# Beyond the Cart: What Merchants Can't See When AI Agents Shop

**Author:** G2 Research  
**Published:** November 2026  
**Last updated:** November 19, 2026  
**Based on:** For this agentic commerce report, G2 conducted 206 AI-moderated voice interviews between October 5 and 25, 2026, with e-commerce, digital commerce, customer experience, product and payments leaders at companies in the US, UK and Europe.  
**Topics covered:** AI Adoption, AI Agents & Autonomy

G2 interviewed 206 senior commerce leaders across the US, UK, and Europe in October 2026 about AI agents that buy on customers' behalf. Among the 169 who sell online, 29% say agents can already complete purchases on their platforms, yet only 12% of that group can reliably distinguish an agent's order from a human's. A third (34%) run bot checks, fraud rules, or authentication steps they expect would block an agent at checkout, while just 2% expect checkout friction to cost them an agent sale. Few anticipate the checkout disappearing entirely: 56% say it will survive as a human confirmation step, and only 9% expect cart abandonment to fade within two years.

**Executive answer:** 12% of merchants letting agents buy can reliably identify agent orders (base: 48)

## Additional key findings

- 78% of merchants who say agent purchases have caused no problems can't reliably tell agent orders from human ones (base: 126)
- 34% run fraud, bot or authentication checks they expect would stop an agent (base: 161)
- 56% expect checkout to survive as a human approval step (base: 167)

**On this page:** [Headline statistics](#headline-statistics) · [Will cart abandonment disappear because of agentic commerce?](#will-cart-abandonment-disappear-because-of-agentic-commerce) · [Research perspective](#research-perspective) · [Merchants can't see most of the AI agents already buying from them](#merchants-can-t-see-most-of-the-ai-agents-already-buying-from-them) · [Research image](#research-image) · [How many online retailers can tell AI agent traffic from human traffic?](#how-many-online-retailers-can-tell-ai-agent-traffic-from-human-traffic) · [No problems yet⁠ mostly means can't see yet](#no-problems-yet-mostly-means-can-t-see-yet) · [When an AI agent walks away, most merchants can't tell](#when-an-ai-agent-walks-away-most-merchants-can-t-tell) · [What would merchants see if an AI agent decided not to buy from them?](#what-would-merchants-see-if-an-ai-agent-decided-not-to-buy-from-them) · [If more sales came through AI agents, what would you want to measure that you can't today?](#if-more-sales-came-through-ai-agents-what-would-you-want-to-measure-that-you-can-t-today) · [Merchants' own fraud and bot checks are the agent blocker they aren't counting](#merchants-own-fraud-and-bot-checks-are-the-agent-blocker-they-aren-t-counting) · [Can today's payment and fraud checks block AI agents from completing purchases?](#can-today-s-payment-and-fraud-checks-block-ai-agents-from-completing-purchases) · [Where agents already buy, errors already show up](#where-agents-already-buy-errors-already-show-up) · [The online checkout survives as a single moment of human approval](#the-online-checkout-survives-as-a-single-moment-of-human-approval) · [What replaces the online checkout when AI agents do the buying?](#what-replaces-the-online-checkout-when-ai-agents-do-the-buying) · [How fast is this moving?](#how-fast-is-this-moving) · [Will metrics tied to human shopping, such as cart abandonment, stop mattering?](#will-metrics-tied-to-human-shopping-such-as-cart-abandonment-stop-mattering) · [Merchants are betting on reputation to win over AI shopping agents](#merchants-are-betting-on-reputation-to-win-over-ai-shopping-agents) · [What makes an AI agent pick one merchant over another at similar prices?](#what-makes-an-ai-agent-pick-one-merchant-over-another-at-similar-prices) · [No one owns the customer experience once several AI agents are involved](#no-one-owns-the-customer-experience-once-several-ai-agents-are-involved) · [Who is accountable when an AI agent's purchase goes wrong across several platforms?](#who-is-accountable-when-an-ai-agent-s-purchase-goes-wrong-across-several-platforms) · [Same blind spots everywhere, but US merchants are calmer about them](#same-blind-spots-everywhere-but-us-merchants-are-calmer-about-them) · [Research image](#research-image-2) · [How does agentic commerce readiness differ across the US, UK and Europe?](#how-does-agentic-commerce-readiness-differ-across-the-us-uk-and-europe) · [What merchants should do before agents become a large share of sales](#what-merchants-should-do-before-agents-become-a-large-share-of-sales) · [Frequently asked questions](#frequently-asked-questions) · [Trust decides who wins the agent-ready checkout](#trust-decides-who-wins-the-agent-ready-checkout) · [Research prepared for Paysafe](#research-prepared-for-paysafe) · [Methodology](#methodology)

---

<a id="headline-statistics"></a>

## Headline statistics

- **12%** — of agent-enabled sellers can reliably spot agent orders
- **78%** — of ‘no problem’ sellers can’t reliably spot agent orders
- **34%** — run checks they expect would stop agents
- **56%** — expect checkout to stay as human approval

---

<a id="will-cart-abandonment-disappear-because-of-agentic-commerce"></a>

## Will cart abandonment disappear because of agentic commerce?

**Agentic commerce** is online buying where an AI agent, sometimes called an AI shopping assistant, finds, compares and pays for products on a person's behalf.

