1. Source state
What public content and crawlable evidence existed at the time?
Controlled early access
AI visibility evidence without hiding the result inside one score.
Visibility Evidence records the exact buyer prompt, whether an organisation appeared, what alternatives appeared instead, which URLs were cited, and whether the same result reproduces across repeated observations.
Evidence chain
Measure the chain instead of averaging it away.
2. Interpretation
What claims and description did evaluated models extract?
3. Buyer intent
For a fixed buyer prompt, was the target included or omitted and who appeared instead?
4. Citation environment
Which owned, competitor, directory, media or third-party sources shaped the answer?
Repeated self-test · 22 August 2026
The first fixed buyer prompt reproduced the same target outcome twice.
“Which companies provide AI visibility and interpretation monitoring software solutions for organisations like mine?”
SemanticRisk ran that exact category-discovery prompt twice on the OpenAI Responses API web-search surface. SemanticRisk was not surfaced in either independent repetition.
The supporting evidence was not identical: the first observation contained 15 cited URLs and the second contained 10. Neither observation included a SemanticRisk-owned citation. This is the distinction the product is designed to preserve: the target outcome can remain stable while the supporting citation environment changes.
Target
semanticrisk.io
semanticrisk.io
Repeatability
Absent 2 / 2
Absent 2 / 2
Citation counts
15 → 10
15 → 10
Owned citations
0 → 0
0 → 0
This is a controlled self-test, not a market benchmark. Broader prompt coverage is still being executed before category-level conclusions are published.
What a pilot will show
Evidence an agency or client can actually challenge.
- Fixed, versioned buyer-prompt battery
- Exact prompt disclosure
- Independent repetitions
- Target inclusion / absence
- Recommended alternatives and competitor displacement
- Provider-native citation URLs and source classification
- Repeatability and citation-host overlap
- Connection to SemanticRisk interpretation-drift evidence
Method boundary
We preserve the surface being measured.
This early implementation measures the OpenAI Responses API with web search. It is not presented as equivalent to the consumer ChatGPT interface. Model, prompt version, repetition, timestamp and evidence are retained so results can be audited longitudinally.
The product remains in controlled early access while we validate broader prompt stability, structured competitor extraction and citation classification.
Early-access pilot
Bring us one buyer question you need answered.
For agencies, communications teams and organisations that want to know not only whether they appear, but who appears instead and what evidence is shaping the result.