Zero-click took the visit. Zero-read takes the answer. The action-taking agent takes the whole transaction — finding, comparing, deciding and buying while the user watches a summary. For a UK brand this is a new contest with new rules, new rails and, uniquely, a regulator that has already written the law for it.

The post-click conversation has been running for two years, and it has been one loss at a time. First the agent answered the query in place, and the visit disappeared — zero-click. Then the answer itself absorbed the reading, and even the summary went unread as users skimmed a single line — call it zero-read. The action-taking agent completes the sequence by absorbing the transaction. It does not hand the user a list of vendors to evaluate; it evaluates them, chooses one, and completes the purchase or the booking on the user’s behalf. The user sees a confirmation, not a shopfront.

This is a categorical change, not an incremental one. For most of the web’s history the entire apparatus of marketing existed to move a human through a funnel: impression, click, consideration, conversion. An action-taking agent collapses that funnel into a single machine decision made on the user’s behalf. Earlier work in this area examined how commerce-native AI decides what to surface and how Google’s AI Mode is reshaping the commercial query. This article takes the next step: not what the agent recommends, but what it does — and how a brand earns its way into actions it will never directly witness.

The strategic gap is wide open. Almost everything written for marketers still assumes a human at the end of the chain, persuadable by design, copy and price presentation. When the buyer is an agent, persuasion in the human sense stops mattering and a different set of properties takes over — properties most brands have never deliberately optimised for, because until recently nothing read their site the way an agent now does. What follows is a framework for that contest, the plumbing it runs on, an honest account of how settled it actually is, and the UK legal overlay that, almost uniquely, has already been drafted around it.

1. What “Without a Visit” Actually Removes

Begin by being precise about what is lost, because the instinct is to mourn the click and stop there. The action-taking agent removes three things at once, and each had been doing quiet work for you.

• It removes the visit. No session, no on-site journey, no chance to convert with your own interface. The analytics that most teams still treat as ground truth simply stop recording the most valuable events.
• It removes the read. The user does not read your page; the agent does, and it extracts what it needs in a form built for machines. Your carefully sequenced argument is irrelevant if the agent only parses your price, stock and terms.
• It removes the impression. In the hardest case the user never even sees your name — the agent compares options silently and surfaces only the one it chose. You can win the transaction and gain no brand awareness at all, or lose it and never know you were considered.

What replaces them is a single question asked by software: of all the options I can read and act upon, which one do I complete this task with? That question is answered by properties, not persuasion. The value of a raw click was already eroding as internal journeys shortened; the action-taking agent finishes the job by making the click optional to the outcome entirely. Measuring success in sessions, in this world, measures a shadow of the real event.

A concrete, anonymised version makes the stakes legible. Picture a UK household replacing a broken appliance. The user tells their assistant the model, a budget and a delivery window, and steps away. The agent reads a dozen retailers, discards the ones whose stock data it cannot trust, compares price-including-VAT and delivery against the stated window, and completes the order with the one retailer that was readable, executable, competitively priced and reliable on its terms. Three of the discarded retailers were cheaper on the headline figure but exposed the agent to risk — one priced before VAT, one had stock the agent could not verify, one buried its delivery terms in a script the agent could not parse. None of them lost on price. They lost on the four gates, silently, and they will never see the lost sale in any report they currently run. The winning retailer never earned a visit, a read or even a remembered impression — only the transaction. That asymmetry is the whole game.

2. The Agent-Selection Stack: Four Gates to Being Acted Upon

Here is the instrument the rest of the article supports. Before an agent completes a task with your brand, it passes through four gates in sequence. They are strictly ordered: an agent that cannot read you never reaches the question of whether to transact with you; one that cannot transact never reaches the question of preference; one it does not prefer never reaches the question of trust. Each gate is necessary, and failing the earliest one you fail is where you lose — which is why diagnosis matters more than any single fix.

Gate 1 — Retrievable

An agent can only act on what it can read. A page rendered entirely in client-side JavaScript, a catalogue with no machine-readable feed, a site that blocks the agent’s crawler, or product data too thin to parse — any of these removes you from the candidate set before selection even begins. Retrievability is the agentic descendant of crawlability, and it is unglamorous, structural and decisive. If the agent cannot assemble a clean, current representation of what you sell and on what terms, nothing downstream matters.

