GEO is a narrative problem, not a search problem

A note from our CEO, Nick Loui: 

Every enterprise buyer we talk to now opens with some version of the same question. What are you doing about GEO? Generative engine optimization. Answer engine optimization. AI visibility. The label keeps changing, but the worry underneath it is constant and correct: people are asking ChatGPT, Claude, Gemini, Perplexity, and Grok the questions they used to type into Google, and the answer that comes back is shaping reputation, demand, and trust before a brand ever gets a say.

The category ported the wrong mental model

Look across the GEO tools on the market today and you'll notice they're all running the same playbook. Pick a set of prompts. Fire them at a handful of models. Count how often your brand shows up. Score the sentiment. Rank yourself against competitors. Wrap it in a dashboard and call it share of answer.

That's a real thing worth measuring. It's also, almost exactly, the SEO rank tracker rebuilt for a new surface. The industry looked at large language models, saw something that produces a ranked-feeling output, and reached for the most familiar tool in the drawer. Track the position. Watch the position move. Try to nudge the position up.

The problem is that an AI answer isn't a search ranking. It only looks like one.

You're optimizing a reflection

The rank-tracker framing gets one thing wrong at the root. A model doesn't store a fixed opinion about your brand that you can climb up a leaderboard. It synthesizes an answer, in the moment, from the material it was trained on and the sources it retrieves. That material is the open web, the news cycle, the forums, the reviews, the analyst notes, the social conversation. In other words, the exact narrative ecosystem that has always determined how the world sees you.

What the model says about you is downstream. It's the residue of narratives that already exist out in the world. So when a GEO tool tells you ChatGPT now describes your product as expensive, or names a competitor first, it's handed you a reflection in a mirror. Useful to see. But you can't fix your face by polishing the glass.

This is why so many teams feel stuck after they buy a monitoring tool. They get a number that moves and no real lever to move it with. The standard advice that follows, write more structured content, add schema, chase citations, is fine hygiene. It isn't a strategy, because it never asks the only question that matters: which specific narratives, in which specific places, are teaching the model to say this, and what do we do about them at the source.

There's no editor to call

We saw this play out firsthand. We had a public sector client in the middle of a geopolitical crisis. We knew the news would cover it. We expected the trolling on social, because that part always comes. What we didn't expect was the engine itself becoming a source of the problem.

People were asking Grok about the situation, and it was answering with misinformation, dressed up with citations that didn't exist. The model had invented its sources. To be fair, X has put real investment into mechanisms like community notes, and that work matters. Filing them well, quickly, and with solid sourcing is one of the levers we help clients pull. But it's one lever, not control, and no single correction mechanism moves at the speed of a crisis on its own.

The reframe, for us, was this. When a newspaper gets a story wrong, you can call the editor. There's a masthead, a corrections policy, a person on the other end of the line. There's no editor of Grok to call. You can't file a correction with a model. Every answer is generated fresh, from whatever narrative supply happens to be available to it in that moment.

So the only real lever left is the supply itself. If you can't edit the answer, you change what the answer is built from. That's not a GEO trick. It's a narrative truth that just shows up at its most unforgiving on the AI surface.

And it cuts the other way too, which is the part I find most exciting. Earned media has always been the hardest thing in our world to measure. You place a story, it runs, and then you squint at impressions and sentiment and hope it landed. The real effect on what people believed was slow, indirect, and mostly a guess. Now there's a new audience reading everything, all the time, and turning it into answers. When a narrative shifts, you can watch it move into what the models say, close to real time. The same surface that makes misinformation so dangerous is the one that finally makes narrative impact observable. Earned media just got something like a live scoreboard.

Our thesis: shape the supply, not the reflection

We think GEO is a narrative problem wearing a search problem's clothes. You don't optimize an AI answer directly. You influence the supply of narrative the answer is built from. The unit of work isn't the prompt result. It's the upstream content that feeds it.

That belief isn't new for us. It's the same conviction we've held since we started tracking how information moves online and took PeakMetrics to market in 2020. We've spent years building a system that watches how narratives form, spread, and mutate across media and social and the broader web, and that traces a claim back to where it started. GEO didn't require us to invent a new company. It required us to point the system we already had at a new surface, because that surface is fed by the same rivers we've always been mapping.

What AI Perceptions does differently

This is where the methodology gets concrete, and where we part ways with the point tools.

Start with how we see the world, because the product falls out of it. We don't think in channels. We think in narratives: single stories that form somewhere, travel, and resurface in new places, and our whole job is to follow them wherever they go. That's what narrative intelligence means to us. An AI answer isn't a new discipline that needs its own tool. It's just the newest place a narrative comes to rest.

So in PeakMetrics, AI Perceptions isn't a standalone product bolted onto the side. It lives inside the same repository as everything else we monitor. The same mention object. The same enrichment. The principle we hold across the whole platform is one repository, many ways to see it, and an AI answer is just another channel in that same record. It comes in, it gets enriched, it goes out to wherever your team works. In, enrich, out. The same as a news article or an X post or a podcast mention.

Because it sits in one record, we can do the thing a standalone tracker can't. We can connect the reflection back to its source. When a model says something about you, we're not stuck reporting that it happened. We can show the provenance: the article, the thread, the review, the claim that taught it. We can show you the same narrative surfacing in the news two months ago, then in social, then in the model's answer today, as one continuous story rather than three disconnected dashboards.

That maps to how we think about the work in three moves.

Detect what the AI is saying about you, across the engines that matter, in the prompts your buyers really use.

Decipher why it's saying it, by tracing the answer back to the narratives and sources feeding it, so you're looking at causes and not just symptoms.

Defend by acting upstream, on the specific content and conversations shaping the model, with the rest of your narrative operation in the same place rather than in a separate silo. The brands that get ahead of this are the ones who stop treating owned media, earned media, and AI visibility as three separate problems and start managing them as one picture. The same story runs through all three. Seeing them together is how you steer it instead of reacting to it.

A monitoring tool stops at detect. It gives you the number. We think the number is the least interesting part. The leverage is in decipher and defend, and you can't get there without provenance, and you can't get provenance without the source data already in the same system.

Why integration is the whole point

If you remember one thing from this letter, make it this one.

A standalone GEO tool is, by design, looking at the last mile of a story it never saw the start of. It can tell you the weather. It can't tell you the climate. The moment you treat AI answers as their own isolated channel, you've cut yourself off from the only context that explains them, and you're back to polishing the mirror.

Putting AI Perceptions in the same record as the rest of the narrative ecosystem isn't a packaging convenience. It's the methodology. It's what turns a vanity metric into an actionable one. The brands that win the AI answer era won't be the ones who got cleverest at gaming prompts. They'll be the ones who understood and shaped the narrative environment the models learn from, which means treating AI visibility as one expression of a single reputation problem instead of a new department.

That's the bet. It's the same bet we've been making since long before GEO had a name, now pointed at the surface where more and more decisions are getting made.

If you're evaluating GEO and your main question is which tool counts mentions best, I think you're asking the wrong question. The right one is: when the answer about me changes, will I know why, and will I be standing in a place where I can do something about it. That's the product we built.

Happy to show you what it looks like on your own narratives.

Nick