The GEO Handbook, Part 2: Sustainable Strategy Over Short-Term Tricks

A GEO trick may work today and become worthless tomorrow. A sustainable strategy does not depend on an exploitable gap. It invests in true information, reachable sources, useful content, strong evidence and an operating process that detects change.

By Kacper Ruta · GlasBox IT Studio, Malters/Lucerne · Reviewed: 12 October 2026.

This is Part 2 of the GlasBox GEO Handbook. Part 1 covered opportunities, risks and the trust problem. Part 2 addresses the strategic choice: which work stays valuable when Google, ChatGPT or Claude changes?

01 — The strategy question is not ‘What works today?’#

The wrong question is: ‘How do we force this sentence into an AI answer?’ The better question is: ‘Why should the system use this claim, and can a person verify it?’ The difference looks small. It separates a hack from a strategy.

A hack depends on an unnoticed weakness. A strategy depends on capabilities you own. These include subject expertise, primary data, clear pages, technical access, external evidence and a team that maintains change.

The familiar good-fast-cheap triangle only goes so far. A tightly scoped technical fix can be fast, affordable and good. The real trade-off appears across the lifecycle. Fast manipulation saves work today but creates rework, risk and explanation later.

Long term does not mean slow at any price. Move fast on a sound base. Fix a blocked crawler today. Correct a wrong address today. But do not publish one hundred pages before the offer, source and owner are clear.

02 — Why white-on-white text is not GEO success#

Machines read HTML differently from the way people see a page. Hidden text may therefore appear in a technical retrieval. That proves only that a parser captured text. It proves no quality, stable selection or business value.

Google explicitly lists white text on a white background as hidden-text abuse. Other examples include text behind an image, off-screen text and font size or opacity set to zero. The content is placed to manipulate search systems and kept from human visitors. [1]

Not every show-and-hide pattern is a problem. Google lists accordions, tabs, sliders, tooltips and text intended for screen readers as legitimate patterns when they serve the user. Intent and real usability matter. [1]

The right response to a trick that works briefly is not ‘do more’. Ask which weakness it exploits, which risk it creates and which clean alternative answers the same information need.

03 — Short-term efficiency can become long-term cost#

The chart is not a measured series. It shows the economic logic. A trick starts with little effort. Dependency and repair load then grow. A sustainable system needs more clarification at the start. The load later falls because sources, approvals and tests can be reused.

Calculate total cost of ownership. Include production, review, approval, monitoring, correction, migration and the risk of incorrect claims. Cheap copy with no owner can cost more than a strong page with an annual review.

Speed still matters. Apply one guardrail: no quick action may sacrifice truth, transparency or reversibility. Speed then becomes a capability rather than a risk.

04 — The five foundations of sustainable GEO#

1. Truth: define what is valid

Write down the real offer first. Name the region, audience, scope, exclusions, pricing logic and accountable experts. Mark every field that ages often. Without this layer, you optimise contradictions.

Acceptance is straightforward. An accountable expert confirms every concrete claim. Each volatile detail receives a source, date and next review date.

2. Access: separate technical decisions

Check status codes, canonicals, rendering, language paths, robots.txt, firewalls and bot rules. Treat search, training and user-directed retrieval as separate choices. OpenAI documents OAI-SearchBot, GPTBot and ChatGPT-User for different purposes. Anthropic separates Claude-SearchBot, ClaudeBot and Claude-User. [5][6]

An allowed bot does not guarantee visibility. A blocked search bot can prevent eligibility. Record the decision, purpose, owner and review date for each bot.

3. Utility: help with a real decision

Good content completes a task. It explains selection, limits, process, cost logic and the next step. For generative Search, Google recommends distinctive, useful content with first-hand perspective rather than interchangeable summaries. [2]

Ask after every section: which decision is now easier? If the answer is ‘none’, cut it or remove it.

4. Evidence: support what you claim

Use first-party data, documented processes, inspectable examples, named authorship and consistent external profiles. Google identifies clear authorship, production method and purpose as useful ways to assess reliable content. [3]

Evidence is not a row of logos. It must fit the claim. A membership does not prove project experience. A review does not prove a technical standard. A dated process record can support a specific working method.

