How AI agents are moving from search to checkout, the six payment rails built to let them buy, how they choose products, and the playbook for becoming the brand an agent buys from.
250+
merchant storefronts audited for agent readiness
6
agent payment rails launched in roughly 12 months
$20.9B
agent-influenced commerce GMV in 2026
64
pages of data, protocols, and the merchant roadmap
Published by MaximusLabs.ai Revenue-focused Generative Engine Optimization
Primary data window: 2024 to Q2 2026 Base: 250+ storefronts, consumer delegation survey
The State of Agentic Commerce 2026Contents
What is inside
Six chapters, one question: will an agent buy from you.
This report moves from the behavior shift to the payment rails, to how agents pick products, to trust and fraud, to the merchants already winning, and finally to a twelve month roadmap. Every exhibit cites its source. Every chapter ends with what it means for your storefront.
For ten years we optimized for the click. The click is going away.
I have spent my career helping brands win a moment that is about to disappear: the moment a human looks at a screen, weighs a few options, and clicks. In agentic commerce, that moment is delegated to a model. No homepage. No shelf. No ad. A buyer says "reorder my running shoes" or "find me the best sleep mask under forty dollars," and an agent goes and does it.
When that happens, the entire funnel collapses into a single question: does the agent trust you enough to buy you? There is no second page of results to rescue you. There is not even a click to win. The agent returns one cart, or a shortlist of two or three, and everything else is invisible.
This is the same truth we have been writing about in AI search, only sharper. In generative search, being left out of the answer costs you a visit. In agentic commerce, being left out of the answer costs you the sale, the payment, and the reorder, all at once. The stakes moved from attention to revenue.
Most teams are preparing for this backwards. They are asking which protocol to support, as if the hard part were plumbing. The plumbing matters, and we cover all six rails in this report. But the rails only decide whether an agent can transact with you. They do nothing to decide whether an agent chooses you. That decision is made earlier, in the data the model reads and the trust it has already formed about your brand. Build that, and the agent has to recommend you. Skip it, and no protocol integration will save you.
The MaximusLabs View
Either the agent buys you, or it buys your competitor. There is no page two, and now there is not even a click.
An agent does not browse. It borrows the buyer's trust and spends it on whoever it already trusts. Your job is to be that brand, in machine-readable form, before the question is ever asked.
Stop optimizing to be found. Start engineering to be bought.
The State of Agentic Commerce 202600 · Executive Summary
Executive Summary
Demand is ready. Rails are ready. Merchants are not.
Across 250+ merchant storefronts and a consumer delegation survey, one pattern dominates 2026: shoppers are willing to hand the purchase to an agent, the payment infrastructure to let them exists, and almost no catalog is built for a machine to actually transact against. That gap is the opportunity.
The numbers that frame the year
71%
of consumers would let an AI agent complete a purchase for routine or repeat orders
Delegation survey
7%
of merchants have a catalog and checkout an agent can transact against today
250+ storefront audit
3.6x
higher checkout completion via an agent-ready protocol versus a standard web flow
Flow benchmark
294%
year over year growth in agent-influenced commerce GMV, 2025 to 2026
$5.3B to $20.9B
The five findings
1
Delegated buying is mainstream intent, not a fringe experiment.
71% of consumers say they would let an agent complete a routine or repeat purchase. Willingness is highest where choice is low-stakes and repeat: groceries, refills, travel rebooking, and commodity electronics.
2
The readiness gap is enormous, and it is a moat for whoever closes it first.
Only 7% of merchants expose a catalog and checkout an agent can actually complete. The other 93% are betting their best customers will keep clicking. The data says the opposite is happening.
3
Agent-ready flows convert dramatically better than agents bolted onto a human web flow.
A purpose-built protocol path completes 3.6 times more often than handing an agent a standard checkout. Pop-ups, CAPTCHAs, and multi-step carts are where agent purchases die.
4
The rails arrived faster than anyone adapted to them.
Six payment rails launched in roughly twelve months, from OpenAI and Stripe's Agentic Commerce Protocol to Google's AP2, Visa, Mastercard, Amazon, and PayPal. Infrastructure is no longer the bottleneck. Merchant data is.
5
Discovery is consolidating, and there is no page two.
A handful of assistants now mediate most agent purchase intent, with ChatGPT alone driving the majority. The agent returns one cart or a shortlist of two to three. Trust, formed before the question, decides who is on it.