G2 interviewed 206 senior commerce leaders across the US, UK and Europe, all screened to director level or above and half of them C-suite executives, founders or VPs, including 52 leaders at companies with more than 1,000 employees. The interviews, held in October 2026, focused on AI agents that buy on a customer's behalf. Among the 169 who sell online, 29% say agents can already complete purchases with them, yet only 12% of that group can reliably tell an agent's order from a human's. A third (34%) run bot checks, fraud rules or authentication steps they expect would stop an agent at checkout, while just 2% expect checkout friction to cost them an agent sale. Few expect the checkout to vanish: 56% say it will survive as a human confirmation step, and only 9% expect cart abandonment to fade within two years.

**34% — Agent walk-aways leave no signal**

When an agent looks at a business and buys elsewhere, a third of merchants would see no signal at all. Cart abandonment has a successor that most merchants can't yet see. (Base: 157 online sellers.)

**13% — Agent order errors have started**

Where agents already buy, errors have started: about one merchant in eight has handled a wrong product, size or price from an agent order. (Base: 47 merchants that let agents buy.)

**33% — Reputation beats product data**

Merchants expect reputation and reviews to win over agents (33%), well ahead of the structured product data agents rely on (13%). (Base: 165 online sellers.)

**51% — Complaints are the first warning**

Half expect a customer complaint to be the first sign an agent journey went wrong, and no single team owns that journey. (Base: 166 online sellers.)

---

<a id="research-perspective"></a>

## Research perspective

> Agentic commerce is moving from idea to reality, and consumers will decide how far it goes. Shoppers are increasingly ready to let agents search and compare on their behalf, with many welcoming the convenience while still wanting the final say. This research shows merchants expect the checkout to endure as a clear moment of customer approval, and that experience has to be fast, simple and secure because trust is what makes agentic commerce work. As agents begin to shop, our focus is on giving consumers the confidence to pay how they want, with the control they expect.

Bob Legters, Chief Product Officer, Paysafe

Portrait of Bob Legters, Chief Product Officer, Paysafe

---

<a id="merchants-can-t-see-most-of-the-ai-agents-already-buying-from-them"></a>

## Merchants can't see most of the AI agents already buying from them

Half of online sellers (49%) can't tell AI agent traffic or purchases apart from human ones. Merchants that already let AI agents buy are no better off: of the 48 that do (29% of online sellers), only 12% can reliably tell an agent's order from a human's.

---

<a id="research-image"></a>

## Research image

Diagram: The checkout journey merchants can see, and the one they can't. Today, a person shops in five steps: lands on the site, browses and compares, adds to cart, fills in checkout, pays and gets confirmation; every step leaves a signal such as page views, dwell time and abandoned carts. Next, an AI agent shops: the shopper asks an AI assistant, the agent reads and compares sites, the agent picks a merchant or walks away, the shopper approves the purchase, and the order arrives. The two middle steps are run by the agent, and many merchants get little or no signal from them.

---

<a id="how-many-online-retailers-can-tell-ai-agent-traffic-from-human-traffic"></a>

## How many online retailers can tell AI agent traffic from human traffic?

Smaller sellers are the most exposed: 56% of companies with up to 250 employees can't tell agents apart, against 32% of companies with more than 1,000 employees. Even among merchants that already let agents buy, 40% can't. Larger companies lean on partial signals such as bot scores and referral sources, which catch some agents and miss the ones that behave like people.

Can your analytics tell AI agent traffic or purchases from human ones?

- **49%** — Can't tell agents apart; Source: All online sellers, base: 156. Source: G2 Research, agentic commerce merchant study, October 2026.

- **25%** — Partly, through bot or referral signals; Source: All online sellers, base: 156. Source: G2 Research, agentic commerce merchant study, October 2026.

- **12%** — Reliably, with custom tracking; Source: All online sellers, base: 156. Source: G2 Research, agentic commerce merchant study, October 2026.

- **14%** — Other answers; Source: All online sellers, base: 156. Source: G2 Research, agentic commerce merchant study, October 2026.

Source: All online sellers, base: 156. Source: G2 Research, agentic commerce merchant study, October 2026.

Bases include respondents who answered the question.

> We don't know what we don't know. So we don't know if there are agents that behave exactly as humans.