5. Operations: keep the base alive

Every important page needs an owner. Every measurement needs a cadence. Every change needs a record. Without operations, even a strong launch becomes legacy debt.

Operations must create decisions: continue, change, merge or stop. A dashboard without a decision is decoration.

05 — Quality is not a feeling. It needs criteria#

‘We create high-quality content’ is not verifiable. Define quality so editorial, experts and management mean the same thing.

Google warns against scaled, unoriginal pages produced mainly to manipulate rankings. The policy applies regardless of whether a person or AI produced the copy. Page volume is not proof of quality. [1]

Google also advises against making a separate page for every possible query or fan-out variation. Systems understand meaning. Write for the task, not every phrasing of the same question. [2]

How to make content easy to cite

• Start a section with one complete core statement.

• Name the subject, location, time and limit where relevant.

• Separate fact, example and your interpretation.

• Link the original source directly to the claim.

• Give volatile figures a date or reporting period.

• Use stable URLs and unambiguous H2 headings.

• Avoid superlatives with no comparison basis.

• Do not write tiny fragments for machines. The section must serve people.

Google explicitly states that artificial chunking and special AI copy are not required for Google Search. llms.txt does not improve Search visibility or ranking there either. Structure content because readers need it, not because a myth demands it. [2]

06 — What you control, influence and only observe#

GEO success sits inside a tension. Your sources are under your control. Relationships can be influenced. Platforms and competitors can only be observed. Teams that mix these zones overpromise or invest in the wrong place.

The budget rule is simple: strengthen the inner zone first. Do not buy external reach for a vague service page. Do not request reviews while location and contact information conflict.

Competitors are context, not a control

You cannot prevent a competitor from publishing better data. You can prevent your expertise from remaining invisible. Use competitor research to find gaps. Do not copy their wording or page count.

Measure categories rather than vanity. Which decisions are answered? Which evidence is missing? Which regions are clear? Which action paths work? Those questions produce tasks. A generic share-of-voice score with no method often produces only debate.

Platform change is an operating risk

OpenAI notes that robots.txt changes may take about 24 hours to affect search systems. Anthropic documents separate bots and the possible visibility effects of blocking them. These rules can change. Keep documentation, review date and escalation path. [5][6]

Never build the process around one interface. Store raw data, prompts, screenshots, URL and date. The test remains inspectable when a product changes its interface.

07 — Build a resilient portfolio, not one bet#

GlasBox recommends a 70–20–10 starting allocation. It is a management heuristic, not a platform rule. Adapt it to risk and maturity.

The allocation prevents two failures. Day-to-day work cannot consume every experiment. A trend cannot consume the foundation. If the base is weak, move more budget into the 70% zone.

An experiment must never alter production truth. Use a defined scope, isolated page or limited dataset. Set success, failure and rollback before launch.

08 — Success factors: separate cause, signal and outcome#

A sustainable strategy needs leading indicators and business outcomes. Leading indicators show whether the foundation improves. Observation signals show whether systems respond. Business signals show whether value follows.

The figures in the table demonstrate precise reporting. They are not benchmarks. A figure without a sample, period and definition cannot be managed.

Since 31 August 2026, Google has provided a worldwide report for generative Search features. It reports impressions in AI Overviews and AI Mode by page, country, device and time. An impression remains a visibility signal. It is not a sale. [4]

No single GEO number

A composite score may support an overview when its formula and raw data are open. It must not replace component metrics. Otherwise the score can rise while incorrect answers or irrelevant mentions increase.

Management needs a few separate questions: Are our sources accurate? Are they reachable? Are we represented correctly? Can people act? Is qualified demand emerging?

09 — Lucerne example: eight strong pages instead of 120 location pages#

This fictional example continues the fiduciary firm from Part 1. The team serves small limited companies in Lucerne in German and English. An external proposal recommends 120 automatically generated location and service pages, plus hidden text blocks containing prompt variants.

The proposal looks efficient. It promises output. But the firm does not offer every service in every location. The English pages would be automatic translations. Nobody will review 120 pages. The project would simulate reach and accumulate debt.

