GEO without a playbook

Working note: this is not a GEO framework. It is an observation from trying to work usefully in a discipline that is still being invented.

A lot of digital marketing has a playbook.

That does not mean the playbook is always right, or that following it produces good work. But if you need to learn paid search, email, SEO, conversion optimisation or marketing automation, there is an accumulated body of practice waiting for you. There are books, courses, benchmarks, case studies, certification programmes, experienced practitioners and years of arguments about what works.

You can disagree with the conventional approach after you understand it.

GEO feels different.

The terminology is still unstable. The tools are changing. The answer engines themselves are changing. Vendors describe the problem differently. Tactics that look promising may be based on a very small number of observations. Citation patterns move. Nobody has twenty years of institutional memory because the discipline has not existed for twenty years.

There are useful guides. There are smart practitioners. There are increasingly good tools. But there is not yet the kind of settled handbook that lets you say: learn these principles, follow this sequence, compare yourself with these benchmarks and you will at least understand the established practice.

That creates a different kind of work problem.

I cannot really learn this by copying what has already been done

I have been spending more time on GEO in my current work. Much of it is experimental because I do not have a template to inherit.

I can inspect which questions an organisation appears or disappears for. I can look at the sources AI systems cite. I can compare those answers with what the organisation already knows. I can ask whether the problem is poor structure, missing evidence, weak authority, inaccessible knowledge or simply an answer that does not yet exist in a useful form.

But the interesting part starts when that analysis needs to change real work.

At the moment I am trying something fairly simple around Making Tax Digital and an industry event.

Instead of treating weak AI visibility as an instruction to commission more GEO content, I have taken a small number of questions where the answer environment looks weak or incomplete and asked what is actually missing.

For one question, the basic factual answer already exists in plenty of places. Another summary of the same information would probably add very little. What appears more useful is practitioner evidence: what clients misunderstand, where the work actually appears, what gets chased or corrected and how someone doing the job would explain the distinction in plain English.

That changes the next action. The GEO signal becomes a reporting brief.

I can take the question into an event where practitioners are already present, gather the missing evidence, decide afterwards whether it belongs in an existing article, a direct-answer page, video, social material or somewhere else, and then see whether the answer environment changes after publication.

I do not know yet whether it will work.

That is the point.

The component parts are not new

The absence of a settled GEO playbook does not mean this work has no precedent. What I am finding reassuring is that the individual behaviours already exist in other disciplines. The experimental part is how they connect inside a generative-search problem.

Search listening has long treated search behaviour as a source of audience intelligence rather than simply an optimisation input. The idea is to analyse what people search for across a category and use those signals to make better decisions about campaigns, content and messaging. GEO creates a related opportunity: the answer environment itself can become an intelligence source. Instead of only asking which query has demand, I can ask which important answer is weak, generic, contradictory or missing useful evidence.

There is also established practice around using industry events as research environments. Quirk’s has documented qualitative research at trade shows and conferences, particularly for B2B audiences that are specialised or difficult to recruit. That makes the idea of using an event to investigate a defined question less unusual than the normal content-marketing framing might suggest. The event is not only something to cover. It can be temporary access to a concentrated field of expertise.

Research-led thought leadership provides another precedent. Work documented by the Bloom Group and Rattleback found that stronger thought-leadership programmes frequently rely on primary research, including surveys and interviews, to shed new light on important business problems. Again, the useful principle is upstream: work out what the market does not already know before producing another polished expression of what it does.

And the uncertainty around GEO itself is real rather than imagined. A 2026 critical survey of 45 GEO studies found heterogeneous terminology, metrics and evidence standards, and reported that no reviewed technique had yet demonstrated a stable, longitudinal, cross-platform causal effect on organic discoverability or downstream behaviour. Its recommendations include repeated measurement, controls and human validation. That is much closer to an experimental discipline than to a mature optimisation checklist.