Gate 2 — Eligible

Being readable is not the same as being actionable. To be acted upon, an agent must be able to execute against you: complete a checkout through a supported rail, or hand off cleanly to a transaction it can rely on. A brand whose price, availability and terms are present but not executable — no connected checkout the agent recognises, stock data it cannot trust, terms it cannot parse — is eligible to be recommended but not to be transacted with. As agentic commerce rails mature, eligibility is increasingly a question of which protocols and payment systems you have connected to, not merely whether your data is tidy.

Gate 3 — Selectable

Among the brands an agent can read and act upon, it still chooses one. Selection is decided by the corroborated, earned signals this publication has always tracked: reviews, independent mentions, competitive price and availability, and whether you are the primary or authoritative seller of the thing in question. This is earned authority doing its work in a new venue. An agent ranking three sources of the same product weighs availability, price, quality and primacy — and the brand that has built genuine corroboration is the one it reaches for. The competitive analysis you would run for human SERPs now has an agentic equivalent: which rival does the agent pick, and why?

Gate 4 — Trustable

The final gate is whether the agent — and, in the UK, the law standing behind it — can safely complete the action with you. An agent that cannot rely on your stated price, your stock, your cancellation and refund terms is a liability to the user it serves, and a well-built agent will route around you to avoid it. Trustability is where accuracy, consistency and compliance stop being hygiene and become a selection signal. It is also, as the rest of this article argues, where UK law has moved fastest — turning trustworthiness from a courtesy into an enforceable requirement, and therefore into a moat for the brands that take it seriously. Brands with a history of policy breaches or penalties carry that risk into the agentic layer too.

3. The Plumbing: Rails, Protocols and an Honest Caveat

Gate 2 — eligibility — is the one that depends on infrastructure outside your own site, so it is worth understanding the rails that emerged in 2025 and 2026. They are real, they are converging, and they are also less settled than the breathless coverage suggests.

The first open standard arrived in September 2025, when OpenAI and Stripe launched Instant Checkout and the Agentic Commerce Protocol, an open specification that lets an agent and a merchant complete a purchase together. The merchant keeps control of payments, fulfilment and the customer relationship; the agent passes scoped information and a payment token. Stripe described it as a shared language between businesses and AI agents, and reporting put the merchant fee at around four percent of each completed purchase on top of standard processing. In January 2026 Google answered with its own coalition-backed Universal Commerce Protocol, aimed at AI Mode and Gemini and endorsed by a long list of networks and processors. Most brands of any scale will end up supporting more than one.

Underneath the protocols sit the payment rails, and here the convergence is striking. On a single day in June 2026, Visa announced an agent-payment integration with OpenAI and Mastercard launched Agent Pay for Machines — two of the largest networks moving on agent-initiated payment within hours of each other. The shared mechanism is tokenisation: rather than handing an agent a raw card number, the system issues a token scoped to a specific agent, merchant and amount, with consumer-set spending limits enforced before any payment is authorised. The direction is unambiguous. Software is being given the means to pay, safely and at scale.

Britain is inside this shift, not watching from the sidelines. In March 2026 a major bank, working with Mastercard and a payments partner, completed what was reported as Europe’s first live end-to-end payment executed by an AI agent on real banking infrastructure — explicitly framed as a controlled test rather than a commercial rollout, but a clear signal that agent-initiated payment on UK-relevant rails is moving from slide deck to settlement. Set against Britain’s unusually mature real-time and account-to-account payment infrastructure, the practical implication is that UK consumers may meet live agentic checkout sooner than the US-first headlines imply. A brand treating agent eligibility as a 2027 problem could find the surface arriving in 2026.

The same agentic logic is arriving across every major surface, not just ChatGPT: Microsoft’s Copilot has its own checkout path, Apple’s on-device assistant routes some intent to partner models, Meta’s assistants push commerce closer to the feed, and open-model ecosystems lower the cost of building agents that act. A brand planning for one agent is planning for too few.