The MaximusLabs ViewKrishna Kaanth M, Founder
The bottleneck moved. It is not whether an agent can pay you. It is whether it has any reason to choose you.
Every team is about to spend the next year on protocol integrations. Necessary, but not sufficient. The rails are a commodity the moment everyone has them. The durable advantage is the same one that wins AI search: structured, machine-readable proof that you are the trusted answer. Build the brand the model already believes in, and the protocol is just the cash register.
The classic funnel assumed a human at every step. Agentic commerce removes the human from the middle and keeps them only at the edges: the intent, and the approval.
The shift
Six stages of browsing, comparing, and clicking are collapsing into one delegated decision. When the agent does the shopping, the only thing it can act on is data and trust it already holds.
The State of Agentic Commerce 202601 · Delegated Buying
The Shift to Delegated Buying
A $20.9B channel today, on track to clear $230B by 2030.
Agent-influenced commerce GMV grew nearly fourfold in a single year. The observed curve is steep, and the projection is steeper, because each new payment rail and each newly agent-ready catalog compounds the next.
Exhibit 1.1
Agent-influenced commerce GMV, observed then projected.
USD billions. Solid line is observed, 2024 to 2026. Dashed line is the MaximusLabs projection. Toggle a scenario.
Observed, 2024 to 2026
Base case: ~49% CAGR, agentic GMV reaches $232B by 2030
Source: MaximusLabs analysis synthesizing public agentic and AI-commerce market estimates (eMarketer, Adobe Analytics, Salesforce, Bain). Observed anchors: $1.2B (2024), $5.3B (2025), $20.9B (2026). Projections are MaximusLabs scenarios, not observed data.
Exhibit 1.2
Six steps for a human become one decision for an agent.
The traditional funnel versus the delegated path. The middle, where merchants spend most of their budget, disappears.
6
human steps: awareness, consideration, click, compare, cart, checkout
→
2
edges that remain: the buyer states intent, the buyer approves the result
Edge 1
Intent
A buyer states a goal in natural language.Human in control
→
The collapse
The agent shops
Discovery, comparison, and cart-building happen inside the model, on data, in seconds.No human, no UI
→
Edge 2
Approval
The buyer confirms, or pre-authorized the agent to confirm.Human in control
Source: MaximusLabs model of agentic purchase flows observed across assistant shopping integrations, 2025 to 2026.
294%
Year over year growth in agent-influenced GMV, 2025 to 2026
MaximusLabs analysis
71%
Of consumers would delegate a routine or repeat purchase to an agent
Delegation survey
62%
Of agent shopping queries we observed originate inside ChatGPT
Origin analysis, see Ch.3
What this means for your storefront
Every dollar you spend on the middle of the funnel, the landing pages, the retargeting, the persuasion copy, is spent on a stage the agent skips. The budget does not disappear; it moves. It moves to the data the agent reads and the trust the agent holds. Chapters 3 and 5 show exactly where.
In roughly twelve months, the largest names in payments shipped six different ways for an agent to hold intent, prove authorization, and settle a transaction.
Why it matters
The rails decide whether an agent can buy from you. They are converging fast, and they are becoming a commodity. The advantage is not which rail you pick. It is being ready before your category is.
The State of Agentic Commerce 202602 · Payment Rails
The Rails, Compared
Six rails, two philosophies: delegated tokens and signed mandates.
Every protocol answers the same question, how does a merchant know an agent is authorized to spend a buyer's money. The card networks extended tokenization. The model makers built new intent layers. The table below is the field as of Q2 2026.
Exhibit 2.1
The six agent payment rails and what each one actually standardizes.
Backers, launch window, the core mechanism, and how much work it asks of a merchant to support.
Protocol & backer
Launched
What it standardizes
Merchant lift
Status
Agentic Commerce Protocol OpenAI + Stripe
Sep 2025
A shared cart and a delegated payment token passed from assistant to merchant
Medium
Live
Agent Payments Protocol (AP2) Google + 60 partners
Sep 2025
Cryptographically signed mandates that prove buyer intent across any rail
Medium
Live
Intelligent Commerce Visa
Apr 2025
Tokenized agent credentials tied to a real Visa card and spend controls
Low
Live
Agent Pay Mastercard
Apr 2025
Agentic tokens and a trust program identifying approved AI agents
Low
Live
Buy for Me Amazon
2025
An in-app agent that completes purchases on third-party brand sites
Medium
Limited
Agent Toolkit & checkout PayPal
2025
An agent SDK plus wallet-backed settlement and buyer protection
Low
Live
Source: MaximusLabs synthesis of public protocol documentation and announcements from OpenAI, Stripe, Google, Visa, Mastercard, Amazon and PayPal, 2025 to Q2 2026. "Merchant lift" is MaximusLabs' estimate of integration effort.
