Introducing Risn AI: AI search visibility, measured honestly

Every brand now has a presence in AI search. ChatGPT cites you, or it doesn't. Perplexity surfaces your page, or a competitor's. Gemini picks one source for "best CRM for hospitality" and that source gets the click, or the answer.

This is a measurement problem. And so far, the tools chasing it have answered it the easy way: count the mentions, plot the line, report the number. That's correlation. It tells you what changed. It cannot tell you why. And in a market where the engines themselves are moving under your feet, "why" is the only question that matters.

We built Risn AI to answer the harder question.

What we track

The surface layer is what you'd expect from a serious AEO/GEO platform:

• Brand visibility across ChatGPT, Claude, Gemini, Perplexity and the rest of the answer-engine surface, sampled daily.
• Page-level and site-level scrape diffs, with content versioning and AI-generated changelogs for both you and your competitors.
• Multi-provider SERP collection (Serper, Tavily, Nimble, Exa, Perplexity) so the data isn't hostage to any one search backend.
• Citation, ranking, and answer-quality signals over time, stored in ClickHouse and queryable in milliseconds.

That's table stakes for the category. It is not why we built the product.

What no one else is doing

The dashboards on the market today have a quiet structural problem: they conflate three different causes of visibility movement.

• Your interventions: the page you shipped, the schema you added, the PR you placed.
• Competitor moves: the 47 comparison pages a rival quietly published last Tuesday.
• Engine drift: a model version bump, a retrieval-stack tweak, a re-indexing event the engine never announced.

When the line goes up, the team that pushed last gets the credit. When it goes down, the engines get the blame. Neither claim is verifiable. You ship the next thing and hope.

Risn AI is the first AI-visibility platform built around causal inference rather than correlation. That single shift changes what the product can say:

• Counterfactual decomposition on every visibility movement. You see the line you observed and the line you'd have seen if the change in question had never happened, with the gap between them quantified.
• 95% credible intervals on every effect estimate. No more "your traffic went up 5%; your edit worked." Instead: "+1.2% [-1.8%, +4.1%], posterior 0.42, indistinguishable from zero." When the answer is we don't know, the product says so.
• Every result ships with the tests that try to break it. We re-run the same analysis on the period before your change happened. If it "detects" an effect that couldn't exist, the method is wrong and we flag it. We also measure how strong a hidden, unmeasured factor would have to be to flip the result. The evidence a skeptic would need to challenge the number ships with the number.
• Cause separation. Your edits, your competitors' moves, and engine-side updates are modelled as distinct interventions and ranked by attribution share. When Claude's visibility drops 12%, the product tells you whether it was the engine, the competitor, or you. With a number, not a vibe.
• Posterior probability, never p-values. Bayesian methods don't produce statistical significance, and we won't pretend they do. We tell you how confident the data lets us be.

The research behind it

Risn AI is built by an R&D team of PhDs whose work spans peer-reviewed publications and granted patents in machine learning, information retrieval, and applied statistics. We've spent careers on measurement problems that look easy on the surface and turn out, on closer reading of the data, to be much harder. AI search visibility is one of those problems.

The methods we use draw on decades of work in causal inference and time-series modelling. It's the same statistical lineage that powers ad-attribution systems at the largest internet companies. Bringing those methods to LLM-mediated search is its own research problem: the engines are non-deterministic, the retrieval layers are opaque, the outcome signals are sparse and noisy, and the interventions you care about don't behave like classical A/B tests. The work that goes into making causal estimates trustworthy in this setting is the work the product is built on.

What this means for you

You stop guessing.

The next time your AI visibility moves, you'll know whether it was the page you shipped or the engine that shipped under you. The next time your team wants to celebrate a +5% week, you'll know whether the lift was real. The next time a campaign looks flat, you'll know whether it was the campaign, or whether something in the engine masked a real win.

That's the product. More to come: case studies, deeper looks at the methodology, and a closer examination of why every dashboard you've used so far has been telling you a story that the data doesn't actually support.

The Risn AI R&D team