4. Winning Each Gate

The framework names the gates; this section is how you clear them. Reassuringly, most of the work is the discipline this publication already teaches, redirected at a machine reader rather than a human one.

Make yourself retrievable (Gate 1)

Start with the unglamorous audit: can an agent actually read you? Ensure your important commercial data is rendered server-side or available through a structured feed, not locked behind client-side scripts. Keep a clean, current machine-readable representation of price, availability and key terms. Confirm you are not inadvertently blocking the crawlers these agents use to retrieve live data — a well-intentioned bot block on the infrastructure side is one of the most common ways a brand quietly removes itself from the candidate set. Retrievability is mostly a technical hygiene project, which is exactly why it is so often neglected and so cheap to fix.

There is a subtle internal-architecture dimension here too. The same clarity that helps an agent parse you — clean templates, descriptive structure, important pages reachable without deep navigation — is what classical crawl efficiency has always rewarded. The work overlaps almost entirely with sound internal linking and information architecture; an agent simply punishes the lapses more visibly, because where a search crawler might eventually find a buried page, an agent assembling a shortlist in real time will not wait for it. Treat retrievability as crawlability with a stopwatch.

Become eligible to transact (Gate 2)

Eligibility is an infrastructure decision. Decide which agentic commerce protocols and payment rails are worth supporting for your category, and structure your product data — titles, descriptions, prices with correct currency, availability, eligibility flags — to the standard those rails expect. Treat your feed as a first-class product surface, because to an agent it is your storefront. For UK merchants this also means getting the commercial fundamentals an agent checks — price in the right currency, inclusive of tax, with accurate stock — unambiguously correct, since an agent that detects a mismatch between your stated and actual terms will discount you on reliability grounds alone. Reported integration costs, such as the roughly four percent merchant fee on completed agent purchases, should be modelled against the incremental demand the channel actually brings, not adopted reflexively.

A useful discipline here is to treat eligibility as a portfolio decision rather than an all-or-nothing one. You do not have to be transactable everywhere on day one; you have to be transactable where your highest-intent agentic demand is forming. For most UK brands that means prioritising the surfaces their customers already use to research and the rails the major networks are standardising, then expanding coverage as the channel proves out. The failure mode to avoid is the opposite of over-investment: being readable and well-reviewed, clearing Gates 1 and 3 handsomely, and then losing the transaction at Gate 2 because no agent could actually execute against you. A brand that the agent wants to choose but cannot transact with has done the hard work and forfeited the reward at the last step.

Earn selection (Gate 3)

Selection is won the way visibility has always been won: corroboration the system trusts. Reviews, independent mentions, and being named as the authoritative source for a product or category are what tip an agent toward you when it ranks comparable options. The classic link building strategies still do the foundational work, because the brands an agent finds credible are the ones the wider web already vouches for. Earned editorial coverage — the kind a responsive-source approach like HARO produces — builds exactly the third-party corroboration an agent reads as a trust signal. And the 2026 data picture continues to show that the authority signals driving conventional visibility are the same ones feeding which sources AI systems reach for.

One practical nuance separates agentic selection from human ranking: agents weigh primacy and verifiability more heavily than humans do. A human shopper may pick a familiar brand on instinct; an agent, tasked with protecting its user, leans toward the source it can most readily verify as the genuine, primary seller at the stated terms. That rewards brands that are unambiguously the authoritative origin for what they sell — official product pages, consistent naming, corroboration from sources the agent already trusts — and penalises resellers and lookalikes whose claims the agent cannot cleanly confirm. Being the obvious, verifiable primary source is, increasingly, a ranking factor in its own right.

Prove yourself trustable (Gate 4)

Trustability is earned by being boringly reliable: accurate prices that match at checkout, real stock, clear and consistent refund and cancellation terms, and no manipulative design. An agent built to protect its user will avoid brands whose stated terms cannot be relied upon, and a brand carrying a history of policy violations or exposed to manipulation risks starts at a disadvantage. In the UK this gate is not merely commercial prudence — it is a legal perimeter, and the next section sets out why that turns trustworthiness into the most defensible advantage available.