Exhibit 2.2
A twelve month sprint: how the rails arrived.
The card networks moved first on credentials. The model makers followed with intent layers. Then reality set in.
April 2025
The networks move first
Visa Intelligent Commerce and Mastercard Agent Pay extend tokenization to agents, betting their existing rails are the safest path.
Mid 2025
Wallets and marketplaces follow
PayPal ships agent tooling; Amazon "Buy for Me" lets its assistant purchase across third-party sites, putting a marketplace inside the agent.
September 2025
The protocol wave
OpenAI and Stripe publish the Agentic Commerce Protocol; days later Google launches AP2 with 60+ partners. Intent becomes a standard, not a feature.
Late 2025 to 2026
The reality check
Early instant-checkout launches see limited merchant rollout and public walkbacks. The gap between an announced protocol and a catalog an agent can actually buy from becomes the real story.
Source: MaximusLabs timeline compiled from primary announcements and subsequent reporting, April 2025 to Q2 2026.
The State of Agentic Commerce 202602 · Payment Rails
How Authorization Works
Underneath every rail is the same four-step handshake.
Strip away the branding and all six protocols do the same thing in the same order. The buyer delegates, the agent proves it is authorized, the rail transacts within limits, and everyone gets a record they can reconcile.
Exhibit 2.3
The agent authorization handshake, rail by rail, is one pattern.
Delegated tokens and signed mandates are two ways to do step two. The rest is shared.
Step 1
Delegate
The buyer grants the agent authority with explicit limits: budget, category, and time.Output: a scoped grant
→
Step 2
Prove
The agent presents a signed mandate or tokenized credential. The merchant verifies an approved agent acting for a real buyer.Output: verified intent
→
Step 3
Transact
The agent submits the cart. The rail authorizes payment against the token, strictly inside the buyer's limits.Output: an authorized charge
→
Step 4
Settle
Funds move, the merchant fulfills, and both sides hold a reconcilable record. Disputes route back to the rail.Output: a clean ledger
Source: MaximusLabs reference model abstracted from OpenAI/Stripe ACP, Google AP2, Visa and Mastercard agent credential flows, 2025 to 2026.
6
Distinct agent payment rails went live in roughly twelve months
Ch.2 protocol set
2
Competing authorization models: delegated tokens and signed mandates
MaximusLabs analysis
93%
Of merchants are not yet transactable by any agent, rail or no rail
250+ storefront audit
Reality check, read this before you pick a rail
The announcements ran ahead of the rollout. Several high-profile instant-checkout launches shipped to a handful of merchants, then quietly narrowed or paused. Treat a protocol's existence as necessary, not sufficient. The question is not "which rail launched," it is "which rail are my buyers' agents actually using to reach my category, today." Build for that one first, keep the others a config change away.
The MaximusLabs ViewKrishna Kaanth M, Founder
A rail is a cash register. Owning the best cash register has never once made a customer choose your store.
Within a year, supporting these protocols will be table stakes, the way SSL or a mobile-responsive site became table stakes. Nobody wins on plumbing once everybody has plumbing. The teams that treat integration as the finish line will have built a beautiful checkout that no agent has any reason to reach. Spend the minimum to be transactable, then spend everything else on the reason to be chosen.
An agent never sees your homepage. It reads a feed, weighs a handful of machine-legible signals, and returns a shortlist before a human ever looks. This is the new shelf, and it is invisible.
Why it matters
The signals that decide the agentic shortlist are, almost exactly, the signals merchants invest in least. The brands that close that gap first will be the default answer in their category before competitors realize the shelf changed.
The State of Agentic Commerce 202603 · The Invisible Shelf
The Ranking-Signal Inversion
The signals an agent weights most are the ones merchants are least ready for.
We scored the inputs an autonomous shopping agent uses to build a recommendation against how many of 250+ audited storefronts are actually strong on each. Toggle the two views below. The ranking flips almost perfectly, the highest-leverage signals sit where the fewest merchants are competitive.
Exhibit 3.1
What agents weight, versus what merchants are good at, are nearly opposite lists.
Switch the metric to re-rank the bars. "Agent weight" is the signal's influence on the shortlist; "merchants strong" is the share of audited storefronts that score well on it.
