GEO Contracts and KPI Acceptance: Defining What 'On Target' Actually Means

Disputes over GEO services rarely start with price. They start with a vague definition of success.

The direct answer: the most important part of a GEO contract is not the price, it is the KPI definition. If the contract only promises "a monthly report" and "continuous optimisation of visibility", the two sides can hold completely different expectations and there will be no way to settle the results six months later. This guide sets out precise definitions for five metrics, sample contract clauses, the acceptance process, and the traps to avoid before signing.

For pricing models and what a quotation should itemise, see the Hong Kong GEO service pricing guide. This article focuses on what happens after signing.

Why the KPI definition matters so much

AI visibility is not like an SEO ranking: there is no single public, standard, always-checkable position. The same brand can produce very different results depending on which questions are used, which platform is tested and which day it is measured. That makes an undefined number — such as "improve visibility by 30%" — impossible to verify:

• Which questions were tested?
• How many were tested?
• Was it the Gemini app or Google AI Overview?
• On which date?
• Does "visibility" mean mentioned, or cited?

Without a written definition, there is no KPI.

A verifiable KPI needs five elements

Element	What must be stated	Example
1. Metric definition	How the number is calculated	Coverage = (questions mentioning you ÷ total questions tested) × 100%
2. Sample size (n)	How many questions are tested, and whether the question list is attached	5 target questions, listed in Annex A
3. Date and frequency	When it is measured and how often	Once per month on the last working day, with the date recorded
4. Platform and surface	Which AI platform and interface is measured	Google AI Overview (desktop) primarily, Gemini app secondarily
5. Measurement method	Who runs it, with what tooling, and how subjectivity is avoided	Run by the provider using their own monitoring, with screenshots supplied

Miss any one of these and the number cannot be verified.

Precise definitions of five common AI metrics

Metric	Definition	What it shows
Coverage	Questions mentioning you ÷ total questions tested	Breadth: how many questions you appear in
AI mention rank	Your average position when mentioned	Position: how prominently AI treats you
Share of voice	Your mentions ÷ all brands' mentions	Relative share: you versus competitors
Cited question count	How many answers reference your link or content	Depth: not just named, but used as a source
Mention count	Total times your name appears in answers	Frequency: raw exposure volume

Important: "mentioned" and "cited" are different metrics. A mention is just your name appearing; a citation means AI used your content or link to answer, which is usually more valuable. The contract should require them to be reported separately rather than merged into one figure.

0% and a dash are not the same thing

This is the most common source of misleading reporting in practice:

Display	Meaning	Correct handling
0%	Measured, and you were not mentioned at all	Show it as a genuine gap and analyse the cause in the report
—	No data yet (insufficient sample, not run that day, or out of scope)	Label it clearly as "not measured"; never count it as 0%

Why does the distinction matter so much?

• Treating a dash as 0% makes performance look like it suddenly collapsed, leading to wrong decisions
• Reporting 0% as a dash hides a real gap and sends optimisation in the wrong direction
• Mixing the two in one trend line makes "33% versus 0%" impossible to read as either a real decline or sampling noise

The contract should require reports to distinguish them and to state the sample size (n) and test date next to every figure.

Sample contract KPI clauses

The following can be placed directly into a contract or a quotation annex:

Item	Clause wording
Term	6 months, with the first month as the baseline period and monthly reviews thereafter
Target questions	5 questions, listed in Annex A; changes only by written agreement
Measurement platform	Google AI Overview (desktop) and the Gemini app
Measurement frequency	Once per month, on the last working day of the month
Primary KPI	From month 3, coverage of 60% or above
Secondary KPI	Average AI mention rank of #3 or better
Report contents	Coverage, AI mention rank, share of voice, cited question count, with sample size, date and screenshots
Missed target	Service term extended until the target is met, at no additional charge
Exit clause	If the target is missed two months in a row, the client may exit on 30 days' notice without penalty
Not guaranteed	No guarantee of 100% recommendation on any platform, and no guarantee of a specific number-one position

The acceptance process: from baseline to monthly review

1. Confirm the baseline: measure once after signing and record the current coverage, rank and share of voice. Both sides confirm in writing; this is the reference point for everything that follows.
2. Lock the question list: fix the target questions as an annex. Any change requires written agreement — otherwise a missed target can be turned into a hit simply by editing the question list.
3. First optimisation round: content, structured data and brand entity consistency work, typically taking several weeks.
4. Monthly measurement: same questions, same platform, same method, recording n and the date.
5. Target assessment: the primary KPI decides, with secondary KPIs as reference. Consecutive misses trigger the handling clause in the contract.

Seven traps to avoid before signing

• Only "a monthly report" is promised: no success criteria, so no report can ever be verified.
• Promises of 100% recommendation across all platforms, or guaranteed first place: recommendations are produced by the model, and nobody can guarantee them.
• The question list can be changed at will: this removes any constraint from the KPI.
• Reports show one aggregate number only: this hides per-question gaps, which should be listed individually.
• No sample size or date recorded: figures cannot be compared, and trends cannot be separated from noise.
• Treating "mentioned" as "cited": this inflates results; the two should be reported separately.
• No missed-target arrangement: "we will do our best" is not a commitment.

Common questions

Q: How long should the contract term be? A: Six months is a sensible starting point. The first month establishes the baseline and completes upfront optimisation; citation rates start to move between weeks 4 and 8; only after that is there enough data to judge performance. Terms shorter than three months rarely allow enough time to show results.

Q: How high should the KPI be set? A: It depends on competition and the website's baseline. For highly competitive categories (medical, finance, legal), 60% coverage by month three may already be a good result; for niche categories it can be set higher. The key is that the KPI is derived from baseline data rather than picked out of thin air.

Q: What if the provider hits the KPI but the business does not grow? A: That is a KPI selection problem. Consider adding commercial metrics to the contract, such as whether the cited questions are high-intent, and whether they drive enquiries or quote requests. Chasing visibility numbers alone does not necessarily produce business.

Q: Can we ask the provider for the raw data? A: Yes, and you should. Ask for the test dates, the question list, the platform, the screenshots and the sample size so you can reproduce and check them yourself. A provider who refuses to share raw data makes performance very hard to verify.

Summary

What matters in a GEO contract is not the price but the definitions: which platform is measured, how many questions are tested, by what method, on what date, and what happens if the target is missed. Just as importantly, "mentioned" and "cited" must be reported separately, and "0%" must be distinguished honestly from "—". Get those written down and both sides will share the same acceptance criteria six months later.

To establish your own baseline first, request the free AI visibility report or ask for a quote.

Further reading: Hong Kong GEO service pricing guide, Measuring GEO performance: coverage and share of voice, How to choose a reliable GEO agency in Hong Kong, How to get Gemini to recommend your brand