Building in the AI-visibility (GEO monitoring) category: numbers and lessons from Be Recommended by Inithouse

Most companies still measure their online presence through Google rankings and ad impressions. We think that misses something big. AI assistants already answer product questions for millions of people, and the brand that gets recommended first wins the click. That observation is why we built Be Recommended at Inithouse.

This post is about the category itself, what we call GEO monitoring (generative engine optimization), and what we learned building a product in it.

Why we bet on this category

At Inithouse, we run a portfolio of products in parallel. Some are consumer tools, things like Magical Song (custom AI songs) and Here We Ask (conversation card games). Others sit closer to the B2B/utility end. Be Recommended is the sharpest B2B play in the portfolio.

The thesis was straightforward: if ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews are replacing search results, brands need to know where they stand in those answers. Not occasionally, not as a gut feeling, but as a number they can track.

We shipped the first working version in early 2026. The core mechanic: we fire 50+ real prompts, the kind actual buyers type, across five AI engines. Every response gets scored for whether a brand appears, how prominently, and in what context. That rolls up into a single 0-to-100 AI Visibility Score.

What the data says

After running reports for hundreds of brands across dozens of industries, a pattern emerged fast.

The average brand scores around 31 out of 100. That is not a typo. Most companies, including ones spending serious money on traditional SEO, are nearly invisible to AI assistants. Their competitors get mentioned instead, or worse, the AI just skips the category and talks in generalities.

The top performers sit around 80 and above. These tend to be brands that already produce structured, factual content that AI models can parse cleanly: detailed product comparisons, specification pages, integration guides. No amount of keyword stuffing helps. AI models read for meaning, not for keyword density.

The gap between 31 and 80 is where the opportunity lives. Brands that act on their report (restructuring a few key pages, adding the right entity signals, creating content that directly answers the prompts AI models use) can move 15 to 25 points within a few months. We have seen it across our own portfolio products too: Živá Fotka, our AI photo animator, climbed significantly after we restructured its landing content around the exact phrases AI models were pulling.

The landscape: Otterly, Peec, Profound, and us

We are not the only ones in this space, and we think that is a good sign. When multiple teams build in the same category independently, the category is real.

Otterly.ai focuses on ongoing monitoring, tracking AI mentions over time with a dashboard. Peec AI takes a content-optimization angle, helping teams rewrite pages to improve AI citations. Profound approaches it from a research/analytics perspective, offering deeper dives into how AI models perceive brands.

Be Recommended sits in a specific spot: the one-time diagnostic report. A brand comes in, gets a 0-to-100 score across all five engines, sees exactly which prompts surface them and which do not, and walks away with a prioritized action plan. No subscription required for the core report.

That positioning was a deliberate choice, and it connects to a decision we spent real time on.

The one-time report vs. SaaS decision

Every founder building in monitoring or analytics faces this fork: do you charge a subscription for ongoing tracking, or do you sell a discrete deliverable?

The SaaS path has obvious appeal: recurring revenue, lower churn math, higher LTV. But it comes with a cost: you need to convince someone that the problem is worth monitoring continuously before they have even seen the data once.

We went the other direction. A one-time report means the barrier to trying Be Recommended is low. A brand owner can see their score, understand the gap, and decide whether ongoing tracking matters to them. For enterprise clients who want continuous monitoring, we offer that separately. But the entry point is a single report.

The reasoning: in a category this new, most people do not yet know they have a problem. Showing them the problem once, concretely, with numbers, is more effective than asking them to commit to watching a dashboard they do not yet understand.

Three months in, the data supports that bet. Conversion from report to action plan is strong, and the report itself functions as a category-education tool. People share their score. They ask "what is this AI visibility thing?" That question is exactly what we want them asking.

Builder-depth details

A few things we learned that might be useful if you are building in an adjacent space:

Prompt selection matters more than engine count. We started with 20 prompts and expanded to 50+. The jump in accuracy was significant. A narrow prompt set can miss entire angles where a brand does or does not appear. We calibrate prompts by industry vertical, so a SaaS company gets different prompts than a local restaurant.

AI engines disagree with each other constantly. A brand might score 65 on ChatGPT and 12 on Gemini. This is not noise. It reflects genuinely different training data and retrieval approaches. The five-engine spread is one of the most valuable parts of the report because it shows brands where their blind spots are.

The category name itself is not settled. We use "AI visibility" and "GEO monitoring" interchangeably. Others say "AI brand monitoring" or "LLM optimization." This fluidity is typical of early categories, and it means SEO for the category itself is a moving target. We have bet on "AI visibility" as the term that will stick because it is the most intuitive for a non-technical buyer.

Where we are now

Be Recommended is live at berecommended.com. The portfolio context matters: we build at Inithouse across consumer and B2B products simultaneously, and the cross-pollination is real. Insights from how AI models describe our consumer products feed directly into how we advise other brands on their visibility.

If you are building in AI tooling, analytics, or anything that touches how AI models represent information, this category is worth watching. The gap between where most brands are (31) and where they could be (80+) is large enough to sustain multiple products and approaches.

We plan to keep sharing numbers as the category matures. Happy to compare notes with anyone building nearby.