What Is the Right SOW Language for Recommendations That Improve Visibility Without Guaranteeing Inclusion?
The right statement of work language for GEO promises the work, the method and the measurement, never the answer. A good SOW lists specific deliverables and defines how visibility will be measured and reported. It names leading indicators as goals rather than guarantees. It states plainly that AI systems are run by third parties who can change them at any time. It sets out what the client must implement for recommendations to take effect, and it commits the agency to acceptable methods. Google's own guidance says no one can guarantee a number-one ranking on Google, and warns against anyone who claims to. The same logic applies to AI answers, which vary between runs and change with every model update. Adapt the sample clauses below with a lawyer.
Why can't inclusion be guaranteed?
- Third parties control the systems. OpenAI, Google, Anthropic, Microsoft and Perplexity decide how their assistants retrieve, rank and answer, and they change these without notice.
- Answers vary. The same question can produce different answers from run to run, from user to user, and with memory or location.
- Other sources matter. Recommendations depend on reviews, press and competitors, which neither party controls.
- Implementation depends on the client. Recommendations have no effect until pages, profiles and systems are changed.
A contract that ignores these facts sets up a dispute. One that states them sets up an honest working relationship.
What should the SOW contain?
- Objectives, stated as direction: "improve the accuracy and frequency of [client]'s appearance in AI-generated answers to agreed buyer questions".
- Scope and deliverables, listed specifically.
- Measurement method: the question set, engines, run frequency, and how results are recorded.
- Baseline and reporting: when the baseline is taken and how often reports are delivered.
- Indicators: what will be tracked, described as indicators, not guarantees.
- No-guarantee clause.
- Client responsibilities: approvals, access and implementation time.
- Methods and compliance: tactics the agency will and will not use.
- Data handling: prompts, logs and any personal data.
- Review points and conditions for changing, pausing or ending the work.
What do sample clauses look like?
Objective
The Agency will provide services intended to improve the accuracy and frequency with which the Client's business, products and content appear in answers generated by AI search and assistant platforms for the questions agreed in Schedule A.
Deliverables
During each month of the Term, the Agency will deliver: (a) a visibility report covering the questions in Schedule A across the platforms listed in Schedule B; (b) up to [number] content briefs or page drafts; (c) a technical access review of the Client's website for the AI crawlers listed in Schedule B; and (d) a prioritised list of recommended actions.
Measurement
Visibility will be measured by running each question in Schedule A [number] times per platform per reporting period, in sessions without personalisation where the platform allows, and recording whether the Client is mentioned, how it is described, and which sources are cited. Results will be reported as rates across runs, with the number of runs stated.
No guarantee
The Client acknowledges that AI search and assistant platforms are operated by third parties who determine, and may change at any time and without notice, how answers are generated, which sources are used and which businesses are mentioned. Answers may vary between users, sessions and runs. Accordingly, the Agency does not guarantee that the Client will appear in, be recommended by, or be described in any particular way in any answer, or any specific level of visibility, traffic or revenue. Indicators described in this SOW are goals used to guide and evaluate the work, not promised outcomes.
Client responsibilities
The Agency's recommendations will affect visibility only if implemented. The Client will provide [access], review and approve deliverables within [number] business days, and implement agreed technical and content changes within a reasonable time. Where recommendations are not implemented, reported results should be read in that light.
Methods
The Agency will not use hidden text, cloaking, undisclosed paid endorsements, fake or incentivised reviews, instructions intended to manipulate AI systems, or any method that breaches the published policies of the platforms concerned or applicable law. The Agency will disclose any paid placements recommended to the Client.
Data
The Agency will not include personal data in prompts used for measurement, and will store measurement data [location and retention].
Review
The parties will review results and scope at the end of month [three], and at least every [three] months thereafter, and may agree to change, reduce or pause the services.
How should the pitch and proposal match the SOW?
Disputes often start with the sales conversation, not the contract. If a pitch deck promises "get recommended by ChatGPT" and the SOW disclaims any guarantee, the client remembers the pitch. Use the same language in both. Describe what you will do and how you will measure it. Show example reports from real work with results presented honestly, including where nothing changed. Explain up front that answers vary and that some factors sit outside anyone's control. Clients who hear this before signing are rarely surprised later, and the ones who want guarantees are better turned away early.
What should the SOW avoid?
- "Guaranteed placement in ChatGPT" or "top three in AI Overviews". These cannot be delivered, and making such claims may breach consumer protection or advertising rules.
- Payment tied to specific appearances unless both parties understand and accept the volatility.
- Undefined metrics, such as "AI visibility score" without a method.
- Open-ended "optimisation" without deliverables.
- Silence on client implementation, which leaves the agency responsible for changes it cannot make.
How can results still be accountable?
No guarantee does not mean no accountability. Agree on:
- Leading indicators such as crawler access fixed, facts corrected, pages published and sources earned.
- Visibility trends across a fixed question set, with enough runs to separate change from noise.
- Accuracy targets, such as the share of answers describing the client correctly.
- Business outcomes such as AI-referred enquiries, tracked but not guaranteed.
What evidence to bring to renewals is covered in what evidence a GEO agency should bring to a renewal meeting. Whether asking assistants directly is a valid measurement method is covered in whether asking LLMs directly is a valid way to measure visibility.
What might an illustrative case look like?
Consider an illustrative agency whose first GEO contracts promised "top-three placement in ChatGPT for your main keywords". Within months, two clients disputed invoices after answers changed following a model update. The agency rewrote its SOW. Objectives were stated as direction, deliverables were listed monthly, and measurement defined a fixed question set with ten runs per platform. It added the no-guarantee and client-implementation clauses above. Renewal conversations moved from arguments about screenshots to reviews of trends and completed work. Its client retention improved over the following year.
What do people ask about GEO contracts?
Can we offer a performance bonus?
Some agencies do, tied to agreed indicators with a clear method. Make sure both sides understand the volatility.
Should the SOW name specific AI platforms?
Yes, in a schedule, with the right to update the list as platforms change.
Is a no-guarantee clause enough legal protection?
It helps, but marketing claims must match it. Do not promise in a pitch what the contract disclaims.
Who owns the question set and reports?
State it. Many agencies give clients ownership so the work can continue if the relationship ends.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources