You're either in the answer, or you don't exist.
ORION asks ChatGPT, Claude, Gemini and Perplexity the questions your customers actually ask, then shows you their answers verbatim, tells you who got named instead of you, and shows you the exact pages they cited. Then it watches it change.
Same brand. Same day. Two questions.
Below are real answers from a real scan. One question makes this business the hero. The other never mentions them at all. Both were asked on the same afternoon, of the same engine, about the same company.
• Vessel: a functioning tall ship, not a motorised yacht.
• Views: marketed as the "best sunset table in Singapore."
• Price: around SGD 191 per person.
• Skyline Luge on Sentosa: race down illuminated tracks.
• Night Safari at Mandai: a night walk among nocturnal animals.
• Art jamming, pottery workshops, kayaking at Pulau Ubin.
Find out exactly where you fall out.
"Invisible in AI" is not one problem, it is four, and each one has a different fix. Guessing between them wastes months. These are the real numbers from that same scan: 12 engine responses across 5 questions.
Read it left to right. The engines went and searched, nearly every time. They named the brand in just over half the answers. But they only pulled a page from the brand's own website in 42% of them, and the specific page being checked was cited in 17%. So the brand is known, and largely quoted from other people's pages. That is a completely different problem from being unknown, and it has a completely different fix.
The engines were reading. Just not you.
For the question they lost, here are the pages the engines actually cited to build the answer. Every one of them is a Singapore roundup this business is absent from, while their competitors are on it.
Sorted by what you can actually do about it.
ORION collects every page the engines cited, then buckets them by who controls the page. A non-marketer can act on this on a Monday morning without asking anyone's permission.
Your own pages
The engines fetched your site and used it, but only for one question. Usually structure, not content: no direct answer near the top, no clean headings. Fastest win, needs nobody's approval.
Roundups and directories
Third-party pages the engines lean on that take submissions and updates. In the real scan above, this was five named publications, and the brand was on none of them.
Someone else's page is the source
The engine treats a competitor's own content as the authority. You will not take that page. You can build the thing it does not cover.
Then it watches, so you find out before your pipeline does.
AI answers are not stable. A model updates, a competitor gets added to a roundup, and the answer your buyer sees changes without a single thing changing on your site. A one-off audit cannot see that. ORION re-asks the same questions on a schedule and tells you the day it moves.
This is the difference between a report and a product. A report tells you where you stood. A monitor tells you the morning you were replaced, while there is still time to do something about it.
Citability, measured five ways.
Every score decomposes into five pillars. Each has defined signals, a visible weight, and a reproducible calculation. These weights are read live from the scoring engine, so this page cannot disagree with the product.
Originality
Does this page say anything a model can't get elsewhere?
Retrievability
Can an engine actually fetch and parse it?
Information Extractability
Can a model lift a clean, self-contained answer out of it?
Observed Visibility
Do engines actually cite it, when asked?
Notability
Does the wider web treat this source as real?
SEO measured clicks. ORION measures citations.
When informational queries resolve inside an LLM answer, the click never happens. The new unit of distribution is the citation. ORION is the scorecard built for it.
The SEO era (1998 – 2024)
- Rank for a keyword
- Earn a click from a SERP
- Count impressions and CTR
- Measure what humans type into a search box
- Optimize for crawlers that index pages
The Citability era (2025 →)
- Be cited by an LLM answer
- Earn the citation, the click is the answer
- Count Observed Visibility across ChatGPT, Claude, Perplexity
- Measure what humans ask in conversation
- Optimize for retrievers that score chunks
We're in Research Release, and we say so on every score.
Honest measurement is worth more than convenient marketing. Until ORION meets all four validation thresholds, every live score carries a Research Release disclosure.
What has to be true before we call this calibrated.
Until every one of these clears on held-out data, ORION does not call itself calibrated and does not call a score a probability. It is currently a research release, and every score says so.
See your Visibility Shortfall.
Sign up free. Paste your URL. ORION scores you across C1, C2, Q, T, then runs live LLM queries to measure Observed Visibility, compute the Gap, and rank the highest-lift changes to close it.
⚡ Get started free