What Makes a GEO Tactic Worth Your Investment?

Imagine a marketing report lands on your desk with three recommendations: fix a page that search engines cannot index, publish a guide explaining your implementation process, and pursue inclusion in an industry roundup.

All three might deserve attention. But if the explanation for each is simply “this will improve your AI visibility,” you still don’t have enough information to decide what to fund first.

What was actually observed? Why should the proposed action help? How confident should you be? And what would count as a useful result?

Those are the questions I want businesses to ask about generative engine optimization, or GEO: the work of improving how a business appears in AI-generated answers.

As a founder of PaperHorse, I have a commercial interest in businesses taking that work seriously. I also believe anyone selling GEO advice owes customers an explanation of its basis. That includes us.

My standard is straightforward: a recommendation should identify the problem it addresses, the evidence supporting the action, and how we will judge the result. The confidence of the advice and the size of the investment should reflect what we actually know.

Be precise about what “works” means

Suppose your team fixes an indexing restriction on a product page. A later check confirms that the page is indexed. The repair worked against its immediate objective.

Whether that repair earns additional AI recommendations is a separate question.

Google’s guidance for AI Overviews and AI Mode, updated December 10, 2025 and checked October 12, 2026, makes the boundary clear. A page must be indexed and eligible for a search snippet to appear as a supporting link. Meeting the requirements does not guarantee that Google will serve it.

That gives a business a strong reason to fix an unintended restriction on an important page. It gives us no basis for promising a particular increase in recommendations afterward.

The same discipline applies to other work. Publishing a useful guide, getting it cited, having your business recommended, and receiving a qualified inquiry are different outcomes. Before approving a tactic, name the outcome it is supposed to influence.

Microsoft makes a similar distinction in its February 2026 introduction to AI Performance in Bing Webmaster Tools. Its page-level citation counts describe how often a page is cited; they do not establish its importance, ranking, or placement within an answer.

A measurable result becomes useful when we understand exactly what was measured.

Ask what kind of evidence supports the advice

I would approach those three recommendations differently.

Recommendation	Evidence I would want	What that evidence supports
Fix an indexing restriction	A verified restriction affecting an intended public page, plus the relevant platform’s documented requirements	Correcting a specific eligibility problem
Publish an implementation guide	Important buyer questions, an inspection showing missing answers, and accurate information the business can contribute	Addressing a concrete information gap, then evaluating its effect
Pursue an industry roundup	The exact page recurring in relevant AI answers, its editorial criteria, and evidence that the business belongs in the comparison	Investigating a relevant opportunity and deciding whether to pursue it

Documentation, direct observation, and a reasoned hypothesis each have a place. Trouble starts when the explanation quietly changes categories: a page appeared in several answers, so a placement is sold as a reliable way to get recommended.

An observation can justify an experiment without establishing its outcome in advance.

Even official guidance needs careful reading. Google says no special structured data is required for its AI search features. A proposal to fix inaccurate markup can still be reasonable. A proposal sold as mandatory “AI schema” needs a different explanation. The platform, the specific defect, and the claimed benefit all matter.

When someone cites a study, I want to know what it tested. Did researchers change something and compare the result with a credible control? Or did they inspect pages that were already successful and describe their common features? Was the sample drawn from businesses like yours, on the engines and questions you care about?

A characteristic shared by successful pages is a clue. Those pages may also have stronger brands, better distribution, or more useful information. Copying one characteristic does not establish that you have copied the reason they succeeded.

New content deserves a specific business case

Consider a hypothetical IT services company. Prospects regularly ask AI whether switching providers will interrupt their work. The IT company has a documented transition process, but its public site says only that onboarding is “seamless.”

I would recommend the IT company create a substantive implementation guide. Explain the stages, who handles each task, what the customer must prepare, and where downtime might occur. Include a real example if the company has permission to share it.

The business case is clear: a consequential buyer question currently lacks a useful public answer. Publishing the guide creates something prospects, salespeople, and search systems can inspect.

