Agentic commerce 2026: key findings

AI has already changed where many buying journeys begin. The next change is more consequential: assistants are moving from answering questions to comparing offers, interpreting policies, assembling baskets and, in early implementations, participating in payment.

So how ready are brands for AI agents in 2026?

The industry evidence points to a clear answer: customer adoption is ahead of merchant readiness. AI-assisted discovery is growing quickly, but the product data, measurement, governance and payment controls required for dependable agent-led commerce are still uneven.

The strongest signals are:

48% of online shoppers used AI to research their most recent purchase, according to Visa Acceptance Solutions and PYMNTS Intelligence. Yet only 15% of merchants had structured product data ready for agents, and only 23% could distinguish AI-driven traffic from human traffic.
Salesforce reported that agentic search as the first step in a shopping journey grew 200% year over year. Only 28% of commerce organisations were already using agentic AI, although another 44% planned to adopt it within six months.
Capgemini found that 58% of consumers had replaced traditional search engines with generative AI tools for product or service recommendations, while 71% wanted generative AI integrated into their shopping experiences.
The transaction layer is earlier. Visa reported that only 19% of merchants had solutions in place to accept agent-initiated payments, while 63% were exploring or implementing support.

These studies do not use one shared definition of “agentic commerce,” and they do not create a universal readiness score. Together, however, they reveal the same structural gap: consumers are learning to shop through AI faster than most brands are learning to serve machines.

The defining readiness question of 2026 is not simply “Can an AI find us?” It is “Can an agent interpret our offer, verify the conditions, act with permission and produce a measurable outcome?”

From search traffic to delegated action

Agentic commerce is not another name for a chatbot. It describes commerce in which software can perform meaningful parts of a buying task for a person: researching a need, comparing options, checking live price and availability, applying constraints, assembling a cart, requesting approval and initiating a purchase.

The market is moving through those stages at different speeds.

AI discovery is already mainstream behaviour

Capgemini Research Institute surveyed 12,000 consumers across 12 countries in October and November 2024. It found that 58% had replaced traditional search engines with generative AI tools as their preferred source for product or service recommendations, up from 25% in 2023. Seventy-one percent wanted generative AI integrated into their shopping experiences, and 68% wanted AI to combine information from search engines, social platforms and retailer websites into one set of purchase options. Read Capgemini’s 2025 consumer research.

Visa Acceptance Solutions and PYMNTS Intelligence found the same shift in early 2026. Across 5,241 consumers in the United States, Brazil and the United Arab Emirates, 48% used AI while researching their most recent purchase. Thirty percent used ChatGPT or another OpenAI tool, 19% used Gemini and 22% used a merchant’s own AI assistant. Read the 2026 Agentic Commerce Deep Dive.

Salesforce’s fourth State of Commerce study, based on 3,450 commerce professionals, 4,690 consumers and behavioural data from more than 1.5 billion shoppers, reported that use of agentic search as the first step in shopping grew 200% year over year. Read Salesforce’s State of Commerce findings.

These figures measure different populations and behaviours, so they should not be compared as if they were one time series. But they agree on the direction: AI interfaces are becoming an important place where consumers form intent and narrow their choices.

AI-referred shopping is not the same as autonomous commerce

This distinction matters. A visit from ChatGPT, Gemini or another assistant is AI-referred commerce. A recommendation or service interaction that contributes to an order is AI-influenced commerce. An agent that acts within explicit permissions to assemble and complete a transaction is agent-led commerce.

Adobe Analytics provides strong evidence for the first category. Based on more than one trillion visits to US retail sites, Adobe reported that traffic from generative AI sources increased 1,200% between July 2024 and February 2025. Its companion survey found that these visitors were highly engaged, although AI referrals still represented a small share of total traffic at that point. Read Adobe’s analysis.

That is evidence of a fast-growing discovery channel. It is not evidence that autonomous agents completed those purchases. Brands should measure the categories separately rather than relabelling all AI-assisted journeys as agentic transactions.

