start in the U.S., Canada, Australia, India and New Zealand. The message for merchants is direct: attribute completeness is something Google measures.

Perplexity

When Perplexity launched shopping in November 2024, it said its product cards are not sponsored and introduced a free merchant program for large retailers to share product specifications. It said joining increases the chance of being a recommended product, because the products are in its index, and that the program is unrelated to its ad products. Check Perplexity’s current pages for the latest terms.

Claude

Anthropic documents three crawlers. Claude-SearchBot crawls the web to improve search results, and Anthropic notes that blocking it may reduce your site’s visibility and accuracy in user search results. Claude-User fetches pages when a user’s question needs them, and ClaudeBot collects content for model development. As with OpenAI, you control each one separately in robots.txt.

What the documentation adds up to

Across these sources, a few signals repeat:

• Structured product data: price, availability, identifiers, attributes, images and shipping details, ideally from a feed you control.
• Accessibility: product pages the assistant’s crawler can reach, with key facts in text.
• Trust signals: reviews, clear seller identity (maker or primary seller), and consistent prices.
• Freshness: prices and stock that match what the shopper sees when they click.

Assistants also write their own summaries and labels. If your product page leaves out a limitation, a model may fill the gap from someone else’s review. Owning the facts is the best way to be described accurately.

How to make your products easy for AI assistants to recommend

1. Decide your crawler policy on purpose. Search crawlers and training crawlers are separate. Many stores allow search crawlers and make a separate decision about training.
2. Keep a complete product feed. Fill every attribute that applies: brand, GTIN or MPN, size, color, material, dimensions, compatibility, price, availability, shipping cost and time. Keep Merchant Center current, and use OpenAI’s feed or Shopify Catalog if you sell through Shopify.
3. Write specs as text, not images. Size charts, materials and compatibility locked in an image are hard to read for any crawler.
4. State limits honestly. “Not dishwasher safe”, “runs small, size up”, “fits laptops up to 15 inches”. Assistants answer comparison questions, and honest limits make your product the right match for the right shopper.
5. Collect and show real reviews. OpenAI’s feed documentation mentions reviews as an optional ranking signal. Never write or buy fake reviews.
6. Publish clear policies. Shipping times, return window, warranty and payment methods on crawlable pages, consistent with your feed.
7. Match prices everywhere. The same price in your feed, structured data and product page avoids a mismatch at the click.
8. Consider llms.txt as an optional extra. llms.txt is a proposal for a Markdown file at your site root that points AI agents to your key pages. Google says it does not need one for AI features, so treat it as a small, low-cost addition, not a ranking lever.

A quick robots.txt reference for the crawlers named above:

Crawler	Company	Purpose (per the company)
OAI-SearchBot	OpenAI	Surfaces sites in ChatGPT search features
GPTBot	OpenAI	Crawls content that may be used for model training
Claude-SearchBot	Anthropic	Crawls to improve search result quality
Claude-User	Anthropic	Fetches pages for a user’s question
ClaudeBot	Anthropic	Collects content for model development

How to build your own AI shopping assistant in chat

External assistants send shoppers to you with a shortlist. Your own assistant closes the gap between “this looks right” and “I’m sure”. It also catches shoppers who start in your WhatsApp, Instagram DMs or website chat.

Ask before you recommend

Good recommendations come from two or three short questions, not a catalog dump:

Customer: I need a running shoe for my first half marathon.
Assistant: Great goal. Two quick questions so I suggest the right pair:
1) Road or trail?
2) Do you need extra support, or a neutral shoe?
Customer: Road, neutral. Budget around 120.
Assistant: These two fit: the Stride 4 (lighter, firm ride) and the Cloudline 2 (more cushion for long runs). Both are in stock in size 42.
[Compare]  [Buy Stride 4]  [Buy Cloudline 2]

Ground every answer in your catalog

The assistant should recommend only products it found through a lookup in your catalog, with live price and stock. Give it tools to search products, check variants and stock, and fetch details, and a knowledge base for size guides and policies. If nothing fits, it says so and offers the closest option or a person.

Compare honestly

When the shopper asks “which one?”, answer with the trade-off, not a sales pitch: “The Stride 4 is lighter; the Cloudline 2 is softer on long runs.” Honest comparisons build the trust that makes the next recommendation land.

Make the next step one tap

Show two or three options as product cards or a carousel with buttons, then send a checkout or payment link. Our guide to WhatsApp interactive messages covers buttons, lists and carousels.

Hand over high-intent shoppers

Large orders, unusual sizes, gift questions and frustrated shoppers go to a person, with the conversation and selected products attached. The rules are in our AI-to-human handoff guide.

Measure what it sells

Track chats that reach a product card, chats that reach checkout, orders from chat, handover rate and the questions the assistant could not answer. Unanswered questions are your next catalog and content fixes, which also help with external assistants.

How to run an AI shopping assistant in MonoChat

MonoChat’s AI assistants answer on WhatsApp, Instagram, Facebook Messenger, TikTok, Telegram, web chat and more from one inbox, using the model you choose, including your own key for OpenAI, Anthropic, Google or your own LLM endpoint. AI function tools let the assistant call your systems, for example to check inventory or fetch an order, with schemas, permissions and logs. The AI knowledge base holds your size guides and policies.

The AI product finder solution shows the flow end to end: connect your catalog so products, variants and inventory are searchable, design guided questions, and recommend matches as product cards with clear next steps on WhatsApp, Instagram and web chat, with agent takeover for high-intent shoppers. On WhatsApp, carousel messages show up to 10 swipeable product cards, and the assistant can pick the cards that match the shopper’s intent, budget or earlier answers. Payment integrations send a payment link from your own provider account when an order is agreed in chat. Add the web chat widget to your store so the same assistant greets site visitors.

Shop Assistant and Store Assistant

Two ready-made apps from the MonoChat Store cover the most common setups:

• Shop Assistant for online stores: customers ask for a product, size or color and get product cards with photo, price and Buy and Details buttons. Buy opens your shop’s own checkout with the items already in the cart. It also tracks orders, handles return requests and recovers abandoned carts, connected to Shopify or to your own platform through the HTTP API or a CSV file.
• Store Assistant for shops and small chains: product cards with Reserve and Details buttons, stock per store, click & collect reservations and the nearest store with opening hours.

Both answer only from the text and data you approve and hand the conversation to your team when a person is needed. Start on the free Basic plan (one agent, one channel) at app.monochat.ai and compare plans on the pricing page.

Sources

• OpenAI Help Center: Shopping with ChatGPT search
• OpenAI Developers: Agentic Commerce key concepts
• OpenAI Developers: Overview of OpenAI crawlers
• Google Search Central: AI features and your website
• Google Merchant Center Help: Insights for AI-powered shopping experiences
• Perplexity: Shop like a Pro (November 2024)
• Anthropic: Does Anthropic crawl data from the web?
• llms.txt proposal