Google’s people-first content guidance, updated October 5, 2026, asks publishers to assess whether their content contributes original information, substantial coverage, and value beyond a rewrite of other sources. That is a useful standard for the guide’s substance. It does not predict its AI visibility.

The guide might help a prospect evaluate the company even before we observe any change in AI answers. That value belongs in the decision. So does the possibility that it becomes useful source material for relevant answers.

What I would not attach is an invented forecast of how many additional recommendations it will earn.

The same reasoning can justify a new service page, a comparison, a pricing explanation, or a customer case study. The recommendation gets stronger when it names the missing information, the buyer decision it supports, and the evidence the business can supply.

Match the commitment to the uncertainty

Businesses cannot wait for a controlled study of every marketing decision. They can make the size of the commitment reflect the risk.

An inexpensive correction to a verified error may deserve immediate action. A guide answering an important question can justify investment through several uses. An expensive placement whose value depends entirely on more AI recommendations needs stronger supporting evidence and tighter spending limits.

I would start an uncertain tactic with a scope small enough to learn from: one placement, a defined group of pages, or a specific customer need. Agree on the cost and review point before starting.

That review point is a business decision, not a promise about when an AI engine will respond. Consider whether the work has been published, whether the relevant systems can access it, and whether enough comparable observations exist to say anything useful. If the evidence is still thin, say so before renewing or expanding the work.

The question is how much uncertainty the proposed spend can reasonably carry.

Decide what would change your mind

A recommendation becomes difficult to evaluate when every possible outcome is treated as a reason to keep investing.

If visibility improves, the tactic gets credit. If visibility stays flat, it needs more time. If visibility declines, the answer is to do more of it. That reasoning gives a business no way to make a different decision.

Before starting, I would write down what would support continuing, what would require investigation, and what would lead us to stop or change course.

For the hypothetical implementation guide, the brief might look like this:

Decision	What we agree before starting
Problem	Prospects need to understand transition risk, and our public site does not explain the process
Action	Publish an accurate guide reviewed by the people who deliver the service
Immediate check	The page is accessible and answers the agreed questions with specific, verified information
Outcome review	Examine relevant AI answers over repeated checks, alongside prospect feedback and the sales team’s use of the guide
Next decision	Improve unclear sections, investigate continued gaps, or expand coverage when the results justify it

If the page is useful to sales but we observe no AI improvement, report both findings. The business may reasonably keep the page. It has not yet demonstrated a GEO gain.

A better result should survive closer inspection

Suppose recommendations increase after the guide is published. That is encouraging. To judge the evidence, I would ask whether we compared the same questions, engines, locations, and available settings. Did the improvement recur? Did the company also launch a campaign, gain press coverage, or update other pages?

Where practical, comparable questions or pages unaffected by the work can provide additional context. They are imperfect comparisons: different topics can move differently, and one site change can affect more than its intended page. Still, they can help challenge an explanation that credits every improvement to the latest action.

Our article on why AI visibility changes covers that measurement problem in more detail. For a spending decision, the principle is simple: record what changed, preserve the original answers, and describe the result at the level the evidence supports.

An improvement after a change is an observation. Consistent results across comparable checks strengthen the case. A design that credibly rules out alternative explanations is needed for a stronger causal claim.

I would become more confident in a tactic as those pieces accumulate, especially when it also produces useful business outcomes.

The standard I want PaperHorse to meet

PaperHorse starts with a business’s website, builds its profile, measures sampled AI answers, and turns findings into prioritized actions. For eligible recommendations, it can also draft supporting content. Later reports show how the observed answers change.

That workflow makes the quality of the recommendation important. I want customers to understand why an action deserves attention, what remains uncertain, and what they should inspect after doing the work. A completed task and a higher score deserve scrutiny, including when they come from our product.

For your next GEO recommendation, ask the person proposing it to complete this sentence: