Evidence and measurement

How VStok Measures AI Recommendations

VStok treats AI answers as time-stamped samples—not permanent rankings. Every useful result records the question, market, execution mode, evidence and failures needed to interpret it.

Version 1.0VStok ResearchReviewed by VStok Editorial

The measurement contract

A VStok baseline is a defined set of buyer questions executed for a stated brand, category, country and language. Results report whether the brand appeared, which competitors appeared, which sources were visible and whether each sample completed. The score summarizes that contract; it is not a universal ranking across every possible user, model or conversation.

Fields required for an interpretable baseline
FieldRecorded valueWhy it matters
PromptExact buyer question and intentPrevents moving the goalposts
MarketCountry and languageControls regional recommendation context
ExecutionProvider API or simulationSeparates observed and fallback data
SampleModel, date and completion stateExposes variation and failures
EvidenceURLs, domains and ownership classMakes findings inspectable

Provider execution and failures

Provider API

The answer was returned by the named provider during the recorded measurement window. VStok preserves the provider/model label exposed by the integration.

A provider response can still be ungrounded. Grounding and citation presence are recorded separately from execution mode.

Simulation

Simulation keeps the product demonstrable when live provider execution is not available. It is explicitly labelled and must not be presented as observed market evidence.

Research datasets, external benchmarks and before/after claims require live provider samples.

Partial failures remain visible in the denominator and evidence summary. VStok does not silently replace a failed provider sample with simulation or report a complete benchmark when every live sample failed.

How citation sources are classified

Owned

A domain controlled by the measured brand.

Third-party

A source independent from the measured and compared vendors.

Competitor-owned

A domain controlled by another vendor in the comparison.

Unknown

Ownership cannot be established from the available public evidence.

Citation does not imply endorsement. A source may support a positive recommendation, a limitation or a neutral product fact. VStok therefore keeps source ownership, recommendation presence and sentiment as separate observations.

Comparable re-audits

To test whether a content, positioning or technical change worked, VStok reuses the baseline prompts, market and model scope. The changed URL and deployment date should be recorded, and the page should be crawlable before the repeat measurement begins.

  1. 1Save the original prompt set and completed/failed sample counts.
  2. 2Record one bounded intervention and the URLs it changed.
  3. 3Allow the new evidence to become publicly fetchable and, where relevant, indexed.
  4. 4Repeat the unchanged measurement contract.
  5. 5Classify the outcome as improvement, no material change or regression.

Claims this methodology does not make

  • • One answer represents all AI users or future answers.
  • • A citation proves endorsement or caused a recommendation.
  • • llms.txt or special schema guarantees inclusion in an AI answer.
  • • Simulation can substitute for live evidence in market research.
  • • A higher aggregate score proves that one specific content change caused it.

Frequently asked questions

Does one AI answer represent a brand's visibility?

No. VStok treats each answer as a sample. A useful baseline keeps prompts, market, model scope and repetition rules visible so that patterns can be separated from normal answer variation.

Does a citation mean that an AI system endorses the source?

No. A citation can support a recommendation, limitation, factual detail or warning. VStok records citation presence and source ownership separately from recommendation sentiment.

Can simulated answers be used as market research?

No. Simulation is a product fallback and is labelled as such. VStok research and comparable provider benchmarks require live provider API evidence.

Primary references