The Agentic Commerce Tipping Point

For most of 2024, “AI shopping” felt like an experiment – something to watch, but not yet a core part of ecommerce. Fast forward to mid-2025, and the picture looks very different. From surging traffic and conversion rates to industry-wide adoption of open protocols, there’s mounting evidence that agentic commerce has arrived.

This post dives into the data and trends proving that AI shopping assistants like ChatGPT aren’t just hype – they’re becoming a meaningful revenue channel. We’ll explore how consumers are using these tools, why their performance has merchants and platforms jumping on board, and what it all means for brands that want to stay ahead.

1. Consumers are embracing AI shopping assistants

Shoppers are rapidly getting comfortable asking AI agents for purchase advice. OpenAI notes that “every day, millions of people use ChatGPT to figure out what to buy” on its merchant overview page. In fact, one OpenAI/Harvard analysis estimated that roughly 2% of all ChatGPT queries involve shopping – on the order of tens of millions of shopping prompts per day. These aren’t just trivial questions either. Users ask things like “recommend a good laptop under $1,000” or “I need a gift for a 5‑year‑old who loves science,” which is exactly the kind of intent that used to go straight into Google or Amazon.

Even more eye‑opening: consumers are not shy about using generative AI for shopping. In an August 2025 survey covered in Salesforce’s holiday trends report, nearly 60% of U.S. shoppers said they had used a generative AI tool to help with online shopping (ChatGPT, Bard, retail‑site copilots, etc.). The key point is that more than half of consumers have now tried an AI assistant in their purchase journey.

Trust is also surging. Salesforce’s research found that 86% of shoppers who use AI for shopping say they trust AI product recommendations, up from just 46% a few months prior, in its Emerging LLM Search for the Holidays report. Once shoppers try AI helpers and see relevant results, they tend to believe those results. That trust is the foundation for agents not just suggesting products, but actually driving conversions.

All of this paints a clear picture: consumers – especially younger, tech‑savvy cohorts – are rapidly warming up to the idea of letting AI guide their shopping. The days of being wary of a “robot personal shopper” are fading; many shoppers are already treating ChatGPT or similar agents as a natural place to start when they have a need or a question.

2. AI-driven traffic is converting (way better than you think)

It’s not just about engagement – AI‑assisted shopping is turning into real sales. Early data indicates that traffic from AI assistants is highly valuable, often outperforming traditional channels on conversion.

Salesforce’s analysis of millions of shopping sessions in 2025 found that conversion rates from AI assistant channels were over 700% higher than from social media traffic, and 200% higher than from standard search engine or direct traffic, again in the Emerging LLM Search for the Holidays report. In plain English: users coming in via an AI recommendation are far more likely to buy than those coming from a Facebook ad or even a Google search result. This makes sense – someone who starts with a specific query to an AI (for example “best noise‑canceling headphones under $300”) is a highly motivated, qualified shopper by the time they click a result.

The volume of AI‑driven traffic is growing quickly. Salesforce noted that traffic from AI assistants grew 119% year‑over‑year in the first half of 2025. And while it’s starting from a small base, it’s climbing fast.

Independent reporting backs this up. A Modern Retail analysis found that by mid‑2025, ChatGPT was driving over 20% of referral clicks to Walmart and Etsy and around 15% to Target and 10% to eBay within those retailers’ referral traffic mix, based on data from analytics firm Similarweb in “ChatGPT is now 20% of Walmart’s referral traffic — while Amazon wards off AI shopping agents”. Inside the “referral” category, ChatGPT has already jumped to become one of the largest sources of new shoppers for these brands.

Note: Referral traffic from ChatGPT is still a small slice of total visits (most traffic is still direct, search, or paid). But the speed of growth is what’s remarkable. Climbing to ~20% of referral clicks in a matter of months shows how quickly an AI assistant can insert itself into the customer journey. Shoppers are increasingly clicking through links inside AI chat responses, effectively pushing retailers into the conversation at the point of discovery.

Perhaps the biggest validation: Salesforce’s 2025 holiday forecast estimates that AI and shopping agents will drive 21% of all global holiday orders, accounting for around $263 billion in sales, again in Emerging LLM Search for the Holidays. Even if that forecast ends up high or low, it signals a belief that agent‑driven commerce is capturing demand at scale right now.

Why do these AI referrals convert so well? The likely reasons: they capture high‑intent queries, they present very tailored options, and they reduce the friction of decision‑making. By the time a shopper sees “the answer” (a curated product suggestion) in ChatGPT, they’ve essentially skipped multiple steps of the traditional funnel. It’s as if the top and middle of the funnel have been compressed into one interactive dialogue, and only the bottom (checkout) remains.