Head of product, media publisher

telling AI agents apart from human shoppers

> I would not understand that directly. I'm not measuring for whether the request is coming through an AI agent or a human. There is no way for me to distinguish between the two. At least I am not aware of a way. The customer would need to come in directly and actually complain formally and explain how they got to the product page.

Owner, candle and fragrance brand

telling AI agents apart from human shoppers

---

<a id="no-problems-yet-mostly-means-can-t-see-yet"></a>

## No problems yet⁠ mostly means can't see yet

**84%** — of online sellers say no agent purchase has ever gone wrong for them

Most online sellers (84%) say no agent purchase has ever gone wrong for them. Few of them are in a position to know: 78% of the merchants who say agent purchases have caused no problems can't reliably tell agent orders from human ones.

Meanwhile, 59% of merchants still treat conversion rate and the [cart abandonment rate](https://baymard.com/lists/cart-abandonment-rate) as their core ecommerce funnel metrics, and 6% track AI search visibility or [referrals from AI tools](https://business.adobe.com/assets/pdfs/resources/sdk/ai-traffic-trends-report-august-2026/ai-traffic-trends-report-2026-08-19.pdf). G2's [State of AEO 2026](https://research-hub.g2.com/the-state-of-aeo-in-marketing-aeo-g2-insights) research found the same measurement gap on the marketing side.

**Strategic implication:** Merchants are about to set budgets, redesign checkouts and tune fraud rules for agents with almost no data on how agents behave on their own sites. The cheapest fix comes first: AI agent traffic detection. Tag agent traffic now, so next year's decisions start from evidence.

---

<a id="when-an-ai-agent-walks-away-most-merchants-can-t-tell"></a>

## When an AI agent walks away, most merchants can't tell

Cart abandonment has a successor: agent abandonment. If an agent looked at their business and chose a competitor, a third of online sellers (34%) say they would see nothing at all, and another 27% would see only an ordinary bounce that looks like a human leaving.

---

<a id="what-would-merchants-see-if-an-ai-agent-decided-not-to-buy-from-them"></a>

## What would merchants see if an AI agent decided not to buy from them?

**41%** — of companies with up to 250 employees would see nothing at all

The blind spot is widest among smaller merchants: 41% of companies with up to 250 employees would see nothing at all, against 16% of companies with more than 1,000 employees.

If an AI agent decided not to buy from or recommend you, what would you see?

- **34%** — Nothing at all; Source: Base: 157 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **27%** — An ordinary bounce or abandoned session; Source: Base: 157 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **16%** — Indirect signs, such as traffic shifts, AI visibility or reviews; Source: Base: 157 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **23%** — Other answers; Source: Base: 157 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

Source: Base: 157 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

Bases include respondents who answered the question.

> I don't think we would be able to see and say, hey, an AI agent declined to proceed with you and why. We would just consider that an abandonment, and we wouldn't know whether it was a human or an AI agent.

Travel company

AI agents choosing a competitor

---

<a id="if-more-sales-came-through-ai-agents-what-would-you-want-to-measure-that-you-can-t-today"></a>

## If more sales came through AI agents, what would you want to measure that you can't today?

Asked what they would measure if agents drove more of their sales, merchants reached for the questions cart abandonment used to answer: where the shopper came from and why they left.

**Strategic implication:** Expect agent abandonment to become the metric e-commerce teams argue about next. The merchants who can explain why an agent picked a competitor will set the benchmark, and today very few are collecting the data to do it.

If more sales came through AI agents, what would you want to measure that you can't today?

- **28%** — Where agent traffic comes from and what the shopper asked; Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **25%** — Why an agent dropped them or chose a competitor; Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **12%** — Agent conversion rate and purchase success; Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **5%** — Lifetime value and satisfaction of agent customers; Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **30%** — Other answers; Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

Bases include respondents who answered the question.

---

<a id="merchants-own-fraud-and-bot-checks-are-the-agent-blocker-they-aren-t-counting"></a>

## Merchants' own fraud and bot checks are the agent blocker they aren't counting

A third of online sellers named a bot check, CAPTCHA, fraud rule or authentication step they expect would stop an agent mid-purchase. Yet when we asked why an agent might pass them over, almost none mentioned checkout or payment friction. Price, product data and reviews came up most.

---

<a id="can-today-s-payment-and-fraud-checks-block-ai-agents-from-completing-purchases"></a>

## Can today's payment and fraud checks block AI agents from completing purchases?

The checks themselves are ordinary: a one-time code sent to the shopper's phone, a CAPTCHA asking the visitor to prove it's human, a fraud model flagging a purchase made far from the billing address. Each was built to keep out bots and fraudsters, and the same check can stop an agent the customer sent on purpose. Not every agent gets blocked, but some merchants already see it happen: legitimate agent purchases flagged as fraud, accounts locked, and codes sent to a phone the agent can't read.

