When the purchase completes inside the chat, your website stops being the conversion surface. This is what shifts — for attribution, for the customer relationship, and for the work itself.

1. The conversion surface has moved

For the entire history of ecommerce, the website was the place a sale happened. Every discipline — SEO, paid media, conversion-rate optimisation and, yes, link building — ultimately existed to deliver a qualified visitor to a product page and persuade them to buy there. Agentic checkout breaks that assumption. When a shopper asks an assistant for a recommendation and completes the purchase without leaving the conversation, the transaction occurs on a surface the merchant does not own and cannot design. The product page is no longer the destination; it is, at most, a data source the agent consulted on the way to a decision.

This is not a fringe behaviour. Shopify reported that AI-driven traffic to its stores grew roughly eightfold year over year, with orders from AI-powered searches rising by an order of magnitude and new-buyer orders from AI arriving at nearly twice the rate of other channels. Salesforce’s retailer survey found three-quarters of retailers expect AI agents to be essential. The hub article for this cluster, getting your products recommended by AI shopping agents, addresses how to be recommended in the first place. This article takes up what happens next — the transaction and everything it disturbs — and what it means for the people whose job is to build authority and links.

The most useful way to frame the change is a sentence now repeated across the industry: the brand site shifts from being the primary conversion surface to being a post-purchase hub. Discovery and checkout happen inside the AI interface; the site handles fulfilment, brand storytelling, returns, support and the ongoing relationship. Crucially, that relationship is not retained automatically. As the same analyses put it, the brands that consciously own the post-purchase layer keep the customer; the ones that treat agentic checkout as a passive pipe quietly surrender it.

Underlying all of this is a compression of the funnel. The classic journey — awareness, consideration, comparison, decision, purchase — unfolded across days and multiple sessions, with your marketing touching the buyer at each stage. The agent collapses those stages into a single conversational moment: it scans a broader assortment than any human would, compares on the shopper’s stated constraints, and presents a decision — sometimes with a buy button attached. Demand is captured at the instant it forms. That compression is why agent-acquired buyers convert with unusual intent when they do convert, and also why the familiar mid-funnel touchpoints, where brands used to do persuasion and data capture, simply do not occur. You are either present at the moment of decision or absent from it; there is little middle ground to work with.

2. A necessary reality check before you re-platform

It would be easy to read the section above as a call to throw resources at agentic checkout immediately. A more responsible reading is to calibrate. The honest data of early 2026 is that in-chat checkout, for all its strategic significance, currently under-performs a brand’s own checkout on conversion.

The strategic conclusion follows directly. Do not dismantle your own conversion funnel in favour of agentic checkout; the funnel still converts better today. But do not ignore the channel either, because it is capturing genuinely new demand — buyers who would otherwise never have reached you — and the flow will mature. The right posture is to make yourself fully recommendable and transactable now, while continuing to invest in the on-site experience and, above all, in the things agentic checkout puts at risk: attribution and the customer relationship. The rest of this article is about defending those two assets while the channel grows up.

The disintermediation question: who does the customer think they bought from?

Beneath the conversion mechanics sits a deeper question that should concern anyone responsible for brand equity. When an agent recommends a product, renders the cart and completes the purchase, the shopper’s felt experience is that they bought from the assistant — from ChatGPT, from Gemini — not from you. The agent becomes the trusted intermediary, and your brand risks receding into a fulfilment detail behind it. This is the same disintermediation that marketplaces inflicted on brands a decade ago, now arriving one layer higher up, at the point of recommendation rather than the point of listing.

The defence is not to fight the agent but to remain unmistakably present through it. Two things preserve brand salience. First, a distinctive, coherent brand entity — so that when the agent does name a source, it names you clearly, and so that the customer who liked the product can find and return to you directly next time. Second, a post-purchase experience strong enough that the customer’s memory attaches to your brand rather than to the assistant that brokered the transaction. A generic, unbranded fulfilment leaves the agent owning the relationship; a memorable one re-anchors it to you. This is why brand-building and authority work, far from being made redundant by agentic commerce, become the primary insurance against being abstracted into an interchangeable supplier.

3. How agentic checkout actually works

To reason about the consequences, you need a working picture of the mechanics. Two open protocols dominate, and a shared principle runs through both: the agent assembles intent and a cart, then hands a validated, signed transaction to the merchant, who fulfils it and remains the merchant of record.

The two protocols, in brief

OpenAI and Stripe’s Agentic Commerce Protocol (ACP) handles the checkout transaction inside ChatGPT through a small set of REST endpoints and a Stripe-led Shared Payment Token — a single-use, scoped, time-limited credential so the assistant never sees the buyer’s raw payment details. Google’s Universal Commerce Protocol (UCP) covers the wider lifecycle and is protocol-agnostic, with the Agent Payments Protocol (AP2) supplying the signed payment mandate. UCP’s April 2026 update is especially relevant to the conversion path: it added a Cart capability (agents can build multi-item carts from a single store), a Catalog capability (real-time variants, inventory and pricing), and Identity Linking (loyalty and member benefits that carry across platforms).

On platforms such as Shopify, the heavy lifting is handled for the merchant: payment processing, tax calculation, fraud detection and fulfilment all run through the platform, orders land in the normal admin with channel attribution, and — the point that matters most — the merchant remains merchant of record and the customer relationship stays with the merchant across every channel. The degree to which bespoke checkout customisations carry over varies by platform, which is itself a strategic consideration: the parts of your checkout that express your brand may simply not survive the handoff.

Agent verification: the trust layer underneath payment

A quieter but important development is agent identity. When software, not a person, presents a card, the payment networks need to know which agent is transacting and on whose authority — hence Visa’s Trusted Agent Protocol and Google’s AP2 mandate, which supply a signed “permission slip” the network can verify before money moves. For merchants this is mostly invisible plumbing, but it has two practical consequences worth noting. First, it is what makes autonomous checkout safe enough to scale, so expect adoption to accelerate as verification standards settle across Visa, Mastercard and the card rails. Second, it introduces a new category of risk and opportunity — impersonation and fake agents — that brand-protection work will increasingly have to account for. The reputational integrity of your brand entity now extends to how it is represented to and by verified agents, a theme the brand-safety cluster later in Phase 7 develops in full.

A detail link builders should not miss: the referral payload

As the conversion path becomes agent-mediated, the humble referral link is being re-engineered. In the emerging model, a link surfaced inside an answer increasingly needs to carry a structured payload that ties the AI citation to the merchant’s site, along with consent signals confirming the user agreed to the agent acting on their behalf. Conceptually it looks like this (illustrative — adapt; do not paste verbatim):

This matters for link builders because it signals where the discipline is heading: a link is becoming less a pipe for raw clicks and more a carrier of context and consent. The value of an earned placement will increasingly be judged by whether it feeds the citation-and-attribution graph the agents read, not merely by the referral traffic it sends. For the foundations of how links function as signals rather than mere conduits, our explainer on what backlinks are remains the right grounding.

The Agentic Conversion Ledger: four things you must now own

When the website was the conversion surface, ownership was implicit: the visitor was on your property, so the recommendation, the transaction, the data and the relationship all accrued to you by default. Agentic checkout unbundles those four things and distributes them across surfaces you do not control. The discipline now is to consciously re-claim each one. We call this the Agentic Conversion Ledger — four ownership layers, each of which a brand must now actively keep on its own books rather than assume.

4. Layer one — Recommendation: where link building moves

The first ledger entry is the one closest to a link builder’s existing craft, and it is where the discipline’s value migrates rather than disappears. In the old model, a link did two jobs at once: it passed ranking signal, and it sent referral traffic to a page that converted. Agentic checkout severs the second job from the first. The page may never be the conversion point, so a link’s worth as a traffic pipe to a PDP diminishes. But its worth as a trust and consensus signal that earns the recommendation rises, because the recommendation is now the whole game.

