When customers Google your category, ranking position is a proxy for visibility — but it is also visible. You can read the SERP. You can see who outranks you. You can build the page that wins.
When customers ask Claude or ChatGPT or Gemini, the answer is generated in private and read in private. There is no SERP to inspect. There is no DOM to scrape. The model decides who gets named, what gets claimed, and which URLs get cited.
Unless you are watching the model, you have no idea what it is saying about you.
The four-layer framework
We score AI visibility in four layers, drawn from work by Aaron Haynes and refined through our own observations.
- L1 — Entity Establishment. Does the model recognise the brand name?
- L2 — Entity Depth. What does the model claim about you when prompted?
- L3 — Recommendation Visibility. Are you surfaced in answers to category questions?
- L4 — Informational Citation. Which of your URLs does the model cite when grounding its answer?
Each layer fails differently, and the fixes are different. L1 failures look like the model not knowing you exist. L4 failures look like a competitor’s blog post being treated as the canonical source for facts about your product. The framework is the map.
What the instrument does
BotScope runs a watchlist of queries against ChatGPT, Claude, Gemini, Perplexity, and Grok every twenty-four hours. We capture full responses, citations, latency, and metadata. We score the four layers per query, per model, and roll them up using medians so a single anomalous response can’t move the dial.
The output is a daily readout: what changed, what’s missing, where you’re being out-cited, and which gaps would close fastest with a single content fix.
Why now
The model decides who gets named, what gets claimed, and which URLs get cited.
Three things have shifted recently. Models have stopped hedging — they answer with confidence and authority. Citation grounding has become widespread, which means specific URLs win or lose. And usage is broad enough that B2B buyers, B2C shoppers, and researchers are all routinely asking models the questions they used to ask Google.
The result: an entire layer of visibility is now happening behind the model, where you can’t see it. We built BotScope so you can.
What’s next
This journal will be where we publish methodology updates, longitudinal observations, and case studies from inside the observatory. Subscribe via your favourite RSS reader at /blog/feed.xml (coming soon), or check back here regularly.
If you want to talk to us, write to [email protected]. Every email is read by a person.