The readiness gap in four numbers

The most useful industry benchmark comes from Visa Acceptance Solutions and PYMNTS Intelligence because it compares consumers, merchants and payments providers in the same study. The 2026 Agentic Commerce Deep Dive surveyed 5,241 consumers, 1,185 merchants and 150 acquirers across the United States, Brazil and the United Arab Emirates in early 2026.

Industry signal	2026 finding	What it means
Shoppers using AI in recent purchase research	48%	AI is already influencing discovery at scale.
Merchants with agent-ready structured catalogue data	15%	Most product information is not yet prepared for reliable machine interpretation.
Merchants able to distinguish AI traffic from human traffic	23%	Measurement is lagging adoption.
Consumers willing to let an agent search and compare	56%	Delegation is strongest early in the journey.

The trust gradient becomes steeper near payment. More than half of consumers would let an agent search and compare products, but fewer than four in ten would give one access to saved payment credentials. Consumers named clear liability and certified agents as trust signals; merchants prioritised protection against unauthorised transactions.

This is why a brand can be “using AI” and still be unprepared for agentic commerce. Adoption of an internal AI tool, a website assistant or AI-generated copy does not establish that an external agent can retrieve accurate facts, respect a customer’s permissions and complete a governed action.

A practical model for brand readiness

Apirro organises readiness into four connected layers. This is Apirro’s interpretation of the industry evidence, not a score produced by the external studies.

1. Discoverable: can agents find and select the brand?

Discovery increasingly happens before a shopper reaches a brand website. Accenture’s 2026 research found that 90% of frequent AI users in North America would consider switching from a preferred brand if an assistant presented a better alternative. It also projected that up to 45% of shoppers could shift at least half of their commerce activity into agent-mediated environments within two years. The latter is a forecast, not a measured outcome. Read Accenture’s agentic commerce report.

For brands, discoverability now includes:

crawlable, server-readable pages;
clear entity and product information;
authoritative third-party mentions and citations;
current sitemaps and canonical URLs;
content that directly answers real buying questions;
visibility monitoring across major AI assistants.

Traditional SEO remains important because search indexes, trusted publications and structured web content often supply the evidence AI systems use. But ranking in a conventional results page does not guarantee inclusion in an AI-generated shortlist.

2. Understandable: can agents interpret the offer correctly?

This is where the sharpest operational gap appears. Visa’s finding that only 15% of merchants have agent-ready structured data means most merchants still expose product information primarily for human browsing rather than machine decision-making.

An agent needs facts, not just persuasive copy. Titles, identifiers, variants, prices, availability, delivery conditions, returns, warranties and eligibility rules must agree across the product page, structured data, catalogue feed and any API.

A beautiful product page can still be difficult for an agent if important attributes are embedded in images, loaded only after interaction, expressed inconsistently or omitted from structured fields. Likewise, schema markup cannot compensate for stale inventory or contradictory policies.

3. Transactable: can agents act safely and with permission?

The payments industry is building the infrastructure, but deployment remains early. Visa reported in June 2026 that only 19% of merchants had solutions in place to accept agent-initiated payments, while 63% were exploring or implementing them. Read Visa’s assessment of agent payments.

Standards are also emerging. Google launched the open Agent Payments Protocol with more than 60 organisations across payments, commerce and technology to support verifiable mandates, credentials and accountability. Read Google’s AP2 announcement. OpenAI and Stripe introduced the Agentic Commerce Protocol alongside Instant Checkout, initially supporting purchases from US Etsy sellers and announcing expansion toward more than one million Shopify merchants. Read OpenAI’s Instant Checkout announcement.

These launches prove that transaction rails are being built. They do not prove universal merchant adoption, interoperability or consumer trust. Brands still need clear rules for authentication, consent, spending limits, fraud, returns, exceptions and human handoff.

4. Measured: can the brand see what agents are doing?

Only 23% of merchants in the Visa/PYMNTS study could distinguish AI-driven traffic from human traffic. Salesforce separately found that only 32% of commerce organisations had fully defined AI success metrics and key performance indicators.

Measurement should go beyond visits. A useful operating view connects:

prompts and recommendation visibility;
citations and source pages;
agent or tool requests;
product matches and exclusions;
cart and checkout handoffs;
completed orders, failures and returns.

