What a GEO Provider Can and Cannot Promise

That distinction matters when you are deciding whether to pay for the work. A promise that your company will appear in AI answers sounds reassuring. A promise to produce an audit and a monthly report sounds easier to verify. Neither tells you enough about whether the agency will make a useful decision for your business.

Publishing new pages and reporting more brand mentions does not tell you whether customers are getting useful information. If the answers describe the wrong capabilities or recommend the company for a service it does not offer, more visibility has not resolved the problem.

This is why the agreement needs to cover both delivery and evaluation: what will the agency do, why does it expect that work to help, and how will it check afterward?

Set a goal the agency can explain#

A useful AI visibility goal names the customer need and the change you want to see. If the problem is inaccurate product or service information, the goal should include correcting that information in the answers being monitored. More mentions alone would not show whether the problem had improved.

The agency should explain the evidence behind its recommendation and the assumptions behind any expected improvement. A forecast can be useful without being certain. A deadline for completing a page update is a delivery commitment; a deadline by which an AI platform will recommend that page is a prediction about an outside system.

Google makes this distinction explicit in its own guidance. It warns against providers that guarantee first place in Google Search. Its guidance on third-party SEO tools also explains that those tools cannot see Google's internal ranking data or guarantee performance. These statements concern Google; they are not a description of every AI product's policies.

For Google's AI Overviews and AI Mode, even meeting the technical and content requirements does not guarantee that a page will be indexed or served. Making a page eligible to appear and having it selected for an answer are different things.

The practical question for an agency is therefore how it expects its proposed work to improve your situation, and what would make it reconsider that recommendation.

The agency should make a recommendation you can act on#

An agency's analysis should lead to a specific recommendation. If it says product information is unclear, it should identify what is unclear, show the relevant evidence, and explain which changes it recommends first. Your team should not have to turn a collection of screenshots into an action plan on its own.

A recommendation should name the information that needs correcting, the pages or listings affected, and how the agency will verify the change. It should also explain the evidence connecting that work to the problem. Unclear wording is something to investigate; it does not by itself establish what caused an AI answer.

Implementation depends on the engagement. If your team will make the changes, the agency should provide usable guidance and the agreed review of the finished work. If implementation is delegated to the agency, it should name what it will deliver and what access, facts or approvals it needs from you. Consulting and full execution can both be useful, provided the responsibilities are clear. We discuss that choice in when to keep AI visibility work in-house and when to hire an agency.

The work may need to reach beyond your website#

Nixal's view is that an agency should also assess whether customers have useful evidence beyond the company's own claims. When relevant outside coverage is missing, the recommendation may need to include PR, industry media, social media creators or video reviewers, alongside website changes.

Your website can explain the product's features and who it is for. Customers may also want to know how it works in practice, where it falls short, and how it compares with other options. Industry reporting, creators' firsthand experiences, and video demonstrations or reviews can add that information. The agency should identify what is missing and work with suitable PR partners or creators to address it.

The agency should explain which missing evidence it wants to address, which audiences and publishers are relevant, and why. Depending on the engagement, it could coordinate with your PR agency, approach suitable creators, arrange product access or demonstrations, prepare verifiable facts, and follow up on factual questions. If that work needs a separate PR or creator partner, the proposal should name who manages it and whether the cost is included.

The commitment should be specific about the arrangement. Earned editorial coverage depends on the publisher choosing to cover the story. A paid creator agreement can specify deliverables and dates, but the resulting content should be identified as sponsored rather than presented as independent endorsement. Neither arrangement makes future AI citations certain.

What matters is whether the published material adds something customers can use: an explanation, a demonstration, a comparison or a documented experience. The agency should assess that contribution and monitor what happens afterward, rather than treating every placement as an equal success.

Agree how progress will be measured before work begins#

A GEO agency should establish a starting point, define what improvement would mean, and agree how it will compare results after the changes. You should know what will be measured before seeing the first progress report.

Start with the questions customers ask, the AI platforms being checked, and the product or service facts the answers should get right. Preserve the original answers and dates before making changes. Agree the review schedule and who will collect the observations, assess them and recommend the next step.

The report should connect four kinds of evidence:

For external work, distinguish a pitch sent, a collaboration agreed and a piece published. Then assess the published material against its purpose. Record any identifiable referral visits or inquiries separately, and check whether the source appears in the AI answers being monitored. Publication, customer engagement and AI citation are different outcomes; a count of placements alone does not establish the latter two.

These measures need definitions. Mention rate can mean answers naming the company divided by all successfully recorded answers in the agreed sample. Recommendation rate should count answers that actually recommend the company for the stated need, using the same defined sample. A citation to your website is a separate observation: it does not automatically mean the answer recommended your product. For accuracy, identify the facts being checked and distinguish a wrong statement from a fact the answer did not address.

Report the number of answers containing a recommendation alongside the total sample size, rather than showing a percentage alone. A higher recommendation rate in the recorded sample does not establish how often all customers see that recommendation, whether the change will persist, or whether the agency caused it. Show which answers changed and whether the comparison conditions stayed consistent.

Use the same core questions, platforms and repeat-count plan for the comparison, recording the dates and relevant settings such as language and location. Show results by platform before combining them. Record failed checks rather than quietly treating them as answers with no mention. If the agency changes the questions, tool or collection method, it should explain the effect on comparability. New questions can be useful; they should not silently replace an unfavorable baseline.

Monitoring software can collect and organize observations. The agency still needs to read the answers and explain whether a movement matters to the business. If more than one tool is used, identify what each contributes; do not average different scores as if they measure the same thing.

For business results, report the traffic and conversions that can actually be identified, and state what remains untracked. A customer's report that they found you through AI is useful additional evidence, but it should be labeled separately from a tracked visit. Neither a missing referral label nor an increase in mentions settles the business impact.

The review should end with a judgment: what improved, what did not, how confident the agency is in the comparison, and whether to continue, change direction or stop. We explain the distinction between these signals in AI visibility measurement.

What happens if the results do not improve?#

An agency should investigate a disappointing result and recommend a next step. Repeating that AI platforms are outside its control does not explain whether the agreed work was completed, whether the original diagnosis still fits, or whether the investment should continue.

The review might find an unfinished change, an approval that is still outstanding, or a problem that the proposed work did not adequately address. It may also find that the observations are not yet enough to judge. Each situation calls for a different response, and the agency should show what supports its explanation.

A further monitoring period needs a reason and a review point. A revised recommendation needs evidence. If there is no convincing case for more work, the agency should say so. Uncertainty is a reason to make careful decisions, not an unlimited extension of the engagement.

The same standard applies to a pilot. Its value is in completing the agreed work and helping you judge the next investment. Learning that an opportunity is weaker than expected can be useful, but a provider should not redefine success afterward to excuse work it failed to deliver.

Look for a promise you can examine#

When reviewing a proposal, look for a connection between the problem, the recommended work, the expected improvement and the review. These example phrases show where that connection can be missing:

Before signing, you should be able to explain what the agency believes needs fixing, what it will do about it, and what you will learn at the next review. That gives you a practical way to judge the service without requiring the agency to pretend it controls an AI platform.

If you are comparing proposals, see how to match a GEO provider's scope to the work you need.