5. The UK Overlay: Where the Law Got There First

Most markets are improvising their response to action-taking agents. The United Kingdom, unusually, has already written much of the rulebook, and any brand selling to UK consumers through agents needs to understand it — not as a compliance afterthought, but as a source of genuine competitive advantage.

The foundation is the Digital Markets, Competition and Consumers Act 2024, whose consumer-enforcement regime gave the Competition and Markets Authority direct powers — including the ability to impose fines of up to ten percent of a company’s global turnover for breaches such as misleading practices. In its first year under the new regime the regulator reported real enforcement activity and consumer refunds, which is to say these are not theoretical powers. Then, in March 2026, the CMA published guidance and a policy paper dealing specifically with agentic AI and consumers. The starting principle is deliberately simple: the same consumer law applies whether the customer deals with a human or an AI agent, and a business is responsible for what its agent does much as it is responsible for its employees.

That principle has sharp consequences for the four gates, and especially for trustability. Before the specifics, one structural point worth holding: under the Act’s digital-markets limb the CMA can designate the largest platforms with “Strategic Market Status” and impose tailored conduct rules on them, and a cross-regulator forum spanning competition, data, communications and financial authorities has already examined agentic AI together. The agent platforms your brand will depend on are themselves moving inside a regulatory perimeter — which makes their incentives to favour compliant, reliable merchants stronger, not weaker. Three areas stand out for any brand operating in the agentic layer in Britain.

1. Accuracy at scale is now a legal exposure. If an agent acting in your commercial chain misstates a price, a characteristic, or a refund right — and that causes a consumer to take a decision they would not otherwise have taken — it can constitute a misleading practice. Errors that would be trivial for one human transaction become systemic when an agent repeats them across thousands, which the regulator has flagged as a particular concern.
2. Dark patterns do not get an agentic exemption. False urgency, fake scarcity and manipulative choice architecture are squarely in scope under the Act’s treatment of online choice architecture — and the regulator has signalled it will look hard at agent interfaces that pressure or mislead, including any failure to disclose where a seller has paid to be promoted to the agent.
3. Distance-selling rules still bind the seller. The Consumer Contracts Regulations 2013 require that the main characteristics and total price are given clearly and directly before an order, with an explicit acknowledgement that ordering creates an obligation to pay. Those duties fall on the trader entering the contract, so a UK merchant transacting through an agent must ensure the agent-mediated flow still satisfies them.

Read defensively, this is a list of ways to be fined. Read strategically, it is a moat. A brand whose prices, stock and terms are accurate and consistent enough to satisfy the CMA is, by exactly the same properties, the brand an agent finds most trustable and most readily completes a transaction with. UK consumer law and agentic selection reward the same behaviour. The brands that treat the CMA’s agentic guidance as a design brief rather than a threat will clear Gate 4 by construction, while competitors that cut corners expose themselves to both regulatory penalty and silent de-selection by the agents themselves. In Britain, being trustworthy to a machine and being lawful to a regulator have become the same project.

6. Measuring Selection You Cannot See

Action-taking agents are the hardest surface yet to measure, because the defining event — the agent choosing and completing — happens where you have no visibility. Honesty matters here: nobody can hand you a clean dashboard of agentic selection, and anyone claiming a precise share-of-agent number is overreaching. What you can do is triangulate from proxies, read as trends.

• Feed and rail coverage. The most controllable input: what share of your catalogue is actually exposed, current and executable across the protocols that matter for your category? Incomplete or stale feeds are the commonest reason a brand is silently ineligible, and this is measurable on your own side without any agent cooperation.
• Seeded selection testing. Within agents you can operate, run realistic buyer tasks for your products and your two closest rivals, and record who gets selected and acted upon, at what price point, with what stated reason. Run it repeatedly across many seeds; a single run is noise, the distribution across many is signal.
• Agent-referral and post-agent demand. Segment the traffic and conversions that arrive via chat domains, agent hand-offs and assistant referrals, accepting that attribution is imperfect. Watch branded-search lift too: when an agent completes a task with you, some users notice the name and return directly later.
• Reliability defect rate. Track the gap between your stated and actual terms — price mismatches at checkout, stock errors, policy inconsistencies. This is your Gate 4 health metric, and it is simultaneously your CMA-exposure metric. One number, two reasons to drive it to zero.