The takeaway: traffic coming from AI assistants isn’t casual window‑shopping traffic – it’s more like someone walking into your store and saying “the AI sent me and I’m ready to buy.” Few channels deliver that level of intent.

3. Industry heavyweights are all-in on agentic commerce

The rapid consumer uptake and strong conversion metrics haven’t gone unnoticed. Over the last year, we’ve seen a cascade of moves from major platforms and providers to enable AI‑mediated shopping.

OpenAI and ACP

In September 2025, OpenAI announced Instant Checkout in ChatGPT, turning ChatGPT from a recommendation engine into a purchase channel. The announcement post, “Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol”, laid out the vision:

• U.S. users can buy directly from Etsy sellers in chat.
• Support for over 1 million Shopify merchants is “coming soon,” including brands like Glossier, SKIMS, Spanx and Vuori.
• Product recommendations are ranked by price, availability, quality, seller prominence and checkout ease, not paid placement.

This rollout is powered by the open‑source Agentic Commerce Protocol (ACP), which OpenAI co‑developed with Stripe. ACP defines a standard way for AI agents to talk to merchant backends: creating carts, calculating tax and shipping, passing delegated payment tokens, and checking order status. OpenAI intentionally open‑sourced ACP so that any PSP or platform can implement it and plug their merchants into agentic commerce.

Payments: Stripe and PayPal

On the payments side, Stripe was an early partner. It built support for ACP’s delegated payment spec so ChatGPT can send one‑time payment requests to Stripe without ever seeing raw card data, as described both in OpenAI’s Instant Checkout post and Stripe’s announcement.

Then in October 2025, PayPal announced it would adopt ACP and integrate its vast merchant base into ChatGPT. In “OpenAI and PayPal Team Up to Power Instant Checkout and Agentic Commerce in ChatGPT”, PayPal said it would:

• Let tens of millions of PayPal merchants tap into ChatGPT as a shopping and checkout surface.
• Act as an aggregator, so many merchants can participate with no custom development, by routing orders and payments behind the scenes via ACP.
• Eventually enable checkout with PayPal directly inside ChatGPT, leveraging PayPal’s user account network.

In practical terms, this means millions of products that already sit in PayPal’s orbit will become discoverable and purchasable through AI agents, without each merchant having to build a bespoke integration.

Platforms: Salesforce, Shopify, and others

Commerce platforms are also moving. Salesforce Commerce Cloud announced that it would support ACP and work with Stripe and OpenAI so merchants can enable Instant Checkout through Salesforce. Their press release, “Salesforce Announces Support for Agentic Commerce Protocol with Stripe”, frames ACP as a way for retailers to plug into ChatGPT with minimal friction.

Salesforce has also been investing in its own AI tools for merchants. The “Agentforce” toolkit and AI shopping assistants for Commerce Cloud are highlighted across its newsroom, where Salesforce cites research that 48% of shoppers who already use AI for shopping say they are open to having an AI agent make a purchase for them. That stat appears repeatedly in the company’s AI commerce materials.

On the storefront side, Shopify is directly partnering with OpenAI. Shopify’s merchants are among the first wave slated to support Instant Checkout in ChatGPT, via Shopify’s integration with ACP and the OpenAI product feed and checkout specs. For many brands on Shopify, ChatGPT could effectively become an additional sales channel alongside their online store and marketplaces.

Search: Google fights back

After years of consumers defaulting to Amazon for product queries, Google is using AI to claw back share. It has integrated generative AI overviews into search results and is pushing the Gemini assistant as a conversational way to search and shop.

It seems to be working, especially with younger shoppers. A Morgan Stanley survey reported by Business Insider in “Gen Z shoppers are loving Google. That could be bad news for Amazon.” found that in just six months:

• Among 16–24‑year‑olds who already knew what they wanted to buy, the share who went to Google first rose from 21% to 30%.
• The share who went to Amazon first fell from 41% to 34%.

Morgan Stanley explicitly linked the shift to Google’s experiments with AI‑powered search and shopping interfaces, which Gen Z users find more engaging than traditional link lists.

Marketplaces: Amazon protects its turf

No discussion of product discovery is complete without Amazon, but Amazon has taken a different approach.

As Modern Retail details in its piece “ChatGPT is now 20% of Walmart’s referral traffic — while Amazon wards off AI shopping agents”, Amazon has:

• Blocked many AI web crawlers from accessing detailed product data.
• Limited how external AI agents can index its catalog in real time.

The result is that ChatGPT’s shopping answers often omit Amazon listings, instead surfacing Walmart, Target, Etsy, and other retailers that allow their sites to be crawled and indexed by AI. Amazon doesn’t want to lose control of customer traffic (and ad impressions) by letting a third‑party agent sit in front of its catalog.