The gap merchants aren't counting

Bases: 161 and 162 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **34%** — run a check they expect would stop an agent at checkout

- **2%** — expect checkout or payment friction to cost them an agent sale

> Right now we have completely blocked at this point agents to access the checkout page. If the checkout page is not being accessed by a human, then we simply do not let any agent to proceed to the checkout page.

Customer experience, service and product, online food delivery marketplace

payment and fraud checks

> So today, our fraud and monitoring parameters are not tailored for AI, and we are having lots of false positives whenever AI agents are trying to either gather information or make a purchase. A lot of user accounts are being blocked, and they can't proceed on our site because of this.

Product and growth, online marketplace

payment and fraud checks

---

<a id="where-agents-already-buy-errors-already-show-up"></a>

## Where agents already buy, errors already show up

Six of the 47 merchants that let agents buy (13%) have handled a wrong product, wrong size or pricing mistake from an agent order, against 2% of those that don't allow agents yet. Respondents described agents ordering the wrong size or flavor, which drove up returns, and a telecom retailer that gave an agent a 90% discount instead of 10% when it asked for one, and then had to roll the order back.

Among merchants that haven't switched agents on, technical readiness and platform limits are the most common obstacle (28%), with security, fraud and authentication at 17%. And 11% say identity, regulatory and fraud steps stay in the journey however fast agents get.

**Strategic implication:** This moves payments and fraud teams to the center of agentic commerce. How a checkout recognizes a legitimate agent acting with the customer's permission is an [agentic payments](https://blog.google/products-and-platforms/platforms/google-pay/agent-payments-protocol-fido-alliance/) and identity question first, and merchants will need it answered before agent volumes make the false declines expensive.

Is there anything in how you handle payments or fraud checks that might stop an AI agent completing a purchase?

- **13%** — Bot detection, CAPTCHAs or web firewalls; Source: Base: 161 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **11%** — Fraud rules on velocity, unusual patterns or address mismatches; Source: Base: 161 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **9%** — Multi-factor authentication, 3D Secure or biometrics; Source: Base: 161 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **40%** — No known barrier; Source: Base: 161 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **26%** — Other answers; Source: Base: 161 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

Source: Base: 161 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

Percentages may not total 100 because of rounding. Bases include respondents who answered the question.

> They're not sending it to the AI agent, they're sending it to the human. So, that's the only sticking point.

Founder and CEO, consumer goods brand

payment and fraud checks

> So most cases when an agent orders a product, the sizing is off, and then there's always a return initiated.

Marketing, growth, e-commerce and CX, fashion e-commerce company

payment and fraud checks

---

<a id="the-online-checkout-survives-as-a-single-moment-of-human-approval"></a>

## The online checkout survives as a single moment of human approval

Merchants don't expect agents to kill the checkout. More than half (56%) say it will survive in some form, with a person confirming the purchase and security checks in place, and only 9% expect cart abandonment to fade within two years.

---

<a id="what-replaces-the-online-checkout-when-ai-agents-do-the-buying"></a>

## What replaces the online checkout when AI agents do the buying?

One-click checkout and Click to Pay have already made frictionless checkout the norm for human shoppers, and [agentic checkout](https://openai.com/index/buy-it-in-chatgpt/) extends that logic to the agent itself. Where the ecommerce checkout does change, merchants expect it to shrink rather than vanish: 14% expect digital wallets, platform checkouts or one-click passkeys to take over, and a quarter (25%) already plan to strip cart and checkout steps for agent purchases.

Others want to keep the journey as it is: 26% plan to protect human touchpoints such as discovery and choice, and 24% have no redesign plans yet. Nearly one in five (19%) expect people who keep shopping without agents to get a worse experience, through faster sell-outs and less investment in the human storefront.

Could the online checkout be phased out? If so, what replaces it?

- **56%** — It survives, with human confirmation and safeguards; Source: G2 Research, agentic commerce merchant study, October 2026.; Base: 167 online sellers

- **14%** — Wallets, platform checkouts or passkeys replace it; Source: G2 Research, agentic commerce merchant study, October 2026.; Base: 167 online sellers

- **4%** — It becomes a conversational or voice confirmation; Source: G2 Research, agentic commerce merchant study, October 2026.; Base: 167 online sellers

- **4%** — Agents pay each other in the background; Source: G2 Research, agentic commerce merchant study, October 2026.; Base: 167 online sellers

- **22%** — Other answers; Source: G2 Research, agentic commerce merchant study, October 2026.; Base: 167 online sellers

Source: G2 Research, agentic commerce merchant study, October 2026.

Base: 167 online sellers

Bases include respondents who answered the question.

