"Will AI agents skip my website?" For some purchases, yes. On October 23, 2026, payments publisher PYMNTS reported that 40.2% of US consumers would send an AI shopping agent to a large marketplace instead of a brand's site, even with price, shipping and returns equal. Only 16.1% picked the brand site. Agents still pick brand sites that state prices and policies clearly and give a reason to buy direct.
- The habit gap: 59.7% of recent AI-assisted purchases were completed on Amazon; brand websites or apps took 6.3% (PYMNTS).
- Checkout is not the lure: of marketplace shoppers, 53.9% would buy direct for a lower price and only 12.1% for an AI-handled checkout.
- Research beats buying: a separate vendor study found 61.4% of US online shoppers use AI to research purchases, but 6.8% have let an agent check out.
- Services are different: there is no cart, but agents already shortlist providers and link people to booking pages.
What changed this week
PYMNTS Intelligence surveyed 2,191 US consumers in August and published the results on October 23, 2026, drawn from its October report "From Prompt to Purchase: Who Wins the AI Shopper?". The setup was blunt: an AI assistant is buying an identical item, and every seller offers the same price, shipping and returns. Where should it buy?
| Where shoppers would send the agent | Share |
|---|---|
| Large online marketplace | 40.2% |
| Department store or large retailer | 21.8% |
| Let the AI decide | 21.8% |
| The brand's own website | 16.1% |
Source: PYMNTS Intelligence, published 2026-10-23. US consumers only.
For Canadian readers, the caveat comes first. Every number in this post comes from US shoppers. No Canadian equivalent has been published that we could find. The habits behind the numbers (trusting big marketplaces for returns, buying where you already have an account) are not unique to the US, so read the figures as a direction rather than a Canadian measurement.
Will AI shopping agents skip brand websites?
AI agents skip a brand site when nothing on it beats the marketplace on the things the shopper cares about. When price and shipping were equal, the reasons shoppers gave were easier returns (36.6%), better selection (34.9%) and habit (32.7%). None of those involve the agent. They are the same reasons people already default to Amazon.
The agent is mostly carrying out a preference the shopper already had. PYMNTS found 57.9% of people who shop with AI still usually buy on Amazon even when AI helped them find the product. The agent does the research; the old habit makes the purchase.
The vendor numbers point the same way. FirstPage Sage, an SEO and generative engine optimization agency, published a study on October 22, 2026 built on a panel of 4,936 US online shoppers plus order data from 43 brands. It found 61.4% use AI to research purchases, 6.8% have let an agent complete a checkout with their approval, and 1.3% have let one buy with no approval at all. Treat it as a vendor study: the firm sells AI search services. Even so, it matches PYMNTS on the shape. Research through AI is common. Handing over the wallet is rare.
Put the two studies together and the picture is split. Agents are not yet doing most of the buying. When they do research, the shortlist they bring back reflects whoever is easiest to understand and easiest to trust.
Why do agents prefer marketplaces?
Marketplaces give an agent everything it needs in one consistent format. Price, stock, delivery date, return window and reviews sit in the same place on every listing, and the shopper already has an account there. A brand site often makes the agent dig: shipping costs appear at checkout, the returns policy is a PDF, and sizes live in an image.
Retailers are also choosing very different doors. In its October 5, 2026 roundup, PYMNTS reported that Amazon blocks outside agents while it builds its own, QVC and HSN let agents complete purchases, Kohl's does not, and Walmart lets Google's Gemini show its products before handing shoppers to its own Sparky assistant. We cover how to write your own rules in our guide to an AI agent policy for your website.
Karen Webster, CEO of PYMNTS, put the problem in one line in a September column the October 23 piece quotes: "Recommendation is the easy part. Certainty and trust is the product." Marketplaces win on certainty. A brand site has to earn it on the page.
What makes an AI agent choose your website?