Instead, Amazon is building its own agentic commerce experience: Amazon Rufus. Rufus is a shopping assistant embedded directly in Amazon’s app and website. According to the same Modern Retail reporting, Rufus had already handled over half a billion customer questions by the end of 2024, and Amazon is starting to insert ads into Rufus’s answers. Amazon clearly believes in AI‑mediated shopping – it just wants that experience to live inside its own walled garden.

4. What it means for brands and retailers

All these trends point to one conclusion: AI‑driven shopping is not a future concept; it’s here now, and growing fast. So how should you, as an ecommerce brand or retailer, respond? A few implications are clear.

A new discovery channel is up for grabs

For years, the playbook was to optimize for Google and Amazon because that’s where discovery happened. Amazon still commands a majority of product searches: Jungle Scout’s Amazon Statistics 2024 aggregates multiple studies and puts the share of U.S. consumers who start product searches on Amazon at 50–60% depending on category.

But AI assistants are creating an alternative gateway to consumers. If a shopper asks ChatGPT for “the best budget smart TV” and your product fits – but your data isn’t in the feed – your brand simply doesn’t exist in that moment. On the flip side, if you are integrated, you can appear in results alongside (or even instead of) Amazon or big‑box competitors.

This is a rare chance to win visibility in a new channel before it gets crowded. It’s telling that ChatGPT is already sending notable traffic to Walmart, Target and Etsy (per the Modern Retail article above). It’s inserting those retailers into journeys where Amazon might previously have dominated. Smaller DTC brands can ride the same wave if they play their cards right.

Relevance and data quality trump ad spend

Unlike search results or marketplace listings, you can’t pay an AI agent to show your product – at least not today. OpenAI’s merchant docs and Instant Checkout announcement emphasize that product rankings in ChatGPT are based on “price, availability, quality, seller prominence, and checkout ease”, not paid placement, on the ChatGPT merchants page and in the Instant Checkout post.

That means the playing field is, in some sense, more level. The agent will recommend whichever products best match the user’s request and context. To be in that consideration set, your product data needs to be complete, structured and trustworthy.

Think of this as the new SEO – some people even call it “generative AI optimization.” Instead of keyword stuffing and backlinks, it’s about:

• clean titles and descriptions that reflect actual use cases,
• rich attributes (materials, dimensions, compatibility, fit),
• accurate prices and availability, and
• trustworthy ratings and reviews.

If your feed lacks key info, the model might exclude you or even hallucinate details. If your prices or stock are stale, the agent will favor a competitor with fresher data. Investing in clean, rich product data is no longer just for ads and comparison engines – it now directly affects whether AI agents recommend your brand or someone else’s.

The “AI shelf” is narrow – but meritocratic

Chat‑based recommendations typically show a very limited number of options (often just 1–3). There’s no second page of results. That means competition for those spots is fierce.

But the winners aren’t determined by who pays for an endcap or who has the fanciest website – it’s determined by who best meets the query and offers a good value. OpenAI explicitly calls out price, availability, quality, seller prominence and checkout ease as ranking factors in its merchant docs and Instant Checkout post. The agent doesn’t want to show overpriced, out‑of‑stock or unreliable products, because that would degrade the user experience.

This puts pressure on merchants to be truly competitive on the fundamentals. It also means brands with great products and customer satisfaction have a chance to shine even if they don’t have the biggest advertising budget. In a way, the AI agent behaves like a super‑intelligent personal shopper: it will put the best options on the shelf, not the highest bidders.

Multi-merchant carts and new logistics demands

Today, ChatGPT’s Instant Checkout focuses on straightforward, single‑item purchases, but OpenAI has signaled that multi‑item, multi‑merchant carts are on the roadmap in the Instant Checkout post and ACP discussions. In an agent‑driven scenario, a shopper might say:

“I need decorations, snacks and games for a kids’ party this weekend.”

The agent could respond with a cart that pulls items from three different stores, each fulfilled separately.

As a merchant, that means you may start seeing orders that originate outside your usual web storefront, via different entry points (ChatGPT, voice assistants, third‑party agents). You still own fulfillment and customer service – ACP is explicit that the merchant or PSP is responsible for order lifecycle and support, as described in the ACP docs and PayPal’s integration announcement – but you have to be ready to treat AI agents as another front‑end.

Practically, that implies:

• Your systems need to handle headless checkout flows where the front‑end is an AI agent, not your website.
• Inventory and pricing must sync quickly to all channels, including your agent‑facing feed.
• Your support team needs to recognize orders that came from ChatGPT (“I bought this through ChatGPT, what’s the status?”) and be able to look them up.