Without those links, a brand may know that AI traffic increased but not which assistant sent it, which answer shaped the visit, why an agent rejected a product or whether an agent-assisted journey created incremental value.

What the industry evidence says about readiness

No credible public source currently provides one representative, cross-industry score for brand readiness across discovery, data, transaction and measurement. It would therefore be misleading to claim that “brands are 40% ready” or that a particular percentage will succeed.

The evidence supports a more precise conclusion.

1. Demand is ahead of infrastructure

Consumers are already using AI for research at scale: 48% in the Visa/PYMNTS tri-market study and 58% replacing search with generative AI for recommendations in Capgemini’s 12-country study. Merchant foundations are much less mature: 15% had agent-ready structured catalogue data and 23% could identify AI traffic.

2. Readiness falls as autonomy rises

Consumers are more willing to delegate search and comparison than access to payment credentials. Merchant deployment follows the same curve: many are exploring agentic payments, but only 19% reported having acceptance solutions in place. Discovery is current behaviour; delegated transaction is an emerging capability.

3. Visibility without data quality is fragile

A model may mention a product using public content, but an agent needs current, consistent facts to decide whether it satisfies a request. Inaccurate price, availability, variant or policy data can remove a brand from consideration even when the brand is technically crawlable.

4. Measurement is a strategic blind spot

When fewer than one in four merchants can distinguish AI-driven traffic, most cannot yet build a reliable commercial feedback loop. Teams need attribution definitions before they can compare AI-referred, AI-influenced and agent-completed journeys.

5. Transaction standards are progressing faster than operating models

AP2, the Agentic Commerce Protocol and network-level agent-payment programmes show momentum. The harder work is organisational: deciding which agents to trust, what they may access, what evidence they must provide and who is responsible when an action is wrong.

Six priorities for commerce teams

Priority 1: establish separate baselines

Measure recommendation visibility, technical discoverability, catalogue quality, transaction capability and business outcomes independently. A single blended score can conceal the weakest layer.

Priority 2: make product and service facts consistent

Create one reliable source for titles, identifiers, attributes, variants, price, availability, delivery, returns and eligibility. Reconcile the website, structured data, feeds and APIs continuously.

Priority 3: make important content easy to extract

Put essential information in the initial page response, use descriptive headings and semantic markup, and avoid hiding decisive details only in imagery, downloads or interactions.

Priority 4: expose live information through governed interfaces

Structured data supports interpretation. Maintained feeds and APIs support freshness. A governed agent interface supports controlled action. Each layer solves a different problem.

Apirro connects these through AI Discovery, Agent Optimiser, Agent Link™ and Monitoring & Insights. Agent Link™ provides one governed integration point for multiple assistants and agents, while white-label agents let enterprises, agencies and payment providers deliver an agent experience under their own brand.

Priority 5: design trust before checkout

Define identity, consent, permissions, spending limits, confirmation steps, liability, fraud controls, cancellation and human escalation before allowing an agent to transact. Start with low-risk tasks and expand autonomy only when evidence supports it.

Priority 6: instrument the full agent journey

Track which assistants appear, what customers ask, which sources are cited, what agents retrieve, why products qualify or fail, and what happens after handoff. Use consistent attribution definitions so reported growth is meaningful.

Methodology and limitations

This article is a synthesis of external industry research, not an Apirro customer study. It draws primarily on reports and announcements published in 2025 and 2026 by Visa Acceptance Solutions with PYMNTS Intelligence, Salesforce, Capgemini Research Institute, Adobe Analytics, Accenture, Google and OpenAI.

The cited evidence includes:

consumer and merchant surveys;
aggregated digital-behaviour data;
merchant and commerce-professional surveys;
provider announcements about standards and product availability;
forecasts and stated adoption plans.

These evidence types answer different questions. Survey responses measure reported behaviour, attitudes or plans. Analytics platforms observe activity within their own coverage. Product announcements establish availability, not adoption or performance. Forecasts are scenarios, not recorded outcomes.