Concretely, the work that moves the Recommendation layer is the work that builds entity authority and consensus: placements in the editorial round-ups agents lean on, genuine presence in the review and community ecosystems they read, and the brand mentions that thicken your footprint in the training data and live indexes. This is earned-media work, and it is more valuable in the agentic era, not less. Our guide to link building strategies maps the tactics; the reframing is simply that you now pursue them to win a named recommendation rather than to drive a click to a checkout you control.

There is a subtlety worth flagging. Research into agentic shopping has begun to characterise AI agents as a “new species of decision-maker” with their own biases and exploitable quirks — they are not perfectly rational optimisers. The ethical line matters here: the goal is to be genuinely, verifiably the better recommendation (through real evidence, real reviews, real authority), not to game a model’s quirks. Manipulating an agent’s biases is the agentic-era equivalent of a link scheme, and it will age just as badly.

5. Layer two — Transaction: removing friction the agent can’t

The second ledger entry is largely an execution problem, covered in depth by the feed-optimisation spoke (#290), but it deserves a place here because a recommendation you cannot transact is a recommendation wasted. Once an agent wants to recommend you, the transaction layer decides whether it can follow through: is your data consistent enough that the agent trusts it, are you present on the relevant protocol, is your availability accurate enough that an agent checkout will not fail?

The recurring failure mode is data inconsistency. If your product page and your feed disagree on price or stock, the agent treats the discrepancy as unreliability and may exclude you from the consideration set entirely — you lose the sale before checkout is even attempted. The remedy is the discipline of a single source of truth rendered identically in feed and page, with availability kept genuinely current. The 2× conversion advantage that structured-catalogue traffic enjoys over scraped data is the clearest evidence that this execution layer pays. Own it not because it is glamorous, but because the other three ledger entries are worthless without it.

There is also a direct line from this layer to the conversion gap of Section 2. Some of that ~77% abandonment is not lost interest; it is friction — a variant the agent could not resolve, an availability mismatch surfaced at the last step, a checkout handoff that stalled. Every data inconsistency you remove and every attribute you complete reduces the chance the agent hits a dead end mid-transaction. In other words, the Transaction layer is not only about being eligible to be recommended; it is about not squandering the recommendations you earn. As the channel matures and the protocols smooth the flow, the merchants who entered with clean, complete, consistent data will convert the rising agentic traffic at materially better rates than those still reconciling feeds after the fact.

6. Layer three — Attribution: measuring a dark funnel

Here the agentic conversion path inflicts its first real wound. When a purchase completes in seconds inside a chat, the identity-capture mechanisms a brand relies on — the landing page, the form, the cookie, the considered journey — largely vanish. The customer barely perceives the agent as an intermediary; they simply feel they decided to buy. You are left knowing that a sale happened without confidently knowing why, or whether your marketing caused it.

What you can measure, and what you cannot

The protocols do provide channel attribution. Webhooks and platform admins report what sold, when, the order value and which agent surface facilitated it; on Shopify this flows into the normal admin automatically. That is real and you should capture all of it, alongside the referral payloads discussed earlier. But channel attribution is not causal attribution. Knowing an order came via ChatGPT does not tell you whether your round-up placement, your reviews or simply the shopper’s pre-existing intent produced it. Industry consensus is blunt about the timeline: plan for 18–24 months before agentic attribution frameworks mature, and expect to operate with partial visibility until then.

Incrementality testing: the honest substitute

Because last-click logic collapses in a dark funnel, the credible answer is incrementality testing — measuring lift against a holdout rather than crediting touchpoints. The practical designs are accessible, and notably Google lowered its incrementality-testing threshold to around $5,000 in late 2025, putting them within reach of smaller brands. This is the bridge between this article and the causal-measurement cluster (AR) later in Phase 7.