Don’t wait – early movers are gaining data and traction

The train has left the station, but it’s still early enough to get on board. Brands that embraced things like Google Shopping early, or joined Amazon in its wild‑west era, reaped outsize rewards. We’re at a similar inflection point.

Those experimenting with ChatGPT commerce in 2025 will:

• learn faster about what attributes and offers perform in agentic channels,
• accumulate reviews and performance history that models may use as soft ranking signals,
• and influence how specifications and best practices evolve.

If you wait until agentic commerce is “proven” by all your competitors, it will be harder to catch up – not just in integration, but in the data exhaust (clicks, conversions, satisfaction signals) that models accumulate about your catalog.

The good news is that getting started doesn’t require a massive overhaul. It mostly means:

• ensuring your product feed is agent‑ready (mapped to OpenAI’s Product Feed Spec and similar schemas), and
• enabling a suitable checkout integration using the Agentic Checkout Spec via your PSP or platform.

It does require priority and cross‑functional focus, because it touches product, engineering, marketing and ops. But it’s not a multi‑year replatforming.

5. How Pesto can help

Navigating this new channel can feel overwhelming – there are specs to implement, feeds to clean up, and constant changes to monitor. This is where Pesto fits in. We built Pesto to be the bridge between your store and the AI shopping agents of the world.

Feed optimization and compliance

Pesto connects to your product catalog (whether you’re on Shopify or any other platform) and transforms it into a clean, agent‑ready feed. We map your data to OpenAI’s Product Feed Spec so you can include the required fields (price, availability, GTIN, etc.) and normalize attributes to standard vocabularies.

This dramatically increases the chances that ChatGPT and other agents see your products as eligible, relevant and reliable. We continuously scan for issues – if a product is getting dropped due to a data error, or an attribute looks malformed, we alert you and can even auto‑fix certain problems.

Real-time updates, zero hassle

In agentic commerce, freshness is non‑negotiable. Pesto sets up a live feed that updates every 15 minutes (or faster) with inventory and price changes, so the information AI agents have about your products is always up to date. No more manual CSV uploads or worrying that a customer will order via ChatGPT something that just went out of stock.

We handle the integration with ACP‑compatible endpoints so that your latest data is always flowing to the assistant.

Seamless checkout integration

Implementing the Agentic Checkout Spec on your own can be tricky – you need to handle cart creation, tax/shipping calculations and delegated payment tokens securely.

Pesto simplifies this by interfacing with your existing ecommerce platform or order system. You continue to use your payment processor (Stripe, PayPal, etc.), and Pesto ensures that the handoff from the AI agent to your backend is smooth.

For example:

• If you use Stripe, ACP’s delegated payment support is already implemented on Stripe’s side, and Pesto can wire it into your existing order flows.
• If you use PayPal, Pesto can leverage PayPal’s ACP integration so you can join ChatGPT commerce without custom work on your own infrastructure.

The result: you can start accepting orders from ChatGPT with minimal development work, and without changing your fulfillment or CRM processes.

Monitoring and analytics

As you start getting agent‑driven orders, Pesto provides visibility into what’s happening:

• Which products are being recommended most?
• How many carts and checkouts are originating from ChatGPT or other agents?
• Are any transactions failing due to missing data or integration errors?

We log and surface this data so you can iterate – maybe you discover that a top‑selling SKU isn’t showing up because it’s missing a size attribute, or that agentic buyers favor a certain category. These insights let you refine your strategy and feed, just like you’d refine keywords or bidding in search.

In short, Pesto’s mission is to handle the messy middle of making your catalog AI‑friendly. We keep track of evolving requirements (if OpenAI updates the spec or Google’s AI search needs a certain schema, we handle it) so you don’t have to. You focus on what you sell and how you sell it, and we make sure AI agents can discover, recommend and transact with your products effortlessly.

Agentic commerce is hitting its tipping point. Shoppers are using it. The tech giants and platforms are enabling it. The only question is: will your products be part of it?

By treating your data and integrations as first‑class priorities – and by leveraging tools like Pesto to accelerate the process – you can position your brand to ride this wave rather than chase it.

The next era of ecommerce won’t be won by those who sit back and see what happens. It will be won by those who recognize an inflection point and act. In 2025, that inflection point is here. Brands that get their feeds and systems agent‑ready now will be the ones capturing the new AI‑driven demand (and revenue) in the years ahead.

Don’t be the late adopter in your category. The commerce AI revolution is no longer theoretical – it’s happening in real time, and there’s real money on the table.

Now is the moment to make sure you’re part of the story. Your customers’ new personal shoppers (the AI agents) are waiting to meet your products – give them the data and access they need, and you just might find yourself with a front‑row seat on the AI commerce boom.