An agent picks a brand site when the site answers the questions the agent would otherwise have to guess at, and when there is a reason to buy direct. The survey's own switching numbers name the reasons: 53.9% of marketplace shoppers would buy from a brand for a lower price or exclusive discount, and 49.9% for free shipping as fast as a marketplace's.
| What the agent checks | What loses it | What wins it |
|---|---|---|
| Price | "Contact us" or price shown only in cart | Price in plain text on the product or service page |
| Shipping and delivery | Cost revealed at checkout | Cost and delivery window stated up front |
| Returns or cancellation | Policy buried in a PDF or footer | One clear policy page, linked from every product |
| Product or service details | Specs inside images | Structured data (JSON-LD) that matches the page |
| Reason to buy direct | Same deal as the marketplace | Direct-only price, bundle, or faster shipping |
| Trust | No reviews, no business address | Reviews, address, phone, real people named |
| Being on the shortlist | Never named in AI answers | Named in ChatGPT and Google AI answers for your category |
Two rows in the table deserve more weight. Structured data (labelled facts in your page code that machines can read) is how an agent reads your price and stock without guessing. PYMNTS has reported that AI search built on Shopify's merchant catalogue converts at twice the rate of AI search built on scraped data. Our structured data guide for AI search covers the markup.
The shortlist row comes first in practice. An agent can only choose from the businesses it already knows about. That is AEO, answer engine optimization, meaning getting named inside ChatGPT and Google AI answers. If you are not on the list, the rest of the table never gets checked. How ChatGPT recommends local businesses walks through what gets a business named.
What about service businesses that don't sell online?
Service businesses have no cart to skip, but agents still do the shortlisting and, increasingly, the booking. An accountant, a dentist or a renovation firm loses the job at the comparison stage, long before any payment.
Google has already built this. On November 17, 2025, Google said AI Mode searches "across multiple reservation platforms and websites to find real-time availability" for restaurant reservations, event tickets and beauty and wellness appointments, then gives direct links to finish booking through partners. Local appointments were open to US Search Labs users. In May 2026, Google said it would extend agentic booking to local experiences and services for everyone in the US. Both announcements cover the US only, but the shortlisting part already happens in any AI answer.
For a service business, the same table applies with different words:
- Price: a starting price or a range on the page beats "call for a quote". An agent cannot compare a phone number.
- Availability: a booking link or a stated response time ("we reply within one business day") gives the agent something to act on.
- Scope: say which services you offer, which you don't, and which areas you serve.
- Trust: reviews, credentials and named staff on the page, not only on Google Maps.
Our post on how an AI shortlist ends on your website covers the second half of that journey: once an agent names you, the person still clicks through, and the page has to close.
How can you tell if agents are visiting your site now?
Look in your server logs, not only in your analytics. Most AI agents never run your analytics script, so a normal dashboard undercounts them.
| Signal | Where to look | What it means |
|---|---|---|
| ChatGPT-User in the user agent | Server or CDN logs | Someone asked ChatGPT a question and it opened your page |
| OAI-SearchBot | Server or CDN logs | OpenAI indexing pages for ChatGPT search |
| GPTBot | Server or CDN logs | OpenAI crawling for model training, not a shopper |
| chatgpt.com or perplexity.ai as referrer | Analytics referral report | A person clicked a link inside an AI answer |
| Fast, form-filling sessions with no scrolling | Analytics and form logs | Possibly a browser agent acting for a real user |
OpenAI's bot documentation is explicit that ChatGPT-User is triggered by a person, not an automatic crawl, and that "robots.txt rules may not apply" to it. Blocking every bot in your firewall can mean blocking a customer's agent. We explain that new traffic type in agents acting on behalf of users, and if you also want to know which businesses are browsing you, see how to see which companies visit your website.
What these numbers don't tell you
The PYMNTS figures describe what people say they would do in a hypothetical, not what agents actually did. The survey covers US consumers, and the article does not give a margin of error. FirstPage Sage sells the services its study argues for. Neither source measures Canadian shoppers or service businesses directly. And the shift is early: with 6.8% of shoppers letting an agent check out, most of the effect today is on research and shortlists, not on carts.
Where this leaves your site
The bigger picture is in what AI shopping agents mean for your website. The short version: plain prices, stated policies, structured data and being named in AI answers are the same fixes that help human visitors. If you want to know where your site stands on them, our free site check at /audit looks at it.