So I am not inventing a research method from nothing. I am combining established behaviours in response to a new information environment:

search and answer intelligence → editorial diagnosis → field reporting → original evidence → activation → measurement

The genuinely experimental question is whether GEO data can make the evidence-gathering stage smarter before anything is commissioned, and whether the evidence gathered as a result later changes the answer environment.

In a mature discipline, you learn the method. In an emerging one, you have to learn how to learn

The absence of a settled playbook does not mean anything goes. It means the learning system matters more.

The useful sequence I am arriving at is not especially exotic:

orient → observe → form a hypothesis → intervene → measure → retain what survives

Orient means understanding the language, tools and current claims well enough to know what people are actually arguing about.

Observe means looking at live behaviour rather than relying only on guidance: which sources appear, which answers recur, where an organisation is absent, what the tools are measuring and how unstable those measurements are.

A hypothesis should be specific enough to fail. Perhaps a weak answer is missing practitioner evidence. Perhaps the evidence already exists but is stranded. Perhaps the page answers the question badly. Perhaps the problem is not editorial at all.

Then make the smallest useful intervention you can and watch what happens.

If it works, retain the learning. If it does not, retain that too.

This is much closer to reporting and product discovery than to following an optimisation checklist.

AI makes this kind of exploration much more possible

I doubt I would have approached GEO in the same way without AI.

That is not because AI knows the correct GEO playbook. It does not. In a field this young, confidently repeating somebody else’s emerging theory can be worse than admitting uncertainty.

The useful thing is that AI reduces the cost of entering unfamiliar territory.

It can help me compare competing explanations, interrogate unfamiliar terminology, summarise vendor claims, question my own assumptions, connect a visibility problem to things I already understand about research and editorial work, and keep track of what has been tried. I can move from I do not really know where to start to a testable working hypothesis much faster than I could by reading randomly until I felt qualified enough to begin.

That is a different productivity gain from writing faster.

It makes adjacent expertise more reachable.

The stable layer is still human judgement. My experience is in editorial work, evidence, interviews, customer stories, research and narrative development. AI helps me explore a new technical and strategic environment quickly enough that those existing skills can become useful inside it.

In that sense, the new capability is not GEO expertise appearing from nowhere. It is older expertise becoming portable into a new problem.

This is why I am wary of premature GEO frameworks

There is a temptation in every new discipline to turn early observations into rules.

A tactic appears to work. Somebody names it. A diagram follows. The diagram becomes a framework. Soon the framework is being taught with much more confidence than the underlying evidence deserves.

I do not want Editorial Intelligence to do that.

I have already written about GEO-informed thought leadership and the idea of becoming the GEO journalist, but both are deliberately framed as developing practice. The test is whether they improve real editorial decisions, not whether they produce a convincing diagram.

That is why the experiment I am taking into Accountex Manchester matters to me. It is small enough to be useful either way.

If a visibility gap leads to a better interview question, that is evidence of value. If practitioner evidence materially improves the resulting answer, that is stronger evidence. If the resulting content later changes how the organisation appears in AI answers, that is stronger again.

If none of those things happen, the method needs changing.

The learning system may matter more than the playbook

This is probably the broader Editorial Intelligence point.

Organisations increasingly have to make decisions in areas where practice is moving faster than documentation. AI search is one example, but it will not be the last. New tools, interfaces and behaviours will keep appearing before there is enough history to know what “best practice” really means.

In those environments, expertise cannot only mean knowing the established answer.

It also means being able to build a disciplined route from uncertainty to evidence: preserving what you observe, connecting it to what you already know, forming a falsifiable view, trying something useful and retaining the result so the next decision starts from a better position.

That is already close to the Evidence Engine and the wider Editorial Intelligence cycle. I do not think it needs another large framework yet.

For now, the useful idea is smaller:

When there is no playbook, your learning system becomes the playbook.

The advantage is not being first to declare the rules. It is getting better at discovering which rules deserve to survive.