If you have been feeling that product discovery is getting less predictable, you are not imagining it. chatgpt shopping agents are starting to sit between shoppers and stores, and they do not browse like humans do. They ask, compare, decide, and increasingly they buy.
I have watched a few platform shifts in my career, search to social to creator commerce, and this one feels faster. The reason is simple: agents compress the funnel into a conversation.
Summary / Quick Answer
Shopping agents are moving from “help me research” to “help me purchase.” Amazon’s Rufus, Perplexity’s Buy With Pro, Google’s Buy for Me, and ChatGPT Instant Checkout all point to the same future, autonomous shopping inside the interface where intent is formed. Rufus dominates inside Amazon, Perplexity blends research with one-click checkout, Google turns price tracking into a purchase trigger, and ChatGPT is becoming a neutral front door for millions of merchants via the Agentic Commerce Protocol.
For brands, this changes how you win visibility. Structured product data, reliable policies, and fast fulfillment matter more than clever copy. Think of it as SEO for agents, not just for people.
What shopping agents are actually doing now
Most marketers still picture a chatbot that answers questions. That mental model is already outdated.
A modern shopping agent does three things in sequence. First, it interprets intent in plain language. Second, it gathers external evidence (reviews, specs, policies, price history). Third, it proposes, or even completes, a transaction. That third step is what makes this “agentic commerce” rather than fancy search. McKinsey frames this as a shift from recommendation engines to agents that participate in the whole journey.
Here is the practical difference I keep explaining to founders:
Old flow (search era)
You search, you click, you compare ten tabs, you decide, you buy.
New flow (agent era)
You describe a problem, the agent narrows options, confirms constraints, and checks out.
Why does that matter for growth? Because the “decision surface” moves. It no longer lives on your product page alone. It lives in the agent’s ranking logic. Those rankings are fed by structured data, merchant reliability signals, and the agent’s own guardrails.
Three signals are showing up across platforms:
- Machine readable catalogs. If your variants, shipping windows, return rules, and inventory are easy to parse, you get picked more often. Stripe and OpenAI are basically spelling this out in the Agentic Commerce Protocol docs.
- Trust and policy clarity. Agents reward low ambiguity. If your return policy is buried or full of exceptions, the agent tends to avoid you.
- Context matching beyond keywords. Agents map “use cases” not just nouns. “Shoes for rainy Lisbon in March” is a category plus a scenario. Google’s AI Mode shopping is built around this.
I have seen this kind of signal shift before. When Google started favoring page speed and structured snippets, brands that took it seriously early got a multi year edge. Same pattern, new interface.
chatgpt shopping agents, perplexity buy with pro, amazon buy for me agent compared
Let’s put the big players on the same map. I am not chasing hype here, I am looking at who owns intent, who owns checkout, and who owns the data exhaust.
| Platform | Where intent starts | How it decides | How it buys | Brand risk |
|---|---|---|---|---|
| Amazon Rufus | Inside Amazon app | RAG over Amazon catalog plus web sources | Adds to cart, routes to Amazon checkout | Medium, Amazon keeps shopper |
| Perplexity Buy With Pro | Perplexity search chat | Web wide comparison with merchant program boost | One click “Buy With Pro” on approved items | High, Perplexity becomes the storefront |
| Google Buy for Me | Google Search and AI Mode | Price tracking plus context filters | Agentic checkout via Google Pay | Medium, depends on Merchant Center setup |
| ChatGPT Instant Checkout | ChatGPT conversation | Relevance ranked, unsponsored results | Instant Checkout for ACP merchants | High, ChatGPT becomes the front door |
A few platform notes worth knowing.
Amazon Rufus is the default agent inside the biggest marketplace.
Amazon says more than 250 million customers used Rufus in 2025, with usage and interactions up sharply year over year. What stands out to me is not the number, but the behavior. Rufus nudges shoppers with “Help Me Decide” style guided comparisons and answers scenario questions, not just spec questions. Amazon controls the catalog and checkout, so Rufus is a conversion accelerator for Amazon first, then for sellers.
Perplexity Buy With Pro is closer to a neutral shopping layer.
Perplexity launched Buy With Pro in late 2024, with one click checkout and free shipping on select partner items. What is strategic here is the merchant program. Approved retailers get higher placement and richer product display. In practice, that is a new kind of retail media, except paid placement is disguised as “better data.” If you sell on Shopify, Perplexity’s Shopify feed access makes integration straightforward.
Google’s Buy for Me makes price tracking a purchase trigger.
Google previewed and rolled out agentic checkout in 2025. Users track a product, set a target, and when it hits, Google offers to buy on their behalf via Google Pay. For many categories, that turns “I’m considering” into “done.” If you already use Merchant Center feeds well, you are halfway there. If you do not, you are invisible in agentic mode.
ChatGPT Instant Checkout is the wildcard.
OpenAI launched Instant Checkout in September 2025, powered by the Agentic Commerce Protocol with Stripe, letting U.S. users buy from Etsy and soon Shopify merchants without leaving chat. OpenAI states results are organic and not sponsored. That matters, because it positions ChatGPT as a “trust first” discovery layer. From a brand angle, that is both exciting and scary. Exciting because it is a new channel with massive intent gravity. Scary because you do not own the surface where the decision happens.
If I had to summarize the fight in one line, Amazon wants you to shop on Amazon, Google wants to mediate every price sensitive purchase, Perplexity wants to be your research plus checkout engine, and ChatGPT wants to be the universal shopping companion. That is not a forecast, it is already happening.
Integration approaches and what agents need from brands
Most founders ask me, “So do I need to build my own agent?” Sometimes yes, but usually not first.
