We Spent Months Asking AI About Other People's Brands. Now We Have a Blog About It.
Every product has an origin story. Ours starts with a question that sounded simple and turned out to be anything but: What does ChatGPT actually say about your brand?
Not what you hope it says. Not from a single prompt that happened to go well. What it says on average, across dozens of queries, with enough statistical rigor that you would bet money on the answer.
Turns out, no tool did it the way we needed. And the ones that came close stopped at the numbers, leaving you to figure out what to do next. So we built something better.
The Measurement Gap
Ask ChatGPT a question today and ask it again tomorrow. You will get a different answer. Ask Claude the same thing and you will get a third. We confirmed this empirically: in a study of 3,837 AI queries across 16 brands, the four major AI providers agreed on their top recommendation only about a third of the time.
That means any single observation, from any single provider, is unreliable on its own. You need sample sizes. You need confidence intervals. You need to measure across providers, not just one.
Popsight uses confidence intervals to turn variable AI responses into measurements you can trust. You do not need a statistics background: Popsight handles the math and shows you clear ranges instead of misleading single numbers. 20 samples per query is enough to surface meaningful trends. From there, the Insights tab highlights where to focus first, such as queries where competitors dominate and you are absent.
Where All of This Is Happening
SEO taught marketers to care about search rankings. GEO (Generative Engine Optimization) is the same instinct applied to a new medium: AI-generated responses.
The scale is already significant. ChatGPT alone serves 800 million weekly active users. That is not a niche channel. Those are brand impressions happening entirely outside your analytics dashboard, in a medium that does not expose its responses to any crawler or index.
We built Popsight because we think marketers deserve real numbers about what is happening there, not anecdotes.
What Surprised Us Along the Way
A few things we did not expect when we started building. Some might change how you think about AI visibility.
Smaller AI models track trends just as well as expensive ones. You do not need GPT-4 or Claude Opus to detect whether your brand mention rate is going up or down. Cost-efficient models like Gemini Flash produce statistically comparable results at a fraction of the price. We wrote about this in detail in our research on AI visibility fragmentation.
AI providers do not agree with each other. ChatGPT, Claude, Gemini, and Perplexity each have their own biases. A brand can be the top recommendation on one provider and completely absent on another. If you are only checking one AI engine, you are seeing a quarter of the picture.
Confidence intervals change everything. Without them, a shift from 40% to 35% mention rate looks alarming. With them, you can see that the confidence intervals overlap and the shift is probably noise. Or you can see that they do not overlap and something genuinely changed. Either way, you make a better decision.
Visibility can shift overnight for reasons that have nothing to do with your brand. A provider model update, a change in system prompt weighting, a retraining cycle. What some researchers call LLM perception drift means brands can lose double-digit mention share in a week with no external trigger. Without continuous measurement, you would never know it happened.
Data without direction is frustrating. We assumed marketers mostly wanted more data points. What we heard instead was: "Great, I can see the numbers. Now what do I do?" That question shaped our roadmap. Popsight already surfaces early insights like competitor gaps and broken citations. We are building toward a product that does not just measure your AI visibility but helps you improve it, one recommendation at a time.
What to Expect From This Blog
This blog will mostly be data. We run thousands of AI queries. When we find something worth sharing, we will write it up with the methodology and the numbers, not just the conclusion. Research first, opinion second. But we will not stop at "here is what we found." Where the data points to something actionable, we will say so.
One More Thing
We debated what to call this blog. "The Popsight Blog" felt safe. "AI Visibility Insights" felt like it was optimized for a search engine we do not even target. Someone suggested "The Fish Tank" (our logo is an anglerfish, long story). We went with "Blog" because we are engineers who believe in shipping over naming.
If you have read this far, you are probably the kind of person who cares about measuring things properly and does not trust a single data point. You will fit right in here.
Welcome to the Popsight blog. The water is fine. The anglerfish is friendly. The confidence intervals are tight.
Want to see how AI talks about your brand and where to improve? Download Popsight free and run your first analysis in minutes. 14-day trial, all features, no credit card.