Brand monitoring is the closest GEO has to real measurement

Rand Fishkin published a piece this morning arguing that marketing teams should treat brand monitoring as a core operational habit, not a reputation-management afterthought. He listed seven reasons. The one that matters most for the people I work with is buried in the middle of the list, and he almost undersells it.

If you want to understand how your brand appears in AI answers, the starting point is not a prompt-tracking dashboard. It is a brand mention monitor.

That sounds banal. It isn't. It reframes a problem the industry has spent eighteen months trying to solve with the wrong instruments, and it points at the one workflow that is actually feeding the systems everyone is trying to influence.

What the industry keeps trying to buy

The pitch you have heard from every GEO tool vendor for the past year runs roughly like this. AI systems are answering questions instead of returning links. You need to know when your brand is mentioned in those answers. Therefore you should pay a subscription to a platform that runs prompts against ChatGPT, Perplexity, Gemini and Claude, logs the outputs, and reports on your citation share.

I've written about why this is measuring the wrong number. Prompt-tracking dashboards sample a synthetic universe of queries against a moving target, and the outputs shift for reasons that have nothing to do with your visibility. It's rank tracking in a hi-vis jacket. The number goes up and down, you build a report around it, and neither the number nor the report tells you what to do next.

The deeper problem is that these tools are trying to measure the output of a system without touching the inputs. AI models don't invent citations. They surface things that already exist on the open web — mentions, descriptions, comparisons, category framings, sentiment. If you want to influence what the model says about you, you need to know what the web says about you first. That is a brand monitoring problem, not a prompt sampling problem.

The input layer nobody's watching

Here's what actually happens when an AI system generates an answer that mentions your brand.

Somewhere in its training corpus, or in the live web results it retrieves, there is a description of what you do. That description was written by someone else — a journalist, a Reddit user, a review site, a competitor's comparison page, a directory listing, a podcast transcript. The model synthesises across those sources. It doesn't quote them verbatim. It builds a composite understanding of your brand from whatever weighted mixture of signals it happens to encounter.

If that composite is wrong, your citations will be wrong. If that composite is incomplete, your citations will be incomplete. If that composite doesn't exist because nobody talks about you, you won't be cited at all.

Rand's example in the piece is telling. He says people still describe SparkToro using terminology tied to a product they discontinued years ago — the fake followers tool. Every time that description survives on a real domain, it feeds back into how AI systems describe SparkToro. He notices it because he monitors brand mentions. He then reaches out and asks for corrections. That is knowledge governance done manually. It is also, functionally, GEO work.

The systems that answer questions about your brand are downstream of the systems that describe your brand. Everyone has been optimising the wrong layer.

Why this reframes the measurement problem

Prompt tracking asks: what does the AI say about me today?

The systems that answer questions about your brand are downstream of the systems that describe your brand. Everyone has been optimising the wrong layer.

Brand monitoring asks: what does the web say about me today, and how is that changing?

The second question is answerable. The first isn't, not reliably, because the answer changes with model version, retrieval configuration, prompt phrasing, and the user's context. But the input layer — the web — is stable enough to observe, and the observations correlate with the outputs in a way that prompt sampling doesn't.

If a new comparison article appears next week describing your product as a cheaper alternative to a well-known competitor, that framing will start propagating into AI answers over the following weeks. You can watch it happen. You can respond to it. You can decide whether to lean into the positioning or push back on it. That is a functioning measurement layer. It is not a dashboard with a citation-share percentage on it, but it is real information about a real thing you can influence.

The gap between these two approaches is where most of the wasted GEO budget in the UK right now is sitting. Teams are paying £500 to £2,000 a month for prompt-tracking tools that don't correlate with anything actionable, and paying nothing to systematically monitor the web sources those prompts are being answered from.

What the honest workflow actually looks like

You don't need enterprise software for this. You need three things running consistently.

First, a mention monitor covering your brand name, your key product names, and the ten to twenty descriptions you don't want propagating. Alertmouse is fine. Google Alerts is worse but free. If you have a developer, a simple pipeline against a search API costs less than a coffee per month. What matters is that you see new mentions within a day or two of them appearing.

Second, a review of those mentions on a fixed cadence — weekly for most businesses, daily if you're getting sustained press coverage. Not to react to every mention. To notice patterns. Which framings are catching on. Which competitors are being paired with you. Which outdated descriptions are still circulating. Which categories the industry keeps sorting you into.

Third, a small budget of time each month to correct, engage with, or amplify the mentions that matter. Corrections go to publishers directly. Amplifications go through your own distribution. New positive framings get promoted, cited in your own content, and referenced in the comparison pages you control. This is the part that's slow, unglamorous, and actually moves citations over a six to twelve month horizon.

Notice what's absent from this workflow. There is no citation-share dashboard. There is no prompt-response log. There is no AI visibility index. Those things describe outcomes you can't reliably influence in real time. This workflow describes inputs you can.

The counterargument worth taking seriously

Someone will read this and say: brand monitoring can't tell you which mentions actually make it into the model's training corpus, or which retrieved sources it weights when generating an answer. That's true. You can watch the web without knowing which specific pages are shaping any given citation.

But this cuts against prompt-tracking tools far worse than it cuts against brand monitoring. Prompt tracking can't tell you why an answer says what it says either. It just observes the surface. Brand monitoring at least observes the substrate the surface is drawn from. It is more information, closer to the input, at lower cost, with clearer implications for action.

The honest position is that nobody has a complete measurement layer for AI search. Every serious operator is working with partial visibility. Given that, spend your instrumentation budget on the layer where cause and effect are legible, not the layer where they aren't.

Why so few people frame it this way

Two reasons, I think.

The first is that brand monitoring sounds boring. It has been part of the marketing stack since PR agencies were faxing clip books. "Monitor your mentions" is not the kind of insight that gets an SEO influencer booked at a conference. GEO tools sound futuristic. They have dashboards. They have percentages. They have a story you can tell your board.

The second is that the tool economics don't favour honesty. A brand monitoring subscription costs a fraction of a GEO platform. If you're a vendor selling AI visibility software, you cannot afford to tell buyers that the ninety-nine pound tool sitting next to yours is doing more of the load-bearing work than your own product. So the conversation stays where the margin is.

That's the loop. And it's shaped, as most conversations in this industry are, by whoever is paying to have it.

What this means for the reader

If you're running SEO for a UK business and you've been quietly wondering whether the GEO tool you bought this year is telling you anything actionable — you're right to wonder. Cancel it if the honest answer is no. Redirect the budget into a mention monitor, a monthly review workflow, and a small retainer for someone to do the correction and amplification work.

If you're in-house at a bigger organisation and the board is asking about AI visibility, the right answer is not to buy more instrumentation. It's to build a brand monitoring practice that treats web mentions as the input layer they actually are. Report on trends in how your brand is being described, not on citation-share percentages you can't defend.

If you're an agency, this is a service line that already exists inside your workflow and probably isn't packaged as one. Rebadge it. There is more demand for structured brand monitoring right now than there is supply, because everyone is looking one layer downstream for a solution that lives one layer upstream.

The measurement problem in AI search isn't going to be solved by better dashboards. It's going to be solved, imperfectly and slowly, by the boring practice of watching what the web says about you and doing something about it. That practice has a twenty-year history under a different name. It just turns out to be the closest thing GEO has to a working measurement layer, and it was hiding in plain sight the whole time.