The first win is getting your product and policy layer “agent ready.” This is where Agentic Commerce Protocols becomes a useful mental model. ACP is essentially the Stripe style standard for letting agents talk to your checkout safely. Even if you are not on ACP yet, the direction is clear.
A practical readiness stack looks like this:
1. Product data that answers real use cases.
Agents do not just match “running shoes.” They match “wide toe box for marathon training in wet weather.” That requires clean attributes, rich variant logic, and honest constraints. If you are on Shopify, your feed is already a core input to ChatGPT, Perplexity, and Google’s agentic surfaces. Perplexity AI+1
2. Policy data that is unambiguous.
Refund windows, shipping zones, subscription cancellation, warranty paths. Put them in structured, plain language blocks. I have seen agents “play it safe” and choose a competitor just because a return rule was unclear.
3. Inventory and price freshness.
Agentic systems update fast. Google talks about high frequency data refresh behind Buy for Me. If your feed is stale, your product drops out of consideration.
4. Trust signals that travel.
Think reviews in a canonical form, not just star ratings. Rufus, for example, summarizes reviews and uses them in its recommendations. If you can syndicate reviews through your feed, do it.
There is a hidden upside here. When your catalog is structured well, you also improve your classic SEO, your on site filters, and your ad relevance modeling. One cleanup, multiple benefits.
Market impact, and why acquisition gets weird
The part that makes me sit up is not the tech, it is the incentive shift.
Traffic from generative AI to retail sites has spiked in 2025, and those visitors show stronger intent. But as agents take more of the journey inside their own UI, direct site traffic can drop. That is the disintermediation risk BCG and others keep warning about. You lose the click, you lose the pixel, you lose the chance to upsell.
Here is what changes for growth levers:
SEO becomes GXO (generative experience optimization).
Instead of ranking ten blue links, you are ranking inside an agent’s shortlist. Your snippets are product feeds plus policy clarity plus brand reliability. You can still win by content, but content has to map to intent clusters, not just keywords.
Retail media starts to fragment.
If Perplexity rewards merchant program members, if ChatGPT stays unsponsored, if Amazon stays closed, paid influence becomes platform specific. Your media mix gets more complex.
Brand loyalty softens.
Agents focus on utility. If you are “good enough” price wise and policy wise, you get picked even if the shopper never heard of you. Great for challengers, dangerous for sleepy incumbents.
This is why I keep telling teams to watch agent surfaces the way we used to watch SERPs. Every time a platform changes its “default path to purchase,” marketing rewires around it. We are in that phase right now.
For a deeper adoption view, the trendline is in my Market Trends breakdown, it is moving quickly and unevenly by category.
How I would prepare an ecommerce brand for B2A
If your buyers outsource more decisions to agents, you have a new customer type, not human, but delegated. That is Business to Agents, or B2A. I go into the full playbook in The Complete Guide to B2A Commerce [Business to Agents]: Preparing Your Ecom Brand for the AI-First Era, but here is the short version I would act on this quarter.
Step 1. Audit your “agent readability.”
Pick three products that matter. Ask ChatGPT, Perplexity, and Google AI Mode the same scenario query. Notice what they cite and what they miss. If they miss key attributes, that is a feed problem.
Step 2. Produce an agent first product feed.
This is not glamorous work. It is naming variants clearly, filling attributes consistently, and tagging use cases. Most teams underinvest here. The ones that do invest will be recommended more often.
Step 3. Publish policy pages like they are API docs.
Short headings, plain language, consistent rules. If a human can misunderstand it, an agent definitely will. ACP style thinking helps, even if you are not integrated yet.
Step 4. Instrument “agent sourced” demand.
You may not get full referrer clarity, but you can track coupon codes, landing path fingerprints, and customer surveys that ask “where did you start?” Treat it like early TikTok attribution, imperfect but directional.
Step 5. Decide where you want to win.
Amazon sellers should double down on Rufus ready listings. DTC brands might prioritize ChatGPT and Google surfaces. Perplexity is interesting for higher consideration categories where research matters.
The teams that win will be the ones who accept that an agent is now part of the funnel. Not a threat to deny, just a new stakeholder to serve.
Q&A
Q: Are shopping agents replacing marketplaces or just adding a layer?
Both, depending on the platform. Rufus strengthens Amazon’s marketplace by boosting conversion inside it. ChatGPT and Perplexity add a layer above many merchants, which can divert traffic away from individual stores.
Q: What is the fastest way to show up in agent recommendations?
Make your catalog and policies easy to parse. Clean attributes, accurate inventory, clear shipping and returns, plus strong reviews. Agents tend to avoid ambiguity and stale data.
Q: Will paid ads matter less when agents buy for people?
Paid influence will not disappear, but it will relocate. Expect more platform specific boosts (Amazon style), data driven partner programs (Perplexity), and feed based competitiveness (Google, ChatGPT).
Conclusion
Agents are turning shopping into a conversation that ends with a checkout, often without a browser tab in sight. Amazon Rufus, Perplexity Buy With Pro, Google Buy for Me, and ChatGPT Instant Checkout are not separate experiments. Together they describe the new architecture of commerce.
If you run an ecommerce brand, your job now includes serving two minds, the human buyer and the agent helping them. Start with data readiness, policy clarity, and feed freshness. Then choose the surfaces where your category will be decided.
If you want to go deeper on the mechanics, revisit Agentic Commerce Protocols and the adoption curve in Market Trends. I suspect we will look back at late 2025 as the moment shopping became agent native. The brands that treated it seriously early will have the calmest growth charts later.
