Methodology
How rankyn measures.
rankyn asks the buying questions your shoppers ask, on the AI surfaces they use daily. It records every answer in full and counts how often the AI recommends your product, who it recommends instead, and which sources it cites.
From buying intent to recommendation rate
A measurement is one answer from one AI surface to one prompt. rankyn runs every confirmed prompt on every surface you measure, on each measuring day, then reads each answer the same way.
- Buying intentA buying situation you choose, such as a gentle magnesium for sleep.
- PromptsConcrete questions in shopper language and your market. Suggested by rankyn, confirmed by you.
- AI surfacesEvery prompt runs on every surface you measure, on each measuring day.
- Recorded answerThe full answer, the products named in their order, and the cited sources.
- EvaluationRecommended or not, who instead, which sources. Per surface, per week.Recommends your productRecommends a competitorNo answer: recorded, left out
Where and how often
rankyn measures a pool of 5 AI surfaces, each the way it really works: in the real interface where shoppers type, set to your market and language, or through the provider’s API with web search on. Every plan measures daily, one sample a day.
Surfaces by method
Real interfaceWhere shoppers type, set to your market
ChatGPT (App)
Google AI Overview- Google AI Mode
Gemini (App)
APIWhat apps and agents build on, web search on
Perplexity (API)- ChatGPT (API)Enterprise
Claude (API)Enterprise
One measuring week
The method is always named
Every surface carries its method, real interface or API, so you always know what an answer stands for.
One sample per measuring day
Each prompt runs once on each surface on every measuring day. Spreading the days across the week keeps one odd day from deciding the result.
What the rate counts
The recommendation rate is the share of answers on one surface in a measuring week in which the AI recommends your product, not merely mentions it. Every surface has its own rate, and every rate names who the AI recommended instead.
Recommended, not mentioned
A product that appears in an answer without being put forward does not count as recommended.
Competitors from the answers
Competitors are the brands the AI recommends in your answers. Publications the answer cites are never counted as competitors.
Never one answer
A rate is built from every answer of the week on that surface, so it never rests on a single reply.
Position, the second figure
Position is where your product stands among the products an answer recommends, per surface. It exists only where the AI recommends you and sits beside the rate, which stays the main figure.
Recommendation rate
Per surface · per product or intent · per measuring week
Answers this week
4 / 12answers recommend your product
How a source is counted
rankyn counts the sources an answer cites as the union over at least three runs. AI cites differently from one answer to the next, so an absence only counts when at least three runs back it, and sources a surface only retrieved are never counted as cited.
| Source | Run 1 | Run 2 | Run 3 | Counted |
|---|---|---|---|---|
| 1Test portaltestportal.example | ||||
| 3Health magazineRedirect link resolved to the site | ||||
| 1Reddit threadreddit.com | ||||
| 2Your domainyourstore.example | ||||
| 4Retailer listingReturned as retrieved | Not counted |
- At least three runsThe view combines the sources cited in at least three runs. AI cites differently from one answer to the next, so one run alone could miss a source or show you as absent by chance.
- Your domain, yes or norankyn shows whether your own domain is among the cited sources, and which sources name your brand or your competitor.
- The real domainRedirect links in an answer are resolved to the site behind them, so the source you see is the one the AI actually used.
- Retrieved is not citedWhere a surface returns sources it retrieved rather than cited, rankyn marks them as retrieved and never counts them as citations.
Re‑tests under the same conditions
A re‑test runs the same prompts on the same surfaces in the same market and language as the original measurement, against a before snapshot that is frozen when you start it. When something cannot be matched, rankyn labels the comparison instead of presenting it as clean.
Only your fix changes
Prompts, surfaces, market and language stay as they were, so the difference points to your change.
Labelled, never hidden
If a surface or prompt could not be run the same way, the result says so instead of passing as a clean comparison.
- Same promptsSame prompts
- Same surfacesSame surfaces
- Same market and languageSame market and language
What the measurement does not tell you
rankyn measures a defined set of buying prompts on defined surfaces, not every conversation shoppers have. These are the limits we state with every number.
A sample, not every chat
Real shopper conversations are private and can depend on their history. rankyn measures the prompts you confirm, the same way each time.
Answers vary
The same prompt can get a different answer tomorrow. That is why rankyn repeats every prompt and reads sources over at least three runs.
No search volume
rankyn shows whether AI recommends you for a prompt, not how many shoppers ask it.
Surfaces differ
An API answer can differ from the interface answer. The method is named for every surface.
The AI keeps changing
Providers update their models and the web changes. A re‑test holds rankyn’s conditions fixed, not the AI itself.
Missing is not losing
An attempt that fails or times out is recorded as failed and left out, never read as a lost recommendation.
See the method on your store.
Start with a free audit measured on real AI surfaces, or compare the plans to see how much each one measures.
Questions about the method
Surfaces, cadence, the recommendation rate, sources and re‑tests in short. Can’t find your answer? Write to hello@rankyn.ai.
How does rankyn measure AI recommendations?
rankyn runs the buying prompts you confirm on the AI surfaces you choose daily, and records every answer in full. It then counts how often the AI recommends your product on each surface, who it recommends instead, and which sources it cites.
What is the recommendation rate?
The recommendation rate is the share of answers on one surface in a measuring week in which the AI recommends your product, not merely mentions it. Each surface has its own rate, and each rate names the competitors recommended instead.
How often does rankyn measure?
rankyn measures the Growth plan daily, one sample a day. The Growth plan runs 5,250 measurements a week: 150 prompts on 5 surfaces on each measuring day.
Which AI surfaces does rankyn measure, and how?
rankyn measures a pool of 5 surfaces: ChatGPT (App), Google AI Overview, Google AI Mode, Gemini (App), Perplexity (API). Enterprise adds ChatGPT (API) and Claude (API). Each is measured in the real interface, set to your market and language, or through the provider’s API with web search on, and the method is named for every surface.
How does rankyn count sources?
rankyn combines the sources cited across at least three runs and checks whether your own domain is among them. Redirect links are resolved to the site behind them, and sources a surface only retrieved are marked as retrieved and never counted as cited.
What makes a re‑test comparable?
A re‑test runs the same prompts on the same surfaces in the same market and language as the original measurement, against a frozen before snapshot. When something cannot be matched, rankyn labels the comparison instead of presenting it as clean.
What does rankyn not measure?
rankyn does not see private shopper conversations and does not report how often a prompt is asked. It measures a defined set of prompts the same way each time, and failed attempts are left out rather than counted as losses.