Method
After You Pick a Channel: What to Actually Check
Picking a channel is only the first step. Before you read any visibility number, confirm the measurement exists and know what kind of observation you are looking at.
So you have picked the channel you want to watch — the engine or surface you plan to observe, say Gemini or Google AI Overviews. Now two things need checking before you read anything into the numbers. First, does a visibility measurement exist for that channel at all? Second, what kind of observation are you looking at? Knowing the channel is live is not enough. You have to confirm the measurement is there and that the observation matches the surface you chose. In RankAI, when a visibility measurement is missing, we show it as N/A, never as zero. That is the first check after you pick a channel, and it comes before any reading of the data.
Why the kind of observation matters as much as the measurement itself
A channel is not a generic label. It is the concrete surface you decide to watch. Choose Gemini and you are observing one thing; choose Google AI Overviews and you are observing another. They are not the same observation and should not be treated as equivalent. The same holds for how the data arrives: what comes back from an engine through an API is the API response, not the screen a person sees. Keep both limits in mind when you fill in the check card. They tell you what you can legitimately compare and what you cannot.
The post-selection check card: a RankAI working proposal
The card below is a RankAI working proposal, not a universal standard. Its job is to make one rule immediately usable: compare only like with like. Two measurements can go side by side only if they come from the same kind of observation and neither one is N/A.
- Name the channel you chose. Write the exact name of the surface you are observing (e.g. Gemini, Google AI Overviews). Test: the field cannot hold a generic label like “AI engine”; it has to name a specific surface.
- Record whether the measurement exists. Enter the returned value, or N/A if the measurement is not there. Test: the field can never hold “0” in place of a missing measurement; if the measurement is missing, the card shows N/A.
- Note the kind of observation. State whether you are reading an API response or another kind of observation. Test: two rows on the card are comparable only if the kind of observation matches and neither row is N/A.
A filled-in example
Here is a concrete example so the card is reusable. Row A: channel = Gemini; measurement present = N/A; kind of observation = API response. Row B: channel = Google AI Overviews; measurement present = returned value; kind of observation = API response. These two rows are not comparable, because one of the measurements is N/A. Row C: channel = Gemini; measurement present = returned value; kind of observation = API response. Row C is comparable with row B only if the kind of observation matches and neither is N/A; it is not comparable with row A, because A shows N/A. The example shows how the card blocks bad comparisons before they turn into a distorted reading.
What this piece adds is the check card as an artifact: not an abstract rule, but a three-field form that forces you to declare the channel, whether the measurement exists, and the kind of observation before any comparison. Fill it in right after you pick a channel, while your intent is still fresh and the risk of confusing one surface with another is highest. Its value is that it makes the implicit visible. If a measurement is missing, the card says N/A and stops you from treating it as zero. If two rows have different kinds of observation, the card stops you from putting them on the same footing. The card does not add properties to the product. It turns the limits already known about APIs, Gemini and Google AI Overviews into a reusable format. Whoever fills it in knows what can be compared and what cannot, and can keep it as a record of the check made after choosing the channel.
A limit you can verify
The card does not say whether visibility is good or bad. It only says whether the measurement exists, how it was observed, and whether two rows are comparable. If a measurement is missing, the card shows N/A and produces no substitute value. That is the limit you can check row by row.
For the full picture of how the product works and what checking after you pick a channel means, see the in-depth page.