Report these as a panel, not a score, and resist the urge to over-interpret any single test — agentic systems are non-deterministic, so the same task will not always resolve the same way. The instrumentation overlaps almost entirely with what your existing measurement habits already cover; what changes is the question you point them at — from “did a human convert?” to “was my brand the one the agent acted upon, and could it rely on me when it did?”

7. What Action-Taking Agents Do Not Change

A dramatic shift invites overcorrection, so it is worth stating plainly what has not moved. Several things are exactly as they were, and a brand that abandons them in a rush toward agent-readiness will weaken the very signals agents rely on.

Earned authority still seeds the candidate set. An agent does not invent its shortlist; it assembles it from the corroborated, well-linked sources the open web already trusts. The link graph that underwrites conventional visibility is the same graph that determines whether you are a candidate for selection at all. Agents route through authority; they do not route around it.

The fundamentals of accuracy and trust are not new requirements either — they are old ones the agentic layer simply makes unforgiving. A brand that has always kept its prices honest, its stock accurate and its terms consistent arrives at Gate 4 already qualified. And the non-determinism that has always made AI measurement treacherous has not been tamed; a favourable test result remains one data point, never proof. The vocabulary of gates and rails is new. The disciplines underneath are the ones that have separated durable brands from fragile ones all along.

8. Your Monday-Morning Deliverable: An Agent-Readiness Audit

Translate the framework into something executable this week. The audit below walks the four gates in order, takes a day or two, and ends with a ranked list of where you are losing agentic selection — with a deliberate UK compliance pass folded into Gate 4.

• Test retrievability (Gate 1). Pick five priority products. For each, confirm an agent could read your price, stock and key terms without executing client-side scripts, and check your bot rules are not blocking the agents’ retrieval crawlers. Every product that fails is invisible before selection even starts.
• Test eligibility (Gate 2). List the agentic commerce protocols and payment rails relevant to your category and mark which you actually support today. Audit your feed for completeness and freshness — correct currency, tax-inclusive price, accurate availability. Stale or partial feeds are your most fixable loss.
• Test selectability (Gate 3). Run seeded buyer tasks against agents you can operate, for your products and two rivals. Record who is selected and the stated reason. Where a rival wins, note whether it was price, availability, reviews or primacy — that is your earned-authority target list.
• Test trustability and UK compliance (Gate 4). Reconcile stated versus actual terms across those five products — price at checkout, real stock, refund and cancellation clarity. Then run the CMA pass: would an agent acting on your data ever mislead a consumer, apply false urgency, or obscure the obligation to pay? Each defect is both a selection risk and a DMCC exposure.
• Rank and assign. Order the four gates by where you fail earliest. Fund that gate first, because clearing an earlier gate is worthless if a later one still de-selects you — and clearing a later one is impossible if an earlier one keeps you out of the running.

Most brands discover their earliest failure is Gate 1 or Gate 4 — they are quietly unreadable, or their stated terms cannot be relied upon — and both are addressable without exotic tooling, using the disciplines this publication already documents. The shift required is not new tactics but a new target: stop optimising only to persuade a human who visits, and start optimising to be the option an agent can read, transact with, prefer and safely complete.

Where This Leaves You

The action-taking agent finishes what zero-click began: it removes the visit, the read and sometimes the impression, and replaces persuasion with a machine decision made on the user’s behalf. Winning in that world is a matter of clearing four gates in order — retrievable, eligible, selectable, trustable — on rails that are real but not yet settled, against the durable likelihood that agent-led discovery feeding a merchant’s own checkout, rather than full in-chat purchase, is the shape that lasts. For UK brands there is a rare gift hidden in the regulation: the same accuracy, consistency and honesty the CMA now enforces is precisely what makes an agent willing to act on you. Build to that standard while the surface is new and competitors are still arguing about whether the click is dead, and you become the option Britain’s agents reach for — long after the query that introduced you has been completed, paid for, and forgotten.