Before You Buy GEO, Ask Which Page Will Change

Imagine your team has just received a proposal promising “AI-search visibility.” The deck has familiar words—GEO, AEO, citations, entity signals—and a chart showing where your brand might appear. Before debating the chart, ask a less glamorous question: Which page on our site would you change first, and what would a buyer understand afterward that they cannot understand today?

If the answer is vague, the work may be vague too. An AI-search project should leave your business with clearer public information, a way to check whether people can find it, and a record of what changed. A promised mention in a chatbot is not a substitute for those deliverables.

Make the proposal point to a page

Choose one page that matters to a real buying decision: a service page, a product detail page, a comparison page, or a page explaining who you serve. Ask the provider to annotate it. What buyer question is unanswered? What fact is buried or ambiguous? Who inside your company can verify the proposed answer?

For example, a 30-person B2B company might describe what its software does but never state whether onboarding is self-serve, assisted, or handled by a partner. A proposal that identifies this gap, drafts an accurate answer, and names the person who will approve it has more substance than a promise to “optimize for ChatGPT.” That example is hypothetical, not a customer result.

The page test is not a claim that one edit will earn an AI citation. It is a way to see whether the provider understands your business and can produce something you own after the engagement ends. Google’s guidance on hiring an SEO recommends asking what changes a provider will make, why, and how success will be measured. It also warns against ranking guarantees and tools that imply access to Google’s internal ranking data.

Ask for a deliverable, not a vocabulary list

A useful first-phase scope can fit on one page:

1. One buyer question: the exact decision a prospect is trying to make.
2. One current page: the URL that should answer it, or a documented reason a new page is needed.
3. One evidence gap: the missing or confusing fact, with an owner who can confirm it.
4. One revision: a proposed change to the visible page, with any supporting technical work explained separately.
5. One measurement plan: what will be checked before and after, with the limits of that evidence stated up front.

This does not dismiss technical SEO. A page still needs to be accessible and eligible for the search system in question. But the technical work should serve an answer worth finding. In its guide to generative AI features in Google Search, Google emphasizes useful, non-commodity content and foundational SEO. It says there is no special schema markup required for its generative features and no need to rewrite a page merely for an AI-specific format. That is Google-specific guidance; it does not describe every assistant’s retrieval system.

Separate platform access from platform promises

A vendor should identify which platform a recommendation concerns. Google Search eligibility, ChatGPT search access, and performance in other assistants are not interchangeable. OpenAI’s publisher guidance says that allowing OAI-SearchBot helps make public content available for ChatGPT summaries and snippets. Its search help page also says placement is not guaranteed. Allowing a crawler is an access decision, not a purchase of visibility.

When someone recommends a robots rule, schema change, new article series, or special file, ask: Which platform documents this? What problem does it solve for our page? How will we know the implementation worked? “Because GEO” is not an answer.

Measure in layers

A sensible baseline has at least three layers. First, check whether the page and its facts are published, accessible, and current. Second, observe search visibility and relevant referral traffic where the platform provides it. Third, watch for business actions: qualified visits, inquiries, signups, or purchases. Keep a dated record of the question tested, the answer returned, and the page changed; one favorable answer is not evidence of durable ranking.

This is why a proposal should tell you what it cannot measure. A dashboard may record a query observation, but it cannot credibly promise that every buyer will see the same answer across assistants. Google explicitly cautions that third-party tools do not have access to its internal ranking or AI systems. Treat a tool as a way to organize evidence and work, not as a source of secret placement.

Use the page test before signing

Ask each prospective provider to bring one annotated page to the next meeting. You are looking for a specific buyer question, a factual gap, a feasible revision, an owner on your side, and an honest baseline. If they can do that, you have something concrete to scope and evaluate. If they cannot, a larger “AI visibility” package is unlikely to become clearer after you sign.

Already know which page to inspect? Our five-minute buyer-facts audit gives you a starting checklist. Then try the free one-question AI visibility check as a directional observation—not a ranking guarantee or a substitute for the page work.