Back in March 2023, ChatGPT made its first foray into ecommerce. OpenAI launched plugins, including Instacart, Klarna, and Shopify, which let users search for products, compare prices, and fill a grocery cart without leaving the chat. In April 2025, OpenAI added shopping to ChatGPT search, with product results that showed images, prices, reviews, and links to buy.
Over the same stretch, other AI platforms built their own shopping capabilities. Perplexity launched Buy with Pro in November 2024, with one-click checkout for its paying subscribers in the US. Google brought shopping to AI Mode in May 2025, and a year later expanded its Universal Commerce Protocol so shoppers can check out with Google Pay inside Search and the Gemini app. Microsoft launched Copilot Checkout in January 2026 with PayPal, Stripe, and Shopify. And Shopify now syncs merchants' catalogs to ChatGPT, Copilot, AI Mode, and Gemini from Shopify Admin.
Then the agents started shopping for people. On September 8, 2026, Meta launched Muse, a personal agent that runs in its own app and in WhatsApp, browses the web, and pays for purchases with a single-use card through Stripe Link. Muse logged 2.8 million installs in its first 12 days, and on September 22 it replaced ChatGPT as the top free app in the US App Store. Instinct, an invite-only assistant that people text or call, had been doing the same kind of work since February. On September 28, founder Noah Shinn said on the Invest Like the Best podcast that Instinct handles "a billion dollars a year or more in transaction volume," about half of it travel, and described growth of about 10% a day with no marketing spend. The same week, Instinct raised $1 billion at a $10 billion valuation.
Personal agents change who does the shopping. A ChatGPT user reads a list of products and picks one. A Muse or Instinct user hands over the task as they would to a personal assistant, and the agent researches options, chooses what to recommend, and can complete the purchase once the user approves.
For ecommerce brands, that adds up to two new places where shoppers decide what to buy, and neither works like a search results page. Answer engines pick a handful of products, show a price and a store for each, and send the shopper off to buy. Personal agents narrow the choice for one specific person and, where your site allows it, buy on their behalf. Which products make either list depends on product feeds and catalogs, reviews, third-party mentions, the retailers that carry you, and how easily software can read and use your site. A new category of ecommerce GEO tools has emerged to track your placement on those lists and improve it.
Below, we cover how AI shopping engines choose products, how personal agents like Muse and Instinct shop differently, what to look for in an ecommerce GEO tool, and the five platforms worth considering.
How AI shopping engines decide which products to show
Here's what happens between a shopper's prompt and the product cards they see.
1. The prompt has to trigger shopping results
No engine shows products for every prompt. When a search reads as shopping, Google's AI Mode opens a product panel that updates as the shopper refines the request, and ChatGPT, Copilot, and Perplexity show rows of product cards.
What counts as a shopping prompt is best documented for ChatGPT. In our analysis of 100.7 million prompt runs, naming a physical product made shopping results 2 to 6 times more likely than the baseline, while buying language like "I want to buy" barely moved the rate unless the prompt also named a product.
Category decides the rest. Apparel prompts triggered shopping results 62% of the time, physical products in general 56%, groceries and other consumables 30%, and health and medical products 6%.
2. The engine gathers candidate products
This is where the engines differ most, because each one pulls product data from its own sources:
- Google AI Mode and Gemini draw on Google's Shopping Graph, which Google says holds more than 50 billion product listings, with more than 2 billion refreshed every hour. Merchants feed it through Merchant Center. For a detailed request, AI Mode runs several searches at once to work out what the shopper needs. Google's example is figuring out what makes a bag good for rainy weather and long trips.
- Microsoft Copilot pulls products from Microsoft Merchant Center feeds, which Microsoft now asks merchants to make ready for the Universal Commerce Protocol.
- Perplexity takes product data from retailers through its free merchant program, alongside the web pages it searches.
- ChatGPT uses two sources: product feeds merchants send through the Agentic Commerce Protocol and the product pages it crawls.
Profound's research on ChatGPT shows how the balance between those two sources has shifted. For most of the past year, crawled pages supplied most of the cards. In a June 2026 sample of 812,190 product cards, 87.3% came from web crawl and 12.7% from feeds. Feeds had the edge where it counted, though. About 99.9% of feed-sourced offers appeared as the first offer on their card.
Then, on July 10, alongside the rollout of ChatGPT 5.6, feed-integrated retrieval jumped from 8.26% to 61.54% in a single day. Our analysis of 687 customers' shopping data found that "brands will not be considered for 65% of product recommendations in Shopping unless you are integrated." If you sell through Shopify or Etsy, your catalog is already connected. Other merchants apply through OpenAI's merchant program.
3. Proof decides which products rank first
None of the engines publish their ranking factors. The clearest view comes from ChatGPT, where Profound compared the products that land at the top of the cards with the rest. Top-ranked cards in the June sample had a median of 787 reviews, compared with 352 for bottom-ranked cards. Product pages that won the first offer slot were more likely than those ranked fourth or lower to include FAQs, Q&As, images, and video. Clear, descriptive product titles that mention a use case also beat clever product names.
What other sites say counts too. Reddit is the most cited domain in ChatGPT Shopping, accounting for about a third of all shopping citations, so the threads where people compare products in a category are part of what the engine reads before it recommends anything.
Google's approach points the same way. When AI Mode works out what makes a bag good for rainy weather, a product whose page and feed say it's waterproof gives the engine something to match. That's the idea behind Merchant Center's new conversational attributes, which let merchants describe products the way people ask about them.
4. The engine picks a store
Every product result also links to a store. When several retailers sell the same product, the engine decides which one gets the shopper. ChatGPT is the only engine measured at scale here. In our March 2026 study of 22.5 million buy offers, Walmart held 8.78% of first-position offers and Target 4.93%, and the top 20 merchants accounted for about 40% of all offers.
The shift to feeds concentrated that further. After July 10, the top 10 merchants' share rose from 22.5% to 41.8%, and the number of distinct merchants ChatGPT referenced fell by more than 20%. If your product makes the card but the link goes to a big-box retailer, you've won the recommendation and handed someone else the sale.
How personal agents like Muse and Instinct shop differently
Everything above describes a shopper who reads the answer and makes the choice. With a personal agent, the agent reads the answers, the catalogs, and your product pages, and the shopper sees only its shortlist, or a single purchase waiting for approval. To see how an agent gets from a request to a recommendation, our team ran 50 shopping prompts through Muse. The sample is small, with one retail category and one Muse account, but four differences from the engines above stand out.
1. The agent knows the shopper
Muse starts with a conversation. It may ask follow-up questions, check the shopper's stored shopping profile, and sometimes search its memory of earlier chats. Instinct also stores what it learns about each user to speed up the next task.
That changes what your product page has to do. An answer engine looks for the best products for a prompt. A personal agent looks for the best product for one person, whose budget, sizes, and preferences it may already hold. If your page doesn't say who the product is for, the agent has nothing to match against that person. For that reason, our research recommends making intended uses, compatibility, dimensions, materials, and limitations easy to find on every product page.
2. The agent searches catalogs as well as the web
For each prompt in the study, Muse ran a browser search and a query of Meta's product catalog at the same time. That catalog is the same one behind Instagram and Facebook shopping, and the two sources returned very different volumes:
- The browser typically visited three domains and returned four products from two merchant sites. The catalog returned around 60 products from 30 merchants.
- Muse then filtered both down to about six product cards, typically two from the catalog and four from the browser.
Each source rewards different work. The browser search is narrow, so the two or three sites it reaches supply most of the final picks, and the reviews, third-party mentions, and clear product pages covered above decide which sites those are. The catalog is wide, so a product that isn't in it misses the larger pool entirely. Instinct follows the same pattern with a different catalog, searching Shopify's product catalog alongside the web. Between the two agents, the Meta and Shopify catalogs are the two feeds to check first.
Instinct has also started recommending products before anyone asks. In late September, it launched Instinct Selections, personalized lists of dining, travel, and shopping suggestions.
3. The agent visits your site the way a person does
Both agents can open your site in a live browser instead of relying only on what crawlers indexed earlier. To work out what's on a page, they read the accessibility tree, the description of the page that the browser builds from your HTML and accessibility labels. Semantic HTML and clear labels make that description more complete. A size selector with no label, or an add-to-cart button built from a generic container, gives the agent less to work with, and an agent that can't finish a task on your site may finish it on another one.
Muse also checks out after the shopper approves the purchase. Retailers have started deciding how much of that to allow. When CNN tested Muse in September, Amazon prevented it from browsing Amazon's store, although Muse could still find Amazon product links through search engines. Target let Muse browse but blocked autonomous clicks at checkout, so the tester had to enter personal details by hand. How far agents can get on your site, and whether they can buy, is now a decision your company should make on purpose.
4. The agent leaves little trace
Answer engines cite sources and send referral clicks that show up in your analytics. Muse and Instinct don't name the sources they visit by default, so their conversations contain no citations. When they browse, they don't identify themselves by name. Their traffic runs through residential connections and looks like ordinary human visits.
The traces that do show up are indirect. Some Muse sessions in our study triggered requests from a self-identified "meta-webindexer" bot, from IP addresses Meta hasn't publicly claimed. In the 20 days after Muse launched, those unverified requests rose 1,313% on retail sites in the US and Canada, where Muse is available, and 827% on sites elsewhere, compared with the 20 days before. Our research treats this as a proxy for Muse traffic rather than a count. Still, the jump was 1.6 times larger where Muse is live.
For measurement, assume your referral reports and bot logs undercount personal agents. Track what you can check directly instead: whether your products are in the catalogs the agents search, whether an agent can use your pages, and what the agents recommend when you run your category's prompts through them yourself.
What to look for in an ecommerce GEO tool
Each stage above, for answer engines and personal agents alike, turns into something an ecommerce GEO tool should track. These are the six criteria we advise you use to compare them:
- Product-level tracking. Brand mentions tell you whether the engines know you. The product cards are where the sale happens, so you need to see individual SKUs.
- Repeated runs on real prompts. With 95% of product titles showing up in fewer than 30% of identical runs, a tool that checks once is, at best, surface-level.
- Merchant and source data. You need to know which store the card sends shoppers to, and which reviews and threads the engine cites.
- Whether it fixes what it finds. Some tools stop at a dashboard or a list of recommendations. Agentic platforms draft the fix themselves. That might be an FAQ section for a product page that doesn't have one, optimized product titles, or a correction request to a review site still listing last year's price.
- Crawler and revenue data that works on your platform. Shopify doesn't give merchants access to server logs, so ask how a tool gets crawler visits and AI-referred conversions from your store.
The 5 best ecommerce GEO tools
The platforms below all track your products in AI shopping results. They differ in which shopping engines they cover, how much detail they provide for each product, and what they do with the data once they have it. None of them can track Muse or Instinct conversations, for the reasons above, so the last column covers what each tool offers for personal agents, based on public documentation as of October 2026.
For a quick side-by-side comparison, check out the table below:
1. Profound
- Product-level shopping data, from placement and pricing to the labels the engine attaches, backed by Profound's research at scale.
- Shopping cards and regular AI answers in one view.
- An orchestrator, AI Marketer (Aim), that finds the issues and opportunities in your shopping data, then builds and runs the agents that carry out the work.
- Crawler and conversion data for Shopify stores, without code changes.