Structured data & schemaProduct, Offer, Review markup the agent can parse
92%agent weight
Review volume & recencyRating distribution and how fresh the reviews are
85%agent weight
Brand authority & LLM citationsWhether models already name you in the category
78%agent weight
Real-time inventory & priceLive availability and accurate, current pricing
71%agent weight
Return & warranty clarityPolicies the agent can read and compare on risk
58%agent weight
Price competitivenessWhere the offer lands against the comparison set
Source: MaximusLabs analysis of agentic product-selection behavior across ChatGPT, Gemini and Perplexity shopping flows, mapped against an audit of 250+ D2C and retail storefronts, Q1 to Q2 2026. Weights are MaximusLabs estimates of relative signal influence.
Exhibit 3.2
The agent throws away most of what a conversion team optimizes.
Two columns, one buyer. Everything on the left is built for human eyes the agent never uses; everything on the right is what it actually consumes.
✕ What the agent never sees
Hero images and lifestyle photography rendered for a visitor who never loads the page
A/B-tested CTAs, button color and copy tuned for a click that no longer happens
Pop-ups, urgency banners and exit-intent aimed at a human attention span
Retargeting and remarketing pixels chasing a session the agent never opened
Page-two SEO and keyword density in a world with no page two, and no click
Funnel and landing-page UX polish below a fold the agent does not scroll
✓ What the agent actually reads
Structured data and schema completeness across Product, Offer and Review
Review volume, rating distribution and recency as a trust and quality signal
Brand authority and citations across LLMs, whether models already name you
Real-time inventory, price and availability feeds it can act on with confidence
Return, shipping and warranty terms expressed in machine-readable form
Specification and attribute coverage deep enough to answer a precise query
Source: MaximusLabs storefront audit and agent-readability framework, 2026. Mapping reflects which assets are exposed to an agent operating over a product feed versus rendering a page.
The gap, in one line
The three signals agents weight most, structured data, review depth and LLM brand authority, are exactly where audited merchants are weakest, with fewer than one in four scoring strong. The one thing most merchants are good at, imagery and page design, is the signal an agent values least. Closing that inversion is the entire job of the next two chapters.
The State of Agentic Commerce 202603 · The Invisible Shelf
Where Agent Demand Comes From
Most agent demand flows through two assistants, and they cite the same short list of sources.
If the shelf is invisible, these two charts are the floor plan. The first shows where agentic shopping sessions begin. The second shows what those agents actually pull from when they justify a recommendation. Hover any segment to isolate it.
Exhibit 3.3
Two assistants originate nearly four in five agent shopping sessions.
Share of agent-led shopping sessions by originating assistant. Concentration this high means a handful of surfaces decide who gets discovered.
ChatGPT62%
Gemini / Google AI Mode16%
Perplexity11%
Microsoft Copilot6%
Amazon Rufus & other5%
Source: MaximusLabs analysis of agent-led shopping session origin, synthesizing platform usage data and observed checkout flows, Q2 2026. Shares are directional estimates for shopping intent, not general assistant usage.
Exhibit 3.4
When an agent justifies a pick, it cites sources a brand can shape.
Distribution of source types agents reference when recommending a product. Owned pages and earned reviews together outweigh everything a brand cannot touch.
Brand & retailer PDPs (owned)34%
Marketplaces (Amazon, Walmart, Target)24%
Community & reviews (earned)21%
Editorial & publisher reviews13%
Comparison & aggregator sites8%
Source: MaximusLabs citation analysis of agent product recommendations across ChatGPT, Gemini and Perplexity, Q1 to Q2 2026. "Owned" and "earned" tags reflect a brand's degree of direct influence.
78%
Of agent shopping sessions originate in just two assistants, ChatGPT and Gemini
MaximusLabs analysis
55%
Of what agents cite is brand-influenceable: owned product pages plus earned reviews
Citation analysis
79%
Of all citations come from three source types: owned PDPs, marketplaces and community
Citation analysis
The MaximusLabs ViewKrishna Kaanth M, Founder
The agent does not browse. It verifies. Your only job is to be the brand it can verify fastest, and trust most.
For twenty years we optimized to be found: keywords, rankings, the race to page one. There is no page one anymore. There is not even a click. The agent arrives already knowing the category and asks a narrower question, who here can I trust. You answer that in three places it actually reads: structured data clean enough to parse without guessing, reviews deep and recent enough to defend a choice, and a brand the model has seen named so often it treats you as the safe default. None of that is bought. All of it is engineered. Stop optimizing to be found. Start engineering to be bought.
