What you can promise (and what you can't) about AI visibility without burning your fingers
The client, pen in hand, before signing: "So, how long until I show up in ChatGPT?"
What you answer now determines what the relationship looks like six months from now. If you promise what you don't control, you've signed your own complaint. If you just say "there's no way to know", you look like a smoke seller in reverse: charging for something you won't even stand behind. There's a third way, and it's the one that separates the agencies that renew these contracts from the ones that burn them: promise exactly what you control, and not a millimetre more.
Let's draw that line precisely, and give you the clauses to put it in writing.
The principle: you control the work, not the answer
This tension already existed in classic SEO — nobody serious guarantees position 1 — but in AI visibility it's sharper, because the models' answers are volatile: the same question can give different results today and tomorrow, and the models change without warning and without a changelog.
From that comes the rule that orders everything else:
Key data
You commit to the work and the coverage. Never to the model's decision.
You decide which sources to fix, which content to publish, how many prompts to monitor and how often. ChatGPT decides who it recommends. The first goes in the contract as an obligation; the second goes in the report as an observed result.
What you CAN commit to
Measurement coverage. Number of prompts monitored, across how many AIs, how often. "Weekly tracking of 40 questions in ChatGPT, Gemini and Perplexity" is a verifiable commitment you deliver, come rain or shine. Choosing those questions well takes some technique: here we explain how.
Correction and creation work. Source audit completed in the first month. X corrections applied across website, listings and directories. X pieces of citable content published per month. Structured data implemented. All measurable, all under your control.
Reporting deliverables. Monthly report with trend, dated screenshots and comparison with the main competitor. Monthly meeting. Detection and alert if the AI gives false data about the client.
Response time on errors. "If we detect an AI giving incorrect information about your business, we start the source correction within 5 working days." Notice: you commit to starting the correction, not to the AI rectifying it by a given date.
What you CAN'T commit to (and how to say it without losing the sale)
Appearing in a specific AI on a specific date. Nobody controls what a model cites. Whoever guarantees it is either lying or attaching conditions so rigged that the guarantee is worthless.
Positions or "being the first recommended". In AI there isn't even a stable ranking to guarantee.
That a correction will be reflected within a deadline. You fix the source today; the model picks it up when it picks it up. Weeks, sometimes more.
Volume of traffic or clients from AI. You can cite sector data as context — traffic from AI answers converts at 14.2% versus 2.8% for classic organic, according to figures compiled by roymo.es — but as context, never as a commitment.
And how do you say all this without the client cooling off? With the opportunity figure in front of them: in our study of 9,865 Spanish SMEs, 91% only appear in 1 of the 4 main AIs (full study). The script:
Key data
"I'm not going to guarantee that ChatGPT will recommend you in 90 days, because nobody controls that and whoever promises it is lying to you. What I do guarantee is everything that depends on us: weekly measurement, correction of every source the AIs draw from, and content designed to get you cited. And the reality of the market is that 91% of your competition is doing none of this."
Radical honesty isn't just ethical: it's a competitive advantage. When another agency shows up promising ChatGPT gold, your client will already know what questions to ask them — partly because you'll have taught them. If that situation catches you mid-renewal, we have a whole article on how to defend it.
Sample clauses for your proposal
Adapt them to your template; the structure is what matters.
Scope of service clause:
Key data
"The service comprises: (a) weekly monitoring of [N] representative queries across [ChatGPT, Gemini, Perplexity and Claude]; (b) ongoing audit and correction of the information sources feeding those systems (website, business listings, directories, structured data); (c) creation of [N] pieces of content optimised for citation per month; (d) monthly trend report with dated evidence."
Nature of results clause:
Key data
"The CLIENT understands and accepts that the responses of generative AI systems are produced by third parties (OpenAI, Google, Perplexity, Anthropic) over which the AGENCY has no control, and that they may vary between queries and over time. The AGENCY commits to carrying out the described work and to the transparent measurement of its effect, without guaranteeing mentions, positions or specific results in those systems."
Error detection clause:
Key data
"If monitoring detects incorrect information about the CLIENT in any of the systems, the AGENCY will report it in the next report (or within 48h if the error is serious) and will start the correction of the identified sources within a maximum of 5 working days."
Objectives review clause:
Key data
"At 6 months, the parties will review the evolution of mention coverage relative to the first month's baseline, and may adjust the set of monitored queries, the systems covered and the work plan."
That last one gives you a natural renegotiation point and communicates confidence without promising figures. Which specific KPIs to put in each category — committed vs only reported — we break down in detail in which KPIs to put in a GEO proposal.
For these clauses not to be a dead letter you need the measurement machinery behind them: if you commit to weekly tracking across 4 AIs with dated evidence, someone has to run it every week. With Surfeo for agencies that commitment is met on its own: each client with their prompts monitored weekly across the 4 AIs, full history and PDF reports, from €20/month for the agency account plus €35-79 per client depending on plan.
Frequently asked questions
What if the client demands a results guarantee to sign?
Offer a guarantee on what you control: "If the monthly report shows the committed work hasn't been done, that month isn't invoiced." It's a real, powerful guarantee that you can keep. If they still demand a guarantee of mentions in ChatGPT, that's a client who'll come after you every month over things you don't control — sometimes the best signature is the one you don't get.
Can I show case studies without them counting as a promise?
Yes, with two precautions: that they're real and verifiable (dated screenshots, measured trend) and that they come with the tagline "results of a specific case; every sector and every starting point is different". A real case with that honesty sells more than a hollow guarantee.
What realistic timeframe can I give verbally without committing to it?
A range with honest vagueness: "in sectors like yours we usually see the first movements between the second and fourth month, but there are faster and slower cases". And in writing, only the work plan: audit in month 1, corrections in months 1-2, content from month 1, evaluation at month 6.
How do I handle the client if visibility hasn't improved at 6 months?
With the data on the table: work delivered (committed and met), the competitor's trend (if the whole sector is stalled, the context backs you up) and a revised plan. It's exactly the situation the review clause exists for. We have a specific article for that scenario.
Before writing your next proposal, measure the client's real starting point: take the free AI visibility test and build the promise on data, not on hope.
