The answer and its citations
The full answer text, every URL the engine cited, the model identifier as the run recorded it, the pass number, and the search-grounding status.
How it works / The measurement method
We don’t guess what AI says. We measure it.
Rankovi.ai asks the questions a buyer would ask, requires each engine to search the web before answering, and stores every answer in full with the sources it cited. The result is evidence Rankovi Labs can act on — and show you.
01 / The run
A defined question set, repeated, across four engines.
A measurement run is one question set, one set of engines and settings, and one measurement period. Everything reported traces back to the individual answers in that run.
QuestionsA documented set of buyer questions for the business — some that name the brand, most that don’t — covering comparison, provider search, and problem-aware questions. The set is kept fixed so a later run can be compared.
Engines and pathwaysChatGPT through the OpenAI API with web search; Claude through the Anthropic API with web search; Gemini through the Google API with Search grounding; Perplexity through the Sonar API. These are the measured pathways — an API run does not reproduce every consumer-app experience, and Gemini API grounding is not a direct measurement of Google AI Overviews.
PassesEach question is asked three times per engine. A mention rate is the share of answers that named the business. Three passes are three repeated observations, not three customers.
LocationThe buyer’s location is set where an engine’s API supports it and recorded as unavailable where it doesn’t. In the October 2026 SWFL run, Gemini had no explicit location-setting mechanism recorded.
GroundingEach answer carries the engine’s recorded search-grounding status. An answer recorded as ungrounded is reported separately from the valid-answer count rather than treated as a result.
FailuresA failed query is recorded as failed and re-run for the same run, prompt, engine, and pass. It is never counted as “not mentioned.” Valid-answer coverage is reported alongside every rate.
02 / What is recorded
Every answer, in full, with its sources.
Each figure in a report traces to the recorded answers and sources behind it. Nothing is summarized away before a person can read it.
Named, cited, and who else appeared
Whether the business was named, whether its website was cited, and which other businesses the answer named. A name and a citation are recorded separately; neither is treated as “recommended first.”
Rates with their denominators
Mention and citation rates per question, per engine, and overall — always with the number of valid answers they are drawn from, and with branded and nonbranded questions kept apart.
03 / What it helps decide
Observed. Interpreted. Then acted on.
Rankovi.ai informs the strategy. Rankovi Labs does the work — and keeps the three apart in every report.
Observed: what the returned answers and citation lists contained — which questions surfaced the business, which sources supported those answers, which alternatives appeared, and where results were inconsistent across engines or passes.
Interpretation: what the pattern might suggest — for example, that a provider-search question for one city is answered inconsistently while a delivery question for the same city is not. These are hypotheses, stated as such.
Recommended action: what Labs proposes to investigate, improve, and measure next. After the work ships, the same question set is run again under the same documented settings, and any change in the question set, models, or valid-answer coverage is disclosed rather than smoothed into a trend.
Where a client has them, Labs reads these observations alongside Google Search Console, website analytics, and conversion data. Rankovi.ai does not collect those sources itself; they stay separate, and no single growth score is manufactured from them.
04 / What clients receive today
Delivered as reports inside a Rankovi Labs engagement.
Rankovi.ai is not sold as a separate subscription, and there are no plans or tiers to choose from. Measurement is scoped as part of the engagement it serves.
Clients receive a written report: what we learned and why it matters, the measurement summary with its date, scope, engines, and valid-answer coverage, the question-by-question detail with sources and the businesses named alongside yours, and the methodology and limits. A baseline report labels its recommendations as proposed next steps; a later report records the work completed and compares like with like.
How often measurement repeats, and how results are shared, depends on the engagement. A one-time build or strategy project typically ends with a baseline and a recommended remeasurement; an ongoing growth engagement remeasures after meaningful changes ship.
The client-facing console is not a standard deliverable today.
What the method does not claim
- That every customer sees these answers, or that an API run reproduces a consumer app.
- That a mention is a recommendation, or that a citation is trust.
- Exactly why an engine omitted a business, or that a change in the answers was caused by our work.
- Promised citations or rankings, or revenue inferred from missing mentions.
- Competitor leaderboards built by counting every name variant as a separate business.
See it on a real business
One baseline, fully disclosed.
The SWFL Health & Hydration baseline from October 8, 2026 is published with its denominators, its strongest and weakest questions, and the limits of what it establishes.