The strategic alternative

The team selects eight decision pages: four core services in DE-CH and EN. Each page receives a clear audience, service limit, process, accountable expert, sources, contact path and review date. Local relevance comes from real responsibility and consistent profiles, not copied place names.

Eight pages are not automatically better than 120. They are better because scope matches the real offer and maintenance capacity. A new page follows when a real location or service creates a distinct information need.

Prioritise with an open formula

The team scores each task from one to five. Priority = impact × confidence ÷ effort. The formula is not a law of nature. It forces assumptions into the open.

These scores are also part of the fictional example. They demonstrate the method. A company may change the scales, but it should define them before scoring.

The 90-day plan

The plan guarantees no recommendation. It aims for a better-verified information base and a reasoned decision after 90 days. That can inform a responsible annual budget.

10 — Experiments need their own constitution#

Experiments matter. Without them, the team learns too slowly. Without rules, they turn the website into a laboratory with no safety boundary.

• State one testable hypothesis in one sentence.

• Limit the page, language, system and test period.

• Use no hidden, misleading or unapproved claims.

• Define baseline, success signal and stop condition before launch.

• Assign an owner and rollback path.

• Store raw data, screenshots, date and version.

• Check side effects on people, accessibility and existing search visibility.

• Move only repeatable learning into standard operations.

A failed experiment is not a loss if it creates a clear decision. An apparently successful test with no raw data is little more than a story.

11 — Annual operations: four quarters, four decisions#

Annual operations do not wait twelve months for the first decision. Monthly checks create operational actions. The quarter turns them into a management decision. The annual review sets scope and budget.

12 — Governance: who may change what?#

A small company does not need six departments. One person can hold several roles. The decisions should still remain separate. The writer does not automatically approve factual accuracy.

13 — Test every new GEO idea with seven questions#

• Does the idea help a real person make a concrete decision?

• Is the content visible, understandable and reachable for people?

• Is the claim accurate, and can we support it?

• Does it leave an owned asset: a source, data, process or relationship?

• Can we test benefit and side effect against a baseline?

• Can we roll the change back cleanly?

• Who maintains it after the test?

If two answers are no, stop before production. This small gate saves more budget than a later rebuild.

14 — Frequently asked questions#

Must sustainable GEO always be slow?

No. Sustainable means controlled and maintainable. Fix a clear technical defect at once. GEO becomes slower where truth, evidence or ownership first needs clarification.

Can a company run short GEO sprints?

Yes. Limit the objective, pages, systems, duration and acceptance. The sprint must not create hidden content or false claims. End with a continue, change or stop decision.

Is hidden text always prohibited?

No. Accessible screen-reader text or content inside usable tabs can serve a legitimate purpose. Hidden text designed only to manipulate search or answer systems is the problem. Google explicitly lists white-on-white text as abuse. [1]

How many pages does a GEO strategy need?

As many as distinct decisions, services and audiences justify as maintainable content. There is no useful minimum. One maintained core page may be worth more than one hundred interchangeable variants.

Does every page need structured data?

No. Structured data can support eligible Search features, but it is not a universal GEO requirement. Google says there is no special schema for generative Search. Markup must match visible page content. [2]

Which work best survives platform change?

First-party sources, clear authorship, consistent company data, stable URLs, documented tests and working contact paths. These remain useful when an interface changes.

How often should the strategy be reviewed?

Review operational errors monthly, core sources and pages at least quarterly, and portfolio and budget annually. Run an extra review after a material platform, offer or location change.

What is the strongest reason to stop?

When a tactic works only by deceiving people, lacks expert approval or has no maintenance owner. A short-lived signal does not justify a lasting liability.

15 — Conclusion: quality is the fastest long-term strategy#

Sustainable GEO is not moral decoration. It is risk management. Clear sources reduce correction cost. Strong pages serve several channels. Open tests speed up decisions. Owners prevent silent content decay.

You do not control the platform. You do control what it can find about your company. You control whether the claim is accurate. You control whether a person can act. That is the strategic leverage.

Sources and editorial basis#

Platform rules and technical claims rely on official documentation. The strategy, portfolio allocation, priority formula and worked example are the GlasBox operating model. They are not presented as a platform requirement or client result.