Profound's Shopping feature tracks how your products show up in ChatGPT Shopping, product by product and against your competitors. For each SKU, it shows how visible the product is, where it's placed in the conversation, the images and prices the engine displays, and the citations and keywords tied to each appearance. It also checks how the engine represents the product: whether the price, availability, reviews, and places to buy are accurate, which labels the engine attaches to it (like "best budget pick" or "premium option"), and the sentiment of its descriptions.
At the category level, Shopping shows how often your prompts trigger shopping results, and which wording and intent trigger them, so you know which prompts to go after. Its merchant view shows which retailers control the checkout link on each card, each retailer's share of your category, and gaps in your retailer coverage that are costing you placements. Shopping also connects to the rest of your AI search data, so you can see how a product performs in regular answers as well as on cards, and which patterns on your pages, like FAQs, video, and detailed specifications, drive higher visibility.
Then there's what you do with the data. Profound's Agents are automations that run on everything above. You can build your own in a drag-and-drop builder or start from a template built for ecommerce. The PDP Optimizer, for example, scrapes a product page, extracts the signals that affect ranking, and scores each element against merchant qualification and product ranking benchmarks. A second version, the Research-backed PDP Optimizer, applies the Profound research team's latest findings to recommend specific changes to a product page.
AI Marketer (Aim) decides which agents to run, and when. It's the orchestrator inside Profound, a marketing teammate that works from your AI search data. Every week, Aim brings a ranked list of opportunities, like visibility gaps, sentiment drops, and content gaps against your sitemap, and builds the agents to act on them. You approve the suggestions and review the output before it goes live, whether it's updated product page copy or a new FAQ section.
GR0, an agency that built its business on SEO for ecommerce brands, uses Shopping with clients. "We were on a call with [a major computer hardware company] yesterday and were able to show them where their printers rank in ChatGPT Shopping. They had never seen anything like that before," says co-founder and CEO Kevin Miller. One of GR0's direct-to-consumer health and beauty clients went from about $1,000 to more than $100,000 a month in LLM-attributed revenue.
2. Peec AI
What you'll like:
- Price tracking alongside visibility.
- A quick start from a Shopify connection or a spreadsheet.
Peec AI, known for its brand visibility tracking, launched AI Shopping Analytics in June 2026. It tracks product visibility in ChatGPT Shopping, updated daily by SKU. Peec also shows the price the engine displays for each product, which retailers it recommends most in your category, and how your products compare with competitors'. You connect your catalog by CSV upload, through Shopify, or by API.
Beyond the core metrics, Peec ranks the products gaining and losing placements, shows the dimensions AI uses to compare your products with competitors', and points you to the competitor content that's outranking you, with a note on what to fix. Its shopping data covers ChatGPT only, though, and Peec doesn't document any features for personal agents like Muse or Instinct.
3. Scrunch
What you'll like:
- The DTC share metric, which shows how often you're losing the click to a retailer.
- AXP, for JavaScript-heavy storefronts that AI crawlers can't read well.
Scrunch's Shopping feature tracks ChatGPT and Copilot, with Perplexity listed as coming soon. It reports share of shelf, how often you win the first position, which SKUs are newly appearing or dropping off, and DTC share, meaning how often AI shopping results link to your own site instead of a retailer's. Separately, its Agent Experience Platform (AXP) works through your CDN to serve AI crawlers a simplified, AI-friendly version of your pages. That helps with crawlers that announce themselves. Muse and Instinct browse in a real browser without naming themselves, so they're likely to load your standard pages, and Scrunch doesn't document any tracking of personal agents.
4. AthenaHQ
What you'll like:
- Product tracking across six engines.
- Revenue attribution in Shopify.
AthenaHQ tracks product recommendations at the product, category, or brand level across ChatGPT, Perplexity, and Google AI, whether you sell on your own site, through Amazon, or across other marketplaces. It helps you write AI-optimized product descriptions, reports category share of voice, and tracks the path from AI recommendation to site visit to conversion. Its copilot, Ask Athena, answers questions about your data in plain language.