The strategic point for link builders: in a channel where you cannot prove a click drove the sale, incrementality is how you defend the budget for earned-media and authority work. Learn the methods now; they are how you will justify recommendation-layer investment when finance asks what the link building actually produced.

The metrics that still work

While causal attribution matures, a handful of measures give you an honest, directional read today. Track recommendation share — how often you appear when you sample category prompts across the engines — as the agentic equivalent of impression share; if you are invisible to agents, conversion rate is moot. Track channel-attributed orders and value from the protocol webhooks, even knowing they understate the picture. Track brand-query volume in traditional search, which tends to rise as a brand becomes more recognisable to the models and is a useful leading indicator that your recommendation work is landing. And track repeat-purchase rate among agent-acquired customers specifically: it is the clearest signal of whether you are keeping the relationship (ledger layer four) or merely renting the sale. None of these is a perfect causal measure, but together they form a dashboard far better than the alternative of flying blind or waiting two years for clean attribution.

7. Layer four — Relationship: owning the post-purchase hub

The fourth ledger entry is the one most likely to be quietly lost, and the most expensive to lose. Agentic checkout threatens the customer relationship in two ways. First, it captures less first-party data than your own checkout, because the data, consent and control sit in the agent’s domain during the transaction. Second, it strips out the moments where brands traditionally deepen a relationship — the upsell, the cross-sell, the loyalty enrolment, the curated thank-you — because the buyer never lands in your funnel to receive them.

The defensible response is to deliberately occupy the post-purchase layer the protocols leave to you. The merchant remains merchant of record across ACP and UCP, which means the fulfilment, the packaging, the returns experience, the support and the follow-up are yours to design — and they are now where the relationship is built rather than at a checkout you no longer control. UCP’s Identity Linking capability is a direct lever here: it lets loyalty and member benefits carry across platforms, so a brand that builds a genuine membership proposition can re-anchor the relationship even when discovery and checkout happened on someone else’s surface.

This connects to the wider 2026 reality that first-party data is becoming the decisive competitive asset as third-party tracking erodes. The brands that win the agentic era are the ones whose data architecture can prove who bought, what was delivered and how the relationship is consented — without that governance becoming a tax that kills the experience. For a link builder or brand strategist, the implication is that authority work and relationship work are now two halves of the same defence: authority earns the recommendation upstream, and a strong post-purchase relationship keeps the customer from being re-acquired by a competitor’s recommendation next time.

8. What this means for link building, specifically

Pulling the ledger together, the agentic conversion path does not make link building obsolete — it relocates its value and adds a new responsibility. Three shifts are worth stating plainly.

• Value moves up-funnel. Links and mentions matter more as recommendation signals and less as referral pipes. The placement that gets you into an agent-trusted round-up is now worth more than the one that merely sends traffic to a PDP, because the PDP may not be where conversion happens.
• The metric of success changes. Referral traffic and last-click conversions understate your impact in a dark funnel. Recommendation share and incrementality become the honest measures of whether earned media is working.
• A new job appears. Owning the post-purchase relationship — first-party data capture, loyalty, retention content — becomes part of the same authority mandate, because winning the recommendation once is wasted if the customer is re-acquired by a competitor next time.

None of this is a reason to abandon fundamentals; if anything it raises their importance. The brand with deep, genuine authority — the kind built through the disciplines in our link building strategies and tracked through the 2026 link building statistics — is precisely the brand an agent reaches for when it must name three options. The conversion path moved; the case for authority only strengthened.

9. Composite case study: a UK apparel brand owns the ledger

The following is an anonymised composite, assembled from patterns across several retailers in 2026; it is illustrative, not a single account.