When the buyer is an algorithm, trust stops being a feeling and becomes a protocol. The question shifts from "is this card stolen" to "is this agent authorized, and who pays if it is wrong."
Why it matters
The merchants who get trust controls right will safely accept agent traffic that competitors are forced to block. The real integration decision is not which rail, it is which liability model you can live with.
The State of Agentic Commerce 202604 · Trust & Fraud
The Risk Surface
The fraud surface moves from the stolen card to the stolen mandate.
Tokenization largely solved card theft. Agentic commerce opens a new surface a step earlier, in the authorization itself. The dangerous failure is no longer a leaked number; it is an agent acting with authority it should never have had, or intent no one can prove after the fact.
Exhibit 4.1
The highest-severity agentic risks live in authorization, not payment.
MaximusLabs agentic risk index, 0 to 100, weighting likelihood, blast radius and how hard the vector is to detect after settlement.
Over-scoped mandateauthority beyond intent
92
Prompt injectionmanipulated agent
84
Synthetic agentsimpersonated identity
71
Unprovable intentno signed mandate
63
Dispute ambiguitywho owns the chargeback
58
Catalog poisoningmanipulated price or feed
44
Source: MaximusLabs agentic risk index, synthesizing protocol threat models from OpenAI/Stripe ACP, Google AP2, Visa and Mastercard, plus merchant interviews, Q2 2026. Scores are MaximusLabs estimates.
Exhibit 4.2
Four controls contain almost the entire surface, and they nest.
Each control closes the vectors the previous one leaves open. Skip one and you reopen a column of Exhibit 4.1.
Control 1
Scope
Issue mandates that are narrow by default: budget, category, merchant and time window, never open-ended authority.Contains: over-scoped mandates
→
Control 2
Verify
Accept only agents whose identity is cryptographically attestable and enrolled in a trust program.Contains: synthetic agents, injection
→
Control 3
Cap
Enforce live spend limits and velocity rules at the rail, so a compromised session cannot run away.Contains: runaway spend
→
Control 4
Audit
Keep a signed, reconcilable record of intent and authorization for every order, so disputes resolve on evidence.Contains: dispute ambiguity
Source: MaximusLabs trust-control reference model abstracted from public agent-authorization specifications, 2025 to 2026.
The unsettled question
No rail has fully answered liability. If an agent buys the wrong item, or buys correctly but the buyer disputes it, the chain of responsibility, buyer, agent platform, rail, merchant, is still being written in real time. Until it settles, treat a clean, signed audit trail as your single most valuable asset in any dispute. The merchant with evidence wins; the merchant with a log file does not.
The State of Agentic Commerce 202604 · Trust & Fraud
Controls and Liability
Every rail answers liability differently, and that is your real integration decision.
The protocols look similar on trust mechanics and diverge sharply on who absorbs a bad transaction. Read the last column first. It decides how much risk you are quietly accepting when you switch a rail on.
Exhibit 4.3
Same handshake, different bag-holder, by rail.
How trust is proven, what limits spend, and where the loss tends to land when an agentic transaction goes wrong.
Protocol
How trust is proven
Primary spend control
Loss tends to land on
ACP OpenAI + Stripe
Delegated payment token issued per cart
Token scoped to a single cart
Merchant / PSP
AP2 Google
Cryptographically signed buyer mandate
Constraints encoded in the mandate
Provable intent
Intelligent Commerce Visa
Tokenized agent credential on a real card
Issuer spend and velocity controls
Network / issuer
Agent Pay Mastercard
Agentic token plus agent trust program
Issuer spend and velocity controls
Network / issuer
Buy for Me Amazon
Amazon-mediated, behind its own identity
Amazon-governed limits
Platform-mediated
Agent Toolkit PayPal
Wallet authentication plus buyer protection
Wallet limits and risk engine
Wallet / protection
Source: MaximusLabs synthesis of public protocol documentation, 2025 to Q2 2026. "Loss tends to land on" is MaximusLabs' read of each rail's current dispute posture, not legal advice, and is evolving.
4
Controls, scope, verify, cap and audit, contain the bulk of the agentic risk surface
Exhibit 4.2
2
Trust primitives in market: delegated payment tokens and signed buyer mandates
MaximusLabs analysis
1
Question no rail has closed: who is liable when an authorized agent buys the wrong thing
MaximusLabs analysis
The MaximusLabs ViewKrishna Kaanth M, Founder
An agent will always prefer the merchant it can trust without asking a human. Be that merchant, and you are on the shortlist before price is even discussed.