The Shopify integration is where AthenaHQ goes furthest for ecommerce. According to AthenaHQ's documentation, it imports your Shopify products so you can generate and publish optimized titles, descriptions, metadata, and FAQs, and it publishes articles from Athena's Content Hub directly to your Shopify blog. Since Instinct searches Shopify's catalog, cleaner product data there may help with personal agents too, but AthenaHQ doesn't document any way to see how Muse or Instinct treat your products.
5. Evertune
What you'll like:
- The competitive pricing view.
- Scores built on large samples of responses.
Evertune added Shopping Intelligence in January 2026. It tracks how often product searches in your category trigger ChatGPT shopping results, how often your brand appears when they do, which products and retailers ChatGPT recommends, and how your prices compare with competitors' according to AI models. All of this sits on top of Evertune's AI Brand Index, which prompts each engine thousands of times to score your visibility and position. Its shopping coverage stops at ChatGPT, and it doesn't document any features for personal agents.
From AI shopping to your next marketing move with Profound
When a shopper delegates a purchase, the agent reads your product pages, your catalog listings, and your reviews on their behalf, and the shopper may only ever see the shortlist. Your brand, distribution, and customer relationships still count. You can't assume, though, that an agent understands what makes your products worth choosing, or that it will take the route to purchase you designed.
Getting agents to understand and buy from you is work that crosses teams. Catalog coverage probably sits with commerce, page structure and checkout with product and engineering, and product positioning with marketing. Each team can fix its own piece, but an internet where more and more visitors are agents needs one executive-level owner. The companies that start testing these new front doors now give themselves more time to learn what works than competitors who wait.
That's what we're building Profound for: understanding how AI sees your business, and giving marketing teams the ability to act on it. The data shows how the engines describe and rank your products and which sources and retailers they rely on. Agents draft the fixes, from product page copy to FAQs. And Aim ranks the opportunities in that data and runs the agents that act on them, with your approval. As Profound CEO James Cadwallader said at Aim's launch, "marketing teams don't need another dashboard. They need to know what to do next."
Step into the agentic era of your marketing. Book a demo to learn more.
Ecommerce GEO tools FAQs
What is ecommerce GEO?
Ecommerce GEO is the work of getting your products recommended by AI engines like ChatGPT, Gemini, Google's AI Mode, Perplexity, and Copilot, and by personal agents like Muse and Instinct. With answer engines, it covers two places where shoppers meet your products. One is the text answer a shopper gets while researching; the other is the row of product cards the engine shows when a prompt has buying intent, each with a product, a price, and a store. With personal agents, it also covers whether the agent can find your products in its catalogs, understand who they're for, and complete a purchase on your site.
What are personal agents like Muse and Instinct?
They're AI assistants that carry out tasks for a person instead of only answering questions. Meta's Muse runs in its own app and in WhatsApp, and Instinct is an invite-only assistant that people text or call. Both can compare products, book travel, and make purchases, and both remember what they learn about each user's preferences.
For ecommerce, the difference from ChatGPT or AI Mode is who makes the choice. An answer engine shows a shopper a set of products and lets them pick. A personal agent narrows the options itself, often to a few product cards or a single purchase waiting for approval, so a product the agent never considers never reaches the shopper.
How do I get my products recommended by Muse and Instinct?
Get into the catalogs they search, make your site usable by an agent, and spell out who each product is for. Muse queries Meta's product catalog, the same one behind Instagram and Facebook shopping, so set up or audit your catalog in Meta Commerce Manager. Instinct searches Shopify's product catalog alongside the web.
Next, open your key product pages in a browser and check that an agent could pick a variant, add it to the cart, and reach checkout, which depends on semantic HTML and labeled buttons and fields. Then put the details an agent compares against a person's needs, like intended uses, compatibility, dimensions, materials, and limitations, where they're easy to find. The work on reviews, third-party mentions, and descriptive titles still applies, because Muse's browser search draws on the same web pages the answer engines read.
