GEO Handbook, Part 7: From Measurement to a Better Page

An AI answer mentions other providers. Your page is not linked. Or the answer describes your service inaccurately. The next decision matters: what can you improve on your website, and how will you know the work is complete?

Useful GEO work starts here with a customer question and a person responsible for it. We select an existing page, verify the finding, fill a knowledge gap and record the change. The first result should be a better page for people. Changes in AI answers are observed separately afterwards.

1 — A missing mention is not a diagnosis#

A screenshot shows one answer under particular conditions. It does not explain why a source was selected or omitted. Before rewriting, record the question, platform, mode, language, time and visible sources. Check the specific URL as well as the company name.

The possible task is a hypothesis. A missing link proves neither poor quality nor a technical fault. If your page already answers the question well, one omission does not justify another page. Recording the observation and checking again can be the right decision.

2 — Choose a page that supports a customer decision#

Start with a service you actually deliver and can explain from experience. Use questions from approved conversation notes, quote enquiries or consultations. Personal customer information does not belong in the published article. Questions that prevent a suitable prospect from enquiring are especially useful.

An existing URL is the first candidate when its audience, task and service already match. An additional section may be enough. A new page needs a distinct user task and enough original substance. A different town name or a slightly reworded question alone is not a sufficient reason.

Our priority is to correct false or misleading information first, then close important, verified gaps. Purely stylistic variants come later. This is an editorial decision aid, not a calculated GEO score or a claim about any platform’s ranking system.

3 — The brief: one question, one piece of evidence, one acceptance criterion#

A useful brief makes sense without marketing terminology. It describes the decision a person needs to make, the answer currently missing and the material supporting the addition. The subject expert checks the content; writing and design make it understandable. Publication follows that review.

This example is fictional. It is neither a verified GlasRein quote nor a measured customer result. A real project would record the document version, responsible person and approval date. If the evidence is missing, obtain it first rather than filling the gap with confident wording.

4 — Before and after: turn a promise into an answer#

Before: “We are your reliable partner for professional window cleaning in Lucerne. Our quality speaks for itself. Contact us for an individual quote.” It sounds familiar. A property manager still cannot tell which services to compare or what information an enquiry needs.

After — a template requiring expert approval before use: “For a comparable quote, we record glazing, frames and access conditions for each property. The quote states whether cleaning covers inside and outside and which components are included. Additional work and any access equipment are agreed separately. For an initial assessment, we need the property address, window photographs and the preferred date.”

The improvement is the reader’s ability to decide: scope, conditions and the next step become clear. A table is simply a suitable format here. It does not guarantee a citation. Turning every heading into a question is equally unhelpful if the answer lacks substance.

5 — Design the evidence: where specialisation and creativity matter#

Practical expertise provides details a general guide cannot. Design makes those details accessible. An annotated photograph can identify the components being discussed. A simple drawing can explain access. A short video can demonstrate a task when motion genuinely helps someone understand it.

Choose the medium to suit the question. A decorative image does not prove a capability; a chart without traceable data does not prove success. Real examples need context, a date and a defined scope. Clear the use of customer photographs and anonymised documents. Keep important statements available as readable text too.

Google highlights originality, effort and accuracy when discussing helpful content. This provides a useful editorial question: what can a person learn or accomplish on this page that an interchangeable summary does not offer? [2]

6 — What our example actually improves#

For an internal acceptance check, we define six questions. In this teaching example, an item passes only if the sample page contains a clear answer linked to an assumed approved internal basis. Merely mentioning a subject is insufficient. Both versions use the same checklist.

The remaining gap stays visible. That is more useful than artificial completeness. These are equally counted editorial checks, not equally important ranking factors. Three additional passes cannot be converted into a visibility uplift or a revenue forecast.

7 — Check technical discoverability deliberately#

Existing SEO fundamentals apply to Google AI Overviews and AI Mode. A supporting page must be indexed and eligible for a search snippet. Google requires neither special AI files nor a special schema for this. Meeting the requirements does not guarantee inclusion. [1]

Documentation checked on 25 October 2026. This table is not a blanket recommendation to allow every bot. Public content, internal documents and training preferences require different decisions. Access alone does not make a page relevant. A bot name in a log is not a verified provider request without checking its origin.

Also review the actual page: a working HTTP response, consistent canonical URL, internal links, important information in HTML and accurate structured data. JSON-LD describes visible content. It replaces neither evidence nor expert review. Change server rules only when a concrete finding justifies it.

8 — Accept, publish and record the change#

Expert review and technical checks answer different questions. Is the claim correct? Can people read and use it comfortably on the page? Both are part of completion. Mobile review matters: a clipped table or unreadable image caption weakens the very explanation you set out to improve.

Keep a short change log: “Added scope and access conditions to the quotation section; basis: service specification version X; reviewed by: name; published: date.” A fresh timestamp alone is not a substantive update. If a factual error appears after publication, correcting it takes priority over keeping an experiment tidy.

9 — Check afterwards: page quality and AI outcomes separately#

Editorial acceptance can be checked immediately. A platform’s selection of a source cannot be controlled in the same way. Use the Part 6 measurement plan for follow-up: the same questions, documented conditions and repeated observations. Record other website or service changes as well.

Set the next review date in advance to match your existing measurement cycle. It is an organisational date, not a promised search-system response time. A new mention may coincide with an edit without being caused by it. A useful page improvement may also leave the visible answer unchanged.

Then decide from the evidence: retain the change, make it more specific or reject an incorrect hypothesis. If the page helps people more but is not cited, that alone is no reason to remove useful information. Business outcomes additionally depend on relevant visits, qualified enquiries and feedback from your advisory team.

10 — Turn the course into a way of working#

The first six parts provide the foundations. Part 7 connects them in one small, executable task. You do not need a content factory: one well-chosen page, practical expertise and a verifiable change are a useful starting point.

Our principle remains: strength through simplicity, power through truth. The next improvement should be easier to understand and more precise about what you actually deliver. A future course instalment can build on this: how external sources confirm or distort the picture of a business.

Sources and working method#

Sources checked on 25 October 2026. Provider documentation describes requirements and controls, not the likely success of our sample page. The brief, before-and-after copy, checklist and diagrams are original GlasBox teaching examples. The assessment measures editorial coverage; no new AI measurement series was conducted for it.

11 — Glossary: key terms and how they connect to the course#