GEO needs more journalism

Rand Fishkin posted a useful challenge to AI visibility measurement. His argument, broadly, is that the dashboards look much more certain than the underlying behaviour.

We do not really know how many people are asking particular prompts, or whether the prompts being tracked by GEO platforms are representative of what customers are actually asking. A visibility score can look very definite even when the demand underneath it is not.

I think that criticism matters. But working on GEO has made me reach a slightly different conclusion from the obvious one: I do not think it makes GEO unimportant. I think it shows why GEO cannot just be a technical optimisation discipline.

The dashboard can tell you where to look

I have been working on Making Tax Digital content recently, and one of the areas our GEO analysis had already surfaced as worth investigating was quarterly updates. You could treat that as an optimisation problem: look at the prompts, look at competitor visibility, change the page and watch the score.

There is value in some of that, and I am not dismissing the technical layer. But I wanted to know whether the question existed outside the dashboard, because a weak answer to a synthetic prompt is only interesting if there is a real information need behind it.

So I went to Accountex, listened to HMRC and spoke to accountants and bookkeepers. Then I looked at what came out of an SMB advisory board, and quarterly updates kept appearing there too.

Not as a keyword, but as a real source of confusion. People had questions about what had to be submitted, when, what categories meant, how software connected to HMRC, whether something had been completed successfully and how quarterly updates differed from the final declaration.

That changed how I looked at the original GEO signal. The dashboard had shown me where to look; the journalism told me why it mattered.

That is a different job

This is the part of GEO I think we are in danger of skipping. If the prompt data is uncertain, the answer cannot be to act as though the prompt data is more precise than it is.

We need other evidence: customer conversations, practitioner interviews, search behaviour, support questions, events, research, sales conversations and actual confusion. That evidence does not replace AI visibility data. It gives it context.

A prompt appearing in a visibility platform might be useful because it exposes a gap, but before I decide that gap deserves content, I want to know whether the underlying question is showing up anywhere else. Is somebody actually struggling with this? Is it coming up in customer research, practitioner conversations or regulator guidance? Are support teams seeing the same thing, and do search patterns point in the same direction?

That is a much stronger basis for editorial judgement than a synthetic prompt set on its own. It also gives the content a chance of being better, because you are answering the real problem rather than merely reproducing the language of the measurement tool.

GEO is useful when it becomes a diagnostic layer

The technical layer still matters. Content needs to be findable and understandable, and search systems and AI assistants need to be able to retrieve it. It is useful to know whether your organisation is being cited, mentioned or represented accurately, and where competitors appear to have stronger coverage.

But I do not think those signals are enough to decide what content should exist. A dashboard can tell you that a question exists in the measurement system; it cannot necessarily tell you how important that question is in the market, or whether the answer you are producing reflects the real experience behind it.

That requires editorial work. It requires somebody to decide what evidence is missing, where the claim is too neat, who should be asked and whether the original signal survives contact with the outside world.

The question changes

Once you think about GEO this way, the question changes. It stops being How do we win this prompt? and becomes We know this question matters. Have we produced an answer good enough for people and AI systems to use?

I think that is a much healthier way to work because it removes some of the pressure to pretend we know things that we do not. If we do not know prompt volume, say so. If visibility is directional rather than definitive, treat it that way.

And if a synthetic prompt is only a hypothesis about customer behaviour, go and look for evidence that supports or challenges the hypothesis. That is where the work gets more interesting, because the task becomes investigation rather than optimisation theatre.

This is why I keep coming back to the GEO journalist

I have been using the phrase GEO journalist to describe part of what I think this role is becoming. I do not mean somebody whose job is to optimise text for machines; I mean somebody who uses AI visibility as one signal, then goes and finds out what is actually happening.

The journalist part matters because it changes the behaviour. You do not accept the dashboard as the story; you treat it as a lead, look for corroboration, talk to people and compare what the organisation says with what customers experience.

Then you look for the part of the problem that is missing from the existing content and build the answer around that. This feels much closer to the work I have actually been doing than most descriptions of GEO.

It also connects to something I wrote recently in When search becomes recommendation, evidence becomes the interface. I argued there that as AI systems start doing more evaluation and recommendation, the evidence underneath content becomes more important.

The measurement problem adds another reason to build that evidence layer. Evidence is not only what helps an AI system decide whether an answer is credible; it also helps the content team decide whether the question is worth answering in the first place.

That is a different kind of optimisation, and probably a more useful one.

---

This piece was prompted by Rand Fishkin’s post on AI visibility measurement.