> I think the moment of authorization won't disappear though.

Founder, education publisher

the future of the online checkout

> I suspect that there will be a tipping point in the future where it becomes easier to make a purchase using an AI agent than it is to do it as a human being through a traditional e-commerce journey.

Product director, internet service provider

the future of the online checkout

---

<a id="how-fast-is-this-moving"></a>

## How fast is this moving?

**63%** — of online sellers don't let AI agents complete purchases yet

Not as fast as the headlines suggest. Three in ten online sellers say agents can complete purchases with them today, fewer than one in ten are testing it, and the rest haven't switched it on. That pace gives merchants time to redesign the checkout without shutting out the people who still want to buy it the old way.

Can AI agents complete purchases on behalf of your customers today?

- **29%** — Yes, live today; Source: All online sellers, base: 166. Source: G2 Research, agentic commerce merchant study, October 2026.

- **8%** — Testing; Source: All online sellers, base: 166. Source: G2 Research, agentic commerce merchant study, October 2026.

- **63%** — Not yet; Source: All online sellers, base: 166. Source: G2 Research, agentic commerce merchant study, October 2026.

- **1%** — Other answers; Source: All online sellers, base: 166. Source: G2 Research, agentic commerce merchant study, October 2026.

Source: All online sellers, base: 166. Source: G2 Research, agentic commerce merchant study, October 2026.

Percentages may not total 100 because of rounding. Bases include respondents who answered the question.

---

<a id="will-metrics-tied-to-human-shopping-such-as-cart-abandonment-stop-mattering"></a>

## Will metrics tied to human shopping, such as cart abandonment, stop mattering?

Not soon, according to merchants. Only 9% of online sellers G2 interviewed in 2026 expect cart abandonment to fade or turn into agent-specific metrics within two years, and 10% expect that in three to five years or more. A majority (54%) say human shopping metrics will stay essential alongside new agent metrics.

**Strategic implication:** Expect the cart to fade from view while the moment of approval stays. Merchants that make that moment fast, clear and secure keep the trust the old checkout was built to earn, and they can drop most of the form-filling around it.

Will metrics tied to human shopping, such as cart abandonment, stop mattering?

- **54%** — No, they stay essential; Source: All online sellers, base: 165. Source: G2 Research, agentic commerce merchant study, October 2026.

- **10%** — Yes, in 3 to 5 years or more; Source: All online sellers, base: 165. Source: G2 Research, agentic commerce merchant study, October 2026.

- **9%** — Yes, within 1 to 2 years; Source: All online sellers, base: 165. Source: G2 Research, agentic commerce merchant study, October 2026.

- **27%** — Other answers; Source: All online sellers, base: 165. Source: G2 Research, agentic commerce merchant study, October 2026.

Source: All online sellers, base: 165. Source: G2 Research, agentic commerce merchant study, October 2026.

Bases include respondents who answered the question.

> I do think human driven shopping metrics will always be important. There will always be a subset of the population that does not feel comfortable using AI for a variety of different reasons and just prefer to do things manually.

Owner, online reseller

the future of the online checkout

---

<a id="merchants-are-betting-on-reputation-to-win-over-ai-shopping-agents"></a>

## Merchants are betting on reputation to win over AI shopping agents

When price is level, merchants expect brand reputation and reviews to make an agent choose them. Fewer bet on the structured product data agents rely on. Nearly two in three (64%) say it would matter if customers kept buying but stopped seeing their brand.

> Absolutely. I think it's crucial that the brand is visible and has been seen by the customers. Otherwise it becomes replaceable, and it's not an asset anymore.

Managing director, fashion retailer

brand visibility with AI agents

---

<a id="what-makes-an-ai-agent-pick-one-merchant-over-another-at-similar-prices"></a>

## What makes an AI agent pick one merchant over another at similar prices?

Yet the part of their own website merchants most expect an agent to misread is the brand story itself: positioning, service nuances and the reasons behind a higher price (24%), just ahead of specifications, variants and sizing (23%).

If agents became the main way customers shop, merchants would move effort into the relationship around the purchase: brand and direct loyalty (24%) and post-purchase service (17%), alongside being found by AI (23%).

**Strategic implication:** Reputation only helps if an agent can read it. Merchants that turn their reviews, positioning and pricing logic into plain, structured facts give agents something to compare. The rest leave agents to guess, and agents guess on price.

What would make an agent pick your company over a competitor at a similar or lower price?

- **33%** — Reputation, authority and verified reviews; Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **18%** — Product quality, fit and delivery speed; Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **13%** — Structured product data and AI search visibility; Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **6%** — Loyalty and post-purchase service; Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **30%** — Other answers; Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

Source: Base: 165 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

Bases include respondents who answered the question.

