The Deliverables That No Longer Deliver: Contract Architecture for AI-Mediated Search

Contract renewal season has arrived for a great many UK search retainers, and the paperwork being circulated looks much as it did three or four years ago. That is the problem. The commercial instrument has not caught up with the channel it is meant to govern, and both sides are now exposed to disputes that neither anticipated when the schedule was first signed.

The difficulty is structural rather than personal. Search engine results pages have become answer surfaces. A meaningful share of queries now resolve inside a generated response, and the referable click that a contract might once have assumed is no longer a reliable unit of delivery. Yet most statements of work still measure success in rankings, article counts and links secured. Those were always proxies; in 2026 they are proxies of shrinking fidelity.

What follows is a practical reading of the clauses that most often break, written for procurement teams, in-house marketing leads and agency principals alike.

Why the deliverable schedule is the weak point

A deliverable is a promise of activity. An outcome is a promise of result. Agencies can generally warrant the first and should almost never warrant the second — and yet schedules are routinely drafted in a way that blurs the two, because both parties find the ambiguity comfortable during the pitch.

A clause that reads "twelve optimised pages per quarter, targeting priority commercial keywords" sounds precise. It is not. It does not define optimised, it does not establish whether a page counts once it is published or once it is indexed, and it does not address what happens when the target keyword is a question the engine now answers directly.

Fairer drafting separates the categories explicitly:

• Contracted activities — audited, scheduled and evidenced by dated artefacts in a shared repository.
• Service standards — response times, review cadence, hours of analyst time, escalation routes.
• Anticipated outcomes — expressed as directional targets, clearly marked as forecast rather than obligation.

A useful discipline is to require that every line in the schedule can be evidenced by a file. If the client cannot point to the deliverable, the agency cannot prove it, and the renewal conversation begins from a position of mutual distrust.

Measurement windows and the lag problem

Most retainers assume a twelve-month horizon with a review at month six. That rhythm was calibrated to a channel with a reasonably predictable feedback loop. It is now too slow for change detection and too fast for compounding results.

Consider splitting the window in two. A quarterly performance window governs operational reporting: crawl health, indexation, query mix, share of voice across surfaces, and the proportion of impressions arriving on answer features rather than classic listings. A separate commercial window — eighteen months is not unreasonable for a competitive UK category — governs the assessment of whether the strategy is working.

Risky wording: "Agency guarantees a 30% increase in organic sessions within nine months."

Fairer wording: "Agency will report monthly against the agreed measurement framework. Where organic sessions decline for three consecutive reporting periods, the parties will convene a strategy review within fourteen days, with the agency required to present a remediation plan."

The second formulation gives the client a genuine remedy — a structured conversation with the supplier on the back foot — without converting a volatile channel into a fixed obligation.

Change-control triggers worth naming

The most contested phrase in any search contract is "material change in the search landscape". It is invoked by agencies seeking relief and disputed by clients who suspect it is used to excuse underperformance. The answer is to define it before it is needed.

Workable triggers include: a documented change to how a major engine renders or cites content; a platform policy shift affecting an agreed tactic; a regulatory or consent-driven change to tracking coverage; a client-side replatform, migration or rebrand; and a sustained shift in the ratio of clicks to impressions beyond an agreed tolerance.

Each trigger should specify the consequence — typically a re-scoping meeting within a defined number of working days, and a documented variation rather than an informal understanding.

The client's side of the data bargain

Agencies are frequently held to performance standards while being denied the access required to meet them. Contracts should therefore state the client's obligations with the same precision as the agency's. That means named access to analytics and search console properties, consent for log-file or server-side data where proportionate, nominated approvers with defined turnaround times, and a commitment not to deploy unannounced changes to templates, robots directives or redirect rules.

Where the client's own consent configuration limits measurement coverage, the contract should say so plainly, and should confirm that the agency's reporting is calibrated to that constraint. This is not a technicality. It is the difference between a defensible report and a contested one.

AI content: warranties, indemnities and disclosure

Two questions recur. First, who owns what the agency produces? Second, who bears the risk if the output is inaccurate, duplicated or reputational?

A balanced position is that the client owns final deliverables on payment, while the agency warrants that it has the right to assign them, that it has followed a documented editorial and fact-checking process, and that human review was applied before publication. An indemnity limited to third-party intellectual property claims is normal. An indemnity covering all downstream consequences of AI-assisted drafting is not, and agencies that accept one are pricing a risk they cannot quantify.

For their part, clients should require disclosure of where generative tools were used in the workflow and what controls applied. A short schedule listing tooling, review steps and the person accountable for sign-off is more useful than a sweeping prohibition that nobody can enforce.

Exit, handover and portability

Disputes cluster at the end of a relationship. The contract should therefore specify a handover pack delivered within a fixed window: account ownership transfers, historical reporting data in an open format, content inventories, redirect maps, technical documentation and any tooling licences the client has funded.

Risky wording: "Agency will provide reasonable assistance during offboarding."

Fairer wording: "Within fifteen working days of termination, Agency will deliver the handover pack specified in Schedule 4, including full export of reporting history and administrative transfer of all client-owned properties."

The second version protects both parties: it protects the client from hostage-taking, and it protects the agency from an open-ended obligation that outlives the fee.

A practical renewal checklist

Before signature, confirm that the schedule distinguishes activity from outcome; that measurement windows are split between operational and commercial horizons; that change-control triggers are named and consequential; that client data and approval obligations are written down; that AI-content warranties are proportionate and disclosure is specified; and that exit handover is defined as an artefact rather than an attitude.

None of this requires adversarial drafting. It requires only that the document describe the channel as it now operates, rather than as it operated when the template was written. Contracts are predictions, and predictions about search have become materially harder. The wording should say so.