Important limitations:

Definitions of “AI traffic,” “agentic search,” “AI-influenced orders” and “agent-led payments” vary by source.
Adobe’s strongest behavioural figures concern US retail traffic; Salesforce measures organisations and shoppers represented in its datasets; Visa/PYMNTS covers the United States, Brazil and the UAE; Capgemini’s survey spans 12 countries.
Vendor studies may reflect the coverage and commercial context of the organisation producing them.
Growth percentages may start from a small base and do not reveal total channel share unless the source reports it.
Consumer willingness does not guarantee actual delegation, and merchant plans do not guarantee deployment.
No cited study proves that any individual technical change causes an AI model to cite a brand or a consumer to buy.

For those reasons, this report does not combine the studies into a universal market-readiness score. It uses converging evidence to identify where consumer behaviour, merchant capability and transaction infrastructure are moving at different speeds.

The 2026 conclusion

Agentic commerce is neither a distant concept nor a finished channel.

AI-assisted discovery is already material. Customers are using assistants to research products, compare options and shape shortlists. The industry is also building the protocols and payment controls needed for agents to act. But merchant readiness remains uneven where it matters most: structured catalogue data, trustworthy live information, measurement, governance and payment acceptance.

The brands best positioned for the next phase will not be those that simply add the word “AI” to their commerce strategy. They will be the ones that become:

discoverable when an assistant forms the shortlist;
understandable when an agent evaluates the offer;
transactable when a customer delegates the next step; and
measured when the business decides what to improve.

The opportunity is not just to attract another source of traffic. It is to become a brand that machines can confidently find, interpret, choose and act with.

See how ready your brand is

Run Apirro’s free agent-readiness report to measure your Visibility, Understandability and Transactability signals and receive a prioritised fix plan.

One URL. About ten seconds. No account is needed to see your scores.

Industry sources

This report draws on the following primary research and infrastructure announcements:

Capgemini Research Institute: What Matters to Today’s Consumer 2025 — 12,000 consumers across 12 countries.
Visa Acceptance Solutions and PYMNTS Intelligence: 2026 Global Digital Shopping Index — Agentic Commerce Deep Dive — 5,241 consumers, 1,185 merchants and 150 acquirers across the United States, Brazil and the United Arab Emirates.
Salesforce: Fourth State of Commerce report — 3,450 commerce professionals, 4,690 consumers and behavioural data from more than 1.5 billion shoppers.
Adobe Analytics: generative AI traffic to US retail sites — aggregated analysis of more than one trillion visits.
Accenture: Agentic commerce — Make your brand unmissable — consumer adoption, brand switching and agent-mediated commerce research.
Google: Agent Payments Protocol — open transaction protocol developed with more than 60 organisations.
Stripe: Instant Checkout and the Agentic Commerce Protocol — primary announcement of the OpenAI and Stripe commerce standard and initial merchant rollout.

Frequently asked questions

What is agentic commerce readiness?

Agentic commerce readiness is a brand’s ability to be found, correctly interpreted and safely used by AI agents across discovery, evaluation and transaction workflows. It includes content and structured data, but also live catalogue access, policies, permissions, transaction interfaces and measurement.

Are AI-referred visits the same as agentic transactions?

No. An AI-referred visit starts when an assistant sends a person to a website. An AI-influenced order may involve recommendations or service interactions. An agentic transaction involves software taking an authorised action in the purchase process. These should be measured separately.

Is structured data enough to make a brand agent-ready?

No. Structured data helps machines interpret first-party facts, but it must match current catalogue and policy information. Brands also need authoritative visibility, governed access to live information, transaction controls and measurement.

Does agent readiness guarantee a brand will appear in ChatGPT or Gemini?

No. Technical readiness reduces barriers, but recommendations also depend on the user’s question, the provider’s retrieval system, independent authority, freshness and other factors outside a brand’s control.

How often should a brand test readiness?

Test after major platform, theme, catalogue, structured-data or policy changes, and on a regular schedule. Agent interfaces, standards and crawler behaviour are evolving quickly, so a one-time audit can become stale.