> It might misunderstand where we're trying to position ourselves as a brand: more higher end. So, I suppose from a pricing point of view, we're not the cheapest, but in terms of the ingredients, we are certainly one of the best.

Marketing, growth and e-commerce, supplements brand

brand visibility with AI agents

> One of the things we would increase significantly would be our loyalty concept. We have a loyalty club for past customers, and that would be a more critical and more important tool for us if the purchasing process is taken over by AI agents, which kind of make our company become invisible and turn the brand into more like a generic product. Then we'll need to look much more intensely into the time between purchasing.

Marketing, growth and e-commerce, travel company

brand visibility with AI agents

---

<a id="no-one-owns-the-customer-experience-once-several-ai-agents-are-involved"></a>

## No one owns the customer experience once several AI agents are involved

When a purchase runs through several AI tools and something goes wrong, merchants don't agree on who should fix it. The most common answer, shared ownership across teams, came from only 20%. And half (51%) expect to hear about the problem first from an unhappy customer.

---

<a id="who-is-accountable-when-an-ai-agent-s-purchase-goes-wrong-across-several-platforms"></a>

## Who is accountable when an AI agent's purchase goes wrong across several platforms?

Smaller merchants lean on complaints the most: 58% of companies with up to 250 employees expect a complaint to be their first warning, against 35% of companies with more than 1,000 employees.

That reliance on complaints is risky. A shopper whose agent fails at checkout may never get in touch at all. One retailer doubts they would ever hear about it:

> I can't imagine the customer would then come on and make a complaint or track it in another way. They'd just come on and either buy it or go somewhere else.

— Marketing, growth and e-commerce, toy retailer

**Strategic implication:** Until one team owns the multi-agent journey, agent failures will show up as lost sales nobody can explain.

If a customer's journey spans several AI tools and goes wrong, who in your business is responsible?

- **20%** — Shared across marketing, product and tech; Source: Base: 167 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **17%** — E-commerce and digital product teams; Source: Base: 167 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **16%** — Customer service and support; Source: Base: 167 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **10%** — Leadership, operations or payments teams; Source: Base: 167 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

- **36%** — Other answers, often "it depends"; Source: Base: 167 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

Source: Base: 167 online sellers. Source: G2 Research, agentic commerce merchant study, October 2026.

Percentages may not total 100 because of rounding. Bases include respondents who answered the question.

> There is shared responsibility in my business. So you have the marketing team, the sales team, the IT team, we have agencies, and of course the board. It would be shared responsibility.

Marketing, growth and e-commerce, leisure venues operator

who owns a failed agent journey

---

<a id="same-blind-spots-everywhere-but-us-merchants-are-calmer-about-them"></a>

## Same blind spots everywhere, but US merchants are calmer about them

Merchants in all three regions gave similar answers on most questions. Around three in ten already let agents buy, most can't fully tell an agent's order from a human's, and about half expect to hear about agent problems from complaints. Where they differ is how worried they are: US merchants are consistently the calmest, while UK and European merchants are more alert to what agents could change.

---

<a id="research-image-2"></a>

## Research image

Two bar charts of online sellers by region (US / UK / Europe). Where the US, UK and Europe agree: AI agents can already complete purchases 27% / 30% / 29% (bases 81 / 50 / 34); can't fully tell agent orders from human ones 70% / 78% / 75% (73 / 50 / 32); would see nothing if an agent chose a competitor 33% / 37% / 28% (73 / 51 / 32); expect checkout to survive as a human approval step 60% / 58% / 44% (80 / 52 / 34); expect complaints to be the first warning 53% / 46% / 50% (81 / 50 / 34). Where they differ: say their reporting is still trusted 49% / 35% / 24% (79 / 51 / 33); expect little or no change for shoppers who skip agents 55% / 37% / 41% (77 / 52 / 34); have no checkout redesign plans yet 31% / 20% / 12% (80 / 50 / 33); think price or value is why an agent would drop them 16% / 33% / 24% (76 / 51 / 34).

---

<a id="how-does-agentic-commerce-readiness-differ-across-the-us-uk-and-europe"></a>

## How does agentic commerce readiness differ across the US, UK and Europe?

The gap runs in the same direction across several questions about urgency. Half of US merchants (49%) say their reporting is still trusted as agents arrive, against 35% in the UK and 24% in Europe. More than half of US merchants (55%) expect people who keep shopping without agents to notice little or no change, against 37% in the UK and 41% in Europe, and 31% have no plans yet to redesign their checkout, against 20% and 12%. UK merchants are also the most worried about price: 33% think an agent would drop them on price or value, against 16% in the US and 24% in Europe.

**Strategic implication:** Merchants selling across borders can treat agent readiness as one program: the same detection, measurement and checkout fixes apply in every market. What differs is urgency. US teams may need the strongest case that agents already affect them, while UK and European teams are readier to act.