This is trust transfer, and it is the whole game. The agent borrows trust from sources it already believes, then lends it to the merchant who makes verification effortless. Clean structured data, a signed and reconcilable order trail, enrollment in the trust programs, none of it is glamorous, and all of it compounds. The brands that treat trust as infrastructure, not a banner that says "secure checkout," will quietly become the default the agent reaches for while everyone else is still arguing about which rail to support.
This is not a forecast you can wait out. Smaller brands are already the default answer in categories owned, on paper, by giants. The agent does not award the shortlist to whoever is biggest. It awards it to whoever is most legible and most trusted.
Why it matters
If a sleep-mask startup and a boutique security firm can out-cite incumbents many times their size, the moat was never budget. It was readiness, and readiness is something you can build this quarter.
The State of Agentic Commerce 202605 · Who Is Winning
Proof, Not Theory
Small brands are already out-citing billion-dollar incumbents, because the agent does not care who is bigger.
Three MaximusLabs engagements, three categories, one pattern: the smaller player won the agentic shelf by being the easier brand for a model to read, trust and cite. None of them outspent anyone.
Case Study 01
Oliv AI overtakes a billion-dollar incumbent on citations
Challenge
Win AI visibility in a category dominated by a billion-dollar incumbent with vastly larger brand spend and backlink authority.
Approach
Trust-first, citation-ready content engineered around the entities and questions buyers actually ask assistants, not keyword volume. Clean schema, defensible claims, primary sources.
Result
A 64% citation rate across AI engines within six months, more than double the incumbent's 30%, on a fraction of the budget.
64%
Oliv AI citation rate across AI engines
30%
Billion-dollar incumbent, same window
6mo
Time to overtake on citation share
Case Studies 02 & 03
The same pattern, two more categories
#1
Nidra Goods ranks number one simultaneously on Google, ChatGPT and Perplexity for "best sleep mask," a term it does not pay to own.
$10B+
UnderDefense is cited by assistants ahead of security incumbents many times its size, on the strength of trust signals, not ad budget.
<6mo
Typical time for a prepared challenger to overtake a larger incumbent on agentic citation share.
The throughline
Different categories, identical mechanism. Each winner made itself the brand a model could verify fastest and trust most, then let citation compounding do the rest. The incumbent's size, ad spend and legacy SEO were not assets in the agentic channel. They were not even visible.
The State of Agentic Commerce 202605 · Who Is Winning
The Conversion Lift
Agent-ready storefronts convert agent traffic at roughly 3.6 times the rate of standard ones.
Winning the citation gets you onto the shortlist. Being transactable is what turns that into revenue. Across segments, storefronts built for agents, clean feeds, live inventory, machine-readable terms, convert agent-led sessions at a multiple of conventional sites, which routinely stall an agent at checkout.
Exhibit 5.1
In every segment, agent-ready storefronts clear agent checkout far more often.
Conversion rate of agent-led shopping sessions, standard versus agent-ready storefront, by category. Hover any bar for the exact figure.
2.2%
7.9%
3.1%
11.2%
1.8%
6.5%
1.4%
5.0%
Apparel3.6× lift
Beauty & care3.6× lift
Home & lifestyle3.6× lift
Electronics3.6× lift
Standard storefront
Agent-ready storefront
Source: MaximusLabs analysis of agent-led session conversion across audited storefronts, by category, Q1 to Q2 2026. "Agent-ready" denotes clean structured feeds, live inventory and pricing, and machine-readable policies.
3.6×
Average conversion lift on agent traffic, agent-ready versus standard storefront
Exhibit 5.1
95%
Practical ceiling agent-ready storefronts approach on high-intent agent sessions
MaximusLabs analysis
4/4
Segments where the agent-ready storefront beat standard by at least three times
Exhibit 5.1
The MaximusLabs ViewKrishna Kaanth M, Founder
Budget buys reach. It has never once bought trust. In the agentic channel, only trust is on the shelf, and the shelf is small.
The incumbent's advantages, the ad budget, the decade of backlinks, the brand everyone recognizes, are real, and to the agent they are invisible. What the agent sees is whether your data is clean, your reviews are credible, and your store can actually complete the purchase it wants to make. A focused team can build all three faster than a giant can reorganize to care. That is the opening. It does not stay open forever. The brands that move while the category is still arguing about whether agents matter will be the default answer by the time it stops arguing.
Everything to here was diagnosis. This is the prescription: find where you sit today, follow a twelve-month plan to move, size the prize with your own numbers, and leave with five things you can start on Monday.
How to read this chapter
Locate yourself on the matrix first. Your quadrant decides which half of the plan matters most. Then run the calculator with your real traffic before you commit a rupee or a dollar of budget.
The State of Agentic Commerce 202606 · The Roadmap
Step One: Locate Yourself
Two questions place you on the board: can an agent read you, and does it already trust you.
Brand authority without data readiness makes you a Trusted Ghost, known, but impossible to transact. Data readiness without authority makes you a Challenger, ready, but not yet chosen. Only the top-right quadrant wins by default. Today, six in every hundred merchants are there. Hover a bubble for its profile.
Exhibit 6.1
Most merchants are invisible or ghosts; the Champion quadrant is nearly empty.
Agent Readiness Matrix. Bubble size is the estimated share of merchants in each quadrant. MaximusLabs proprietary framework.
Brand authority in LLMs →
High authority · Low dataTrusted Ghost
High authority · High dataAgentic Champion
Low authority · Low dataAgent-Invisible
Low authority · High dataData-Rich Challenger
61%Invisible
22%Trusted Ghost
11%Challenger
6%Champion
Agent data readiness →
Source: MaximusLabs Agent Readiness framework applied to an audit of 250+ D2C and retail storefronts, Q2 2026. Quadrant shares are MaximusLabs estimates and sum to 100%.
What your quadrant tells you to do
Trusted Ghost (high brand, low data): you have the hard part. Spend the next quarter almost entirely on data readiness, schema, feeds, transactability. Data-Rich Challenger (low brand, high data): you can be transacted but not chosen; invest in citations, reviews and authority. Agent-Invisible: you need both, in parallel, starting now. Champion: defend the lead, because the bubble in your corner is small and everyone else is aiming for it.
The State of Agentic Commerce 202606 · The Roadmap
Step Two: The Twelve-Month Plan
Readiness first, authority in parallel, defense last. The order is not optional.
You cannot earn citations for a store an agent cannot transact with, so data readiness leads. Authority compounds slowly, so it starts early and runs long. Trust controls go in before you scale agent traffic, not after. Hover any bar for its window.
Exhibit 6.2
A sequenced twelve-month path from Agent-Invisible to Champion.
Indicative workstream windows across twelve months. The early blocks unlock the later ones.
M1M2M3M4M5M6M7M8M9M10M11M12
Foundationschema & clean data
M1-M3
Transactabilityfeeds, inventory, one rail
M2-M5
Authoritycitations & reviews
M3-M9
Trust controlsmandates & audit trail
M5-M8
Measure & defendmonitor & iterate
M9-M12
Foundational, unlocks the rest
Core build, runs in parallel
Defend and compound
Source: MaximusLabs agentic commerce engagement model, 2026. Windows are indicative for a mid-sized merchant and compress with dedicated resourcing.
Step Three: Five Moves for Monday
1
Watch an agent try to buy from you.
Ask ChatGPT, Gemini and Perplexity to research and purchase your best-seller. Note exactly where it stalls, that failure point is your number-one priority.
2
Ship schema on your top 20 SKUs this week.
Complete Product, Offer and Review markup on your highest-revenue products. This is the single highest-weighted signal from Chapter 3, and the fastest to fix.
3
Open one clean, live product feed and pick one rail.
Real-time price and inventory, exposed to a single rail your buyers' agents actually use. Be transactable on one before you chase all six.
4
Get named in the sources agents cite.
Target the owned pages, review platforms and editorial that fill the Chapter 3 citation mix. Earned authority is what moves you up the matrix, not ad spend.
5
Stand up a signed order trail before you scale.
A reconcilable record of intent and authorization on every agent order. When liability is still unsettled, evidence is the asset that wins disputes.
The State of Agentic Commerce 202606 · The Roadmap
Step Four: Size the Prize
Put your own numbers in. The gap between standard and agent-ready is the cost of waiting.
This model applies the 3.6 times conversion lift from Chapter 5 to your agent traffic, capped at a realistic 95% ceiling. Drag the inputs to your reality. The annual figure is incremental revenue you are leaving on the table for every month you stay un-ready.
Exhibit 6.3
The agentic opportunity calculator.
Interactive. Set monthly agent-led visitors, your baseline conversion, and average order value.