A UK apparel brand — the kind of mid-market retailer following the path JD Sports cut as the first UK retailer live on Stripe’s agentic commerce suite — enabled agentic checkout expecting a conversion windfall and was initially disappointed: in-chat conversion ran well below their site’s, exactly as the broader data predicts. Rather than retreat, they re-framed the channel through the ledger. On Recommendation, they redirected their digital-PR effort toward the category round-ups agents cited, treating each placement as a recommendation asset rather than a traffic source. On Transaction, they closed the feed-versus-page inconsistencies that had been silently excluding lines. On Attribution, they captured referral payloads server-side and ran a geo holdout to estimate true incremental lift instead of arguing about last-click credit.

The fourth entry proved the most valuable. On Relationship, they invested in the post-purchase experience — a genuinely useful returns and styling follow-up, and a membership programme wired into UCP identity linking so benefits persisted regardless of where the shopper discovered them. Over roughly two quarters, agent-attributed orders grew steadily and, more importantly, repeat-purchase rate among agent-acquired customers rose toward parity with their site-acquired customers — the signal that they were keeping the relationship rather than renting it. The lesson is the ledger itself: the sale is necessary but not sufficient; the business is built in the two layers most brands forget to own.

Two transferable observations came out of the project. First, the team that ran the channel was not the performance-media team but the brand-and-content team — a deliberate choice, because the levers that moved the ledger (authority, entity clarity, post-purchase experience) were brand levers, not bidding levers. Second, the initial disappointment over conversion rate turned out to be the wrong frame entirely: judged on incremental, net-new customers acquired and retained, the channel was clearly additive, even though its headline conversion rate trailed the site. Brands that judge agentic checkout by site-checkout benchmarks will under-invest in it; brands that judge it by incremental demand captured will size it correctly. The metric you choose determines the decision you make.

10. Your Monday-morning action plan

• Calibrate, don’t re-platform. Keep investing in your own converting funnel; treat agentic checkout as additive demand capture while it matures.
• Capture referral payloads server-side. Make sure your stack logs (not strips) agent referral parameters and consent signals, keyed to a citation ID.
• Map your Agentic Conversion Ledger. Score yourself honestly on Recommendation, Transaction, Attribution and Relationship; find the weakest layer first.
• Re-point earned media to the recommendation. Target the round-ups and communities agents cite, judged on recommendation share, not just referral clicks.
• Design one incrementality test. A geo or temporal holdout to estimate true lift — your defence of authority budget in a dark funnel.
• Claim the post-purchase layer. Audit your returns, support and follow-up experience, and wire loyalty into identity linking so the relationship persists.
• Confirm merchant-of-record terms. Know exactly what customer data and relationship rights you retain on each protocol you enable.

11. Frequently asked questions

Does agentic checkout mean my website no longer matters?

No — its role changes. The site is no longer the only conversion surface, but it becomes the post-purchase hub where you fulfil, support, retain and build the relationship. It also remains a primary data source agents consult, so its content and structured data still shape whether you are recommended.

Should I disable my own checkout in favour of in-chat checkout?

Not in 2026. Current data shows in-chat checkout converting well below a brand’s own site — around a third of the rate in Walmart’s figures. Run both: capture the new demand agentic checkout brings while keeping the funnel that still converts better.

Do I still own the customer if they buy through an agent?

Across ACP and UCP the merchant remains merchant of record and the relationship stays with you — but only if you actively claim it. You capture less first-party data at the transaction, so the post-purchase experience and loyalty/identity linking are how you secure the relationship.

How do I prove link building works if I can’t track the click?

Through recommendation share (how often agents name you) and incrementality testing (lift against a holdout). Last-click attribution understates earned media in a dark funnel; incrementality is the honest, defensible measure.

Is this only relevant to large retailers?

No. Platform-native paths (Shopify’s agentic storefronts) lower the technical bar, and the lowered $5,000 incrementality-testing threshold puts measurement within reach of smaller brands. The ledger applies at any size; only the implementation effort scales.