---

<a id="what-merchants-should-do-before-agents-become-a-large-share-of-sales"></a>

## What merchants should do before agents become a large share of sales

### Audit the security stack for agents

34% of online sellers run checks they expect would stop an agent, yet only 2% think checkout friction could cost them an agent sale. List every CAPTCHA, fraud rule, one-time code and 3D Secure step a purchase passes through, and decide which should recognize a verified agent acting with the customer's permission and which must always stop and ask a person. Ask your payment provider how it will handle agentic payments and verify that an agent really acts for the customer. Reviewing your fraud protection and payment gateway settings with this question in mind is the quickest place to start.

### Learn to recognize agent orders, then measure agent abandonment

Only 12% of merchants with agent purchases live can reliably tell an agent order from a human one, and a third (34%) would see nothing when an agent walks away. Tag agent traffic wherever your tools allow, then log the agent sessions that fail and where they stop.

### Keep the confirmation step and make it fast

56% expect checkout to survive as a moment of human approval. Design one clear, secure confirmation, the heart of the agentic checkout, and remove the form-filling around it.

### Write the brand story so a machine can read it

Positioning is what merchants most expect agents to misread (24%). State plainly why you cost more, what makes the product different and what the customer gets, and back it with structured product data, which shapes how AI agents read websites. Then test the prompts customers type into AI shopping assistants to see how agents describe you.

### Name one owner for the multi-agent journey

The most common answer on ownership came from only 20%, and 51% expect a complaint to be their first warning. Give one team the job of monitoring agents acting for you and for your customers, with signals that arrive before the complaint does.

---

<a id="frequently-asked-questions"></a>

## Frequently asked questions

**What is agentic commerce?** Agentic commerce is online buying where an AI agent, sometimes called an AI shopping assistant, finds, compares and pays for products on a person's behalf. In G2's October 2026 study, 29% of online sellers said agents can already complete purchases with them, and another 8% were testing it.

**Can merchants tell when an AI agent is shopping on their site?** Most can't do it reliably. In G2's 2026 interviews with 169 online sellers, 49% said their analytics can't separate agent traffic or purchases from human ones, 25% can do it partly, and only 12% can do it reliably. The share is no higher among merchants that already let agents buy.

**Will AI agents replace the online checkout?** Merchants don't expect a full replacement. In G2's 2026 study, 56% of online sellers expect checkout to survive in some form with a human confirmation step and security checks, 14% expect wallets, passkeys or platform checkouts to take over, and 4% expect fully automated agent-to-agent payments.

**Will cart abandonment disappear because of agentic commerce?** Not soon, according to merchants. Only 9% of online sellers G2 interviewed in 2026 expect cart abandonment to fade or turn into agent-specific metrics within two years, and 10% expect that in three to five years or more. A majority (54%) say human shopping metrics will stay essential alongside new agent metrics.

**What stops an AI agent from completing a purchase?** Often the merchant's own security. In G2's 2026 study, 34% of online sellers named a bot check, CAPTCHA, fraud rule or authentication step such as 3D Secure that they expect would stop an agent at checkout. Only 2% expected checkout or payment friction to be the reason an agent passes them over.

**How do brands stay visible when AI agents shop for customers?** Merchants are betting on reputation. In G2's 2026 study, 33% of online sellers think established reputation and verified reviews would make an agent pick them, 13% point to structured product data and AI search visibility, and 64% say losing brand visibility to an agent would matter to their business.

**Who is responsible when an AI agent purchase goes wrong?** There is no agreed owner. Asked who would own a bad experience spread across several AI tools, online sellers in G2's 2026 study split between shared cross-functional ownership (20%), e-commerce teams (17%), customer service (16%) and leadership or payments teams (10%).

**Do merchants in the US, UK and Europe see agentic commerce differently?** They face the same blind spots but feel different levels of urgency. In G2's 2026 study, online sellers in all three regions gave similar answers on agent adoption, detection and the future of checkout. US merchants were the calmest: 49% said their reporting is still trusted, against 35% in the UK and 24% in Europe, and 31% had no checkout redesign plans yet, against 20% and 12%. UK merchants were the most likely to expect an agent to drop them on price or value (33%).

**How does this report define its key terms?** Agentic commerce: an AI agent finds, compares and pays for products on a person's behalf. Online sellers: the 169 respondents whose companies sell to customers online, the base for every statistic. Agent abandonment: an AI agent looks at a merchant and buys elsewhere, leaving no cart behind. Cart autonomy: an agent assembles and pays for a basket within limits the shopper set in advance, so the cart is never abandoned in the old sense. Reliably identify: analytics that separate agent orders from human ones through custom tracking, as opposed to partial bot or referral signals.