Can I see when a personal agent visits my store?
Not reliably. Muse and Instinct don't identify themselves when they browse, and their traffic comes through residential connections that look like human visitors. Their conversations don't cite sources either, so there's no referral click to attribute.
You can track proxies and inputs instead. Profound's research used requests from an unverified "meta-webindexer" bot as a signal of Muse activity, since those requests grew much faster in the countries where Muse launched. Alongside that kind of signal, check whether your products appear in the Meta and Shopify catalogs, and run your category's shopping prompts through the agents yourself to see what they recommend and where they buy.
Should I block personal agents from my store?
Only if you have a specific reason and know what you'd give up. When CNN tested Muse in September, Amazon prevented it from browsing Amazon's store, and Target let it browse but blocked autonomous clicks at checkout. Both choices keep the retailer in control of the checkout, at the cost of making the agent's job harder, and an agent that can't buy from you still has a shopper waiting for an answer.
Blocking is also harder than it sounds, since the agents don't name themselves when they browse. If you want a middle path, Target's approach is one option: let agents read your pages and products, and keep a human step at checkout. Whichever way you go, make the call with commerce, marketing, and security together, so it's a decision instead of a side effect of a bot rule.
How do I get my products into ChatGPT Shopping?
Connect your catalog first, then give the engine reasons to rank your products. If you sell through Shopify or Etsy, your catalog is already connected to ChatGPT. Large retailers like Target, Best Buy, and Wayfair have integrated directly through the Agentic Commerce Protocol. Everyone else applies through OpenAI's merchant program, which accepts feeds by SFTP, API, or through supported feed providers, and OpenAI has said a self-serve merchant portal is on the way. Shopping is currently live for ChatGPT users in the US.
Getting in is only the first step, because ranking follows. Make sure your robots.txt lets OpenAI's crawlers reach your product pages. Then work on what separates first-position products from the rest in Profound's data: more reviews, FAQs and Q&As on the product page, images and video, and descriptive titles that mention a use case.
Can shoppers buy directly inside ChatGPT?
No. Since March 2026, ChatGPT has sent shoppers to the merchant's own checkout, which opens in a browser inside ChatGPT.
Do I need a product feed to show up in AI shopping results?
For ChatGPT, in most cases, yes. On July 10, 2026, the share of ChatGPT shopping results pulled from integrated feeds jumped from 8.26% to 61.54% in a single day. Profound's analysis of the shift found that "brands will not be considered for 65% of product recommendations in Shopping unless you are integrated." Even before that, feed-sourced offers took the first slot on their product card about 99.9% of the time. Google's AI Mode and Gemini also draw on product data from Merchant Center.
Which products should we optimize first?
Start with the products in categories where shopping results trigger most often, and where you already have demand. From there, check three things for each product. First, confirm the product is in your feeds with complete attributes, including the Meta and Shopify catalogs that personal agents search. Next, compare its product page with the product holding the first position for the same prompts, including review count, FAQs, images, and title. Finally, check which store the card links to. If the card sends shoppers to a retailer instead of your site, decide whether that's a channel you're happy to win through or a gap to fix on your own site.
Is Google Merchant Center's AI performance report enough?
No, but turn it on anyway, since it's free. AI performance insights show your share of voice in AI Mode and AI Overviews compared with the competitors you define, how often different types of searches come up, and how many of your products appear for top terms. It also splits queries into three shopping stages, Discovery, Evaluation, and Ready to Buy, which helps you see where in the journey you're losing ground.
Its limits are what make it insufficient on its own. The report covers only AI Mode and AI Overviews, only organic listings (paid ads aren't included), and only English-language queries for Merchant Center accounts in the US, Canada, Australia, India, and New Zealand. It won't show you ChatGPT, Copilot, or Perplexity, and it won't tell you which third-party pages the engines cite about your products. Use it for your Google view, next to a tool that covers the other engines.