MaximusLabs Model
What agent-readiness is worth to you
Monthly agent-led visitors12,000
Baseline conversion rate2.5%
Average order value$85
Annual incremental revenue
$796K
Monthly orders, standard storefront300
Monthly orders, agent-ready1,080
Incremental orders per month+780
Incremental revenue per month$66K
Model: agent-ready conversion = baseline × 3.6, capped at 95%. Lift derived from Exhibit 5.1. Figures are directional and meant for prioritization, not forecasting.
Source: MaximusLabs agentic opportunity model. Conversion lift from MaximusLabs analysis of agent-led session conversion, Q1 to Q2 2026.
The MaximusLabs ViewKrishna Kaanth M, Founder
The buyer is becoming an algorithm. The brands that win it are the ones deciding, this quarter, to be easy to trust and impossible to skip.
None of this roadmap is exotic. Clean data, real reviews, a checkout an agent can finish, a record you can defend. The reason it is a moat is not difficulty, it is that almost nobody has done it yet, ninety-four out of a hundred merchants are still optimizing a page the buyer will never see. The window where being ready is a genuine advantage is open right now, and it closes the moment your category wakes up. Pick one move from the Monday list. Start it today. Being early is the entire strategy.
The State of Agentic Commerce 202607 · Methodology and Sources
Methodology
How this report was built, and what to trust in it.
This report synthesizes a proprietary audit of agent-readiness across 250+ storefronts with public protocol documentation and market data from the primary sources below. Where a figure is a MaximusLabs estimate or projection, we say so on the chart, not in a footnote.
250+
D2C and retail storefronts audited for agent-readiness
MaximusLabs, Q1 to Q2 2026
6
Agent payment rails analyzed from primary documentation
Protocol set, Chapter 2
2024-26
Primary data window, with projections through 2030
Multi-source synthesis
18
Named primary sources, cited individually below
Full ledger included
Source ledger
Source
Publisher
Date
Informs
Agentic commerce market sizing
eMarketer, Morgan Stanley
2025-26
GMV trajectory, Exhibit 1.1 and Executive Summary
Consumer delegated-buying intent
McKinsey, Salesforce
2025-26
Willingness to delegate purchases, Chapter 1
Agentic Commerce Protocol & Instant Checkout
OpenAI, Stripe
Sep 2025
Rails comparison, Exhibits 2.1 to 2.3
Agent Payments Protocol (AP2)
Google
Sep 2025
Signed-mandate model, Chapters 2 and 4
Agent credentials and trust programs
Visa, Mastercard
Apr 2025
Tokenized credential rails, Exhibits 2.1, 4.3
Marketplace and wallet agents
Amazon, PayPal
2025
Buy for Me and wallet settlement, Chapter 2
Platform commerce integrations
Shopify, Perplexity, Google
2025-26
Channel availability, Chapters 2 and 3
AI traffic and conversion to retail
Adobe Analytics, Bain, BCG
2025-26
Conversion lift context, Exhibit 5.1
Agent-readiness audit and case studies
MaximusLabs (proprietary)
Q1 to Q2 2026
Signal weights, matrix, lift and cases, Chapters 3 to 6
Limitations and disclosures
Market sizing for agentic commerce varies widely by definition and scope; we reference multiple estimates to convey a credible range rather than a single point. Protocol capabilities are moving monthly, and several instant-checkout launches narrowed after announcement, our rails snapshot is current to Q2 2026. Signal weights, the risk index, quadrant shares and the conversion lift are MaximusLabs estimates derived from our audit, labeled as analysis wherever they appear. MaximusLabs client outcomes are proprietary and cannot be independently verified. Forward GMV figures beyond 2026 are scenario projections, not observed data, and are drawn as dashed lines throughout.
When agents do the buying, agent-readiness is the moat.
MaximusLabs is a full-stack AI growth marketing agency that turns AI search and AI commerce into revenue across Google, ChatGPT, Perplexity, Gemini, and Claude. Founded by Krishna Kaanth on the insight that each AI platform has its own algorithm, trust signals, and citation patterns, MaximusLabs pioneered Revenue-focused Generative Engine Optimization. As shopping shifts from humans browsing to agents transacting, we extend that practice into agent-readiness, making catalogs, feeds, trust signals, and checkout legible to the models that now decide what gets bought.
R-GEORAEOAgent-Readiness AuditsTrust-First OptimizationFeed and Catalog IntegrityOff-Page Authority