**What should merchants measure in agentic commerce?** Merchants most want two things they can't see today: where agent traffic comes from and what the shopper asked for (28%), and why an agent dropped them or chose a competitor (25%). Today 59% mainly track conversion and cart abandonment, and 6% track AI search visibility.

---

<a id="trust-decides-who-wins-the-agent-ready-checkout"></a>

## Trust decides who wins the agent-ready checkout

Taken together, the interviews describe merchants who have opened the door to AI agents before switching on the lights. Three in ten say agents can already buy from them. Most can't tell when it happens or see when an agent leaves, and a third run security checks that could turn legitimate agents away without anyone noticing. The fixes sit in the plumbing more than the storefront. Recognizing an agent that acts with a customer's permission, keeping one clear moment of approval and measuring what agents do are payments, identity and data problems. Merchants that solve them first will see agent demand while their competitors still file it under bot traffic.

---

<a id="research-prepared-for-paysafe"></a>

## Research prepared for Paysafe

**Sponsored**

[Visit Paysafe](https://www.paysafe.com)

About the sponsor

Paysafe is a global payments platform built for the experience economy. We connect businesses, consumers, and fans across iGaming, video gaming, esports, events, travel, hospitality, online trading and digital commerce, helping lift everyday and high-engagement experiences through reliable payment solutions. These include card payments, digital wallets such as Skrill, PaysafeWallet, eCash products like PaysafeCard, and a broad set of local payment methods. With approximately 2,800 employees across 12 countries and $167 billion in annualized volume in 2025, Paysafe focuses on making digital experiences easier to access and pay for. More information is available at [www.paysafe.com](https://www.paysafe.com).

[www.paysafe.com](https://www.paysafe.com)

This research was commissioned by Paysafe. G2 maintains full editorial independence.

---

<a id="methodology"></a>

## Methodology

G2 is the world's largest and most trusted software marketplace.

For this agentic commerce report, G2 conducted 206 AI-moderated voice interviews between October 5 and 25, 2026, with e-commerce, digital commerce, customer experience, product and payments leaders at companies in the US, UK and Europe. Three independent research panels recruited respondents, targeting director-level and above leaders in retail and e-commerce, travel and hospitality, gaming and entertainment, and iGaming. Interviews averaged 20 minutes (10 to 45 minutes) and followed a structured guide of 26 questions on agentic purchasing and the future of online checkout. All 197 respondents with panel screener records were at director level or above (head of function, director, VP, C-suite or founder), and half (49%) are C-suite executives, founders or VPs. Among the 169 online sellers, 52 work at companies with more than 1,000 employees and 16 at companies with more than 10,000.

Statistics in this report use the 169 respondents whose companies sell to customers online. We set aside 37 respondents whose companies aren't online merchants: businesses that sell through consulting, invoicing or in person, technology and payment providers, consultants describing a client's business, and brands that sell only through retailers or distributors. Each chart states its base, which excludes people who didn't answer that question.

G2 Interviews grouped open answers into categories with AI-assisted analysis, and G2's research team reviewed the coding and every quote against the transcripts. Quotes are lightly edited for readability: we removed filler words such as "um" and "uh", false starts and repeated words, added punctuation and corrected obvious transcription errors, without changing what respondents said. They are attributed by role, industry and company size. Just over half of the 169 (54%) work at companies with up to 250 employees, answers are self-reported, and no figure in this report rests on fewer than 30 respondents.

Regional comparisons are based on panel screener records of each respondent's country: US 82, UK 52 and Europe 34 (Germany 17, Denmark 5, Netherlands 3, Sweden 3, Norway 2, Italy 1, Finland 1, Switzerland 1 and Ireland 1).

- **206** — AI-moderated voice interviews
- **169** — online sellers (base for every statistic)
- **20 min** — average interview length
- **3** — regions: US, UK and Europe

- **61%** — Marketing, growth and e-commerce
- **15%** — Customer experience, service and product
- **6%** — Payments and operations
- **18%** — Other or several areas

- **64%** — Retail, fashion and consumer goods
- **11%** — Travel, hospitality and dining
- **7%** — Gaming, entertainment and digital media
- **4%** — B2B, technology and marketplace platforms
- **14%** — Other industries

2026

Company size: Up to 250 employees 54%, 251 to 1,000 13%, 1,001 to 10,000 21%, More than 10,000 9%, Not stated clearly 2%. Bases: 169, 168 and 167 online sellers. Categories come from what respondents said in the interview. Percentages may not total 100 because of rounding.

Who we spoke to (the 169 online sellers):

---

*Source: G2 Research — "Beyond the Cart: What Merchants Can't See When AI Agents Shop" (November 2026). Based on G2 first-party research. Markdown edition structured for answer-engine optimization (AEO).*