Answers
Why N/A Is Not Zero: Reading Missing Measurements
An N/A in a visibility table is not a zero. It means the measurement is unavailable. Treating it as zero drags down totals and averages and makes a brand look less visible than the data can support.
Picture exporting a brand's visibility table and finding a row marked N/A. Treat that row as zero and the total drops and the average falls: the brand looks less visible than the available data can support. The rule of thumb is simple: if a row is N/A, the total and the average do not describe the brand. N/A means the measurement is unavailable, not that the observed value is null. When a visibility measurement is not there, RankAI shows it as N/A, never as zero. Zero is information: it says something was observed and produced no results. N/A says something else: that measurement was not collected or is not available for that row.
What changes between absence and zero
The difference is not cosmetic. A zero enters calculations as a value; an N/A does not. Add rows with N/A as if they were zero and you get a total that no observation supports. Calculate an average with zeros in place of missing measurements and the result falls with no data to justify it. So the first step is to separate the N/A rows from the rest and ask whether the total you are about to publish includes only measurements that are actually available.
A checklist, offered by RankAI as a working proposal
The checklist below is a working proposal from RankAI, not a universal standard. It is there to help you decide what to do with a number before you publish it.
The questions to ask
- Is the row N/A, or does it contain an observed zero?
- Does the total I am calculating include only rows with an available measurement?
- Is the average calculated on the number of available measurements or on the number of rows in the table?
- If I publish the total, can I say which rows are excluded and why?
- Does the N/A row concern an engine or a market that this view is actually observing?
What to do when the checklist stays open
If any question has no answer, do not publish the number as a total. Report it as "not calculable" or publish it alongside the list of excluded rows. In practice: do not replace N/A with zero to make the sum add up, and do not present as complete a total that rests on partial rows.
The delicate point is that an N/A is not an error to correct but information to keep. In a visibility table, the temptation is to turn every cell into a number so you can sum and compare. But a total built on artificially filled cells is no longer comparable with a total built on real measurements: you are mixing two kinds of data. RankAI's proposal is to treat N/A as a label that stays visible next to the number, so the reader knows the total excludes something. This holds even when the N/A concerns a single row out of many: that is exactly the case where it is easiest to miss. The checklist above is meant to make that exclusion explicit before publication, not after. It is not a guarantee that the product does or does not attribute a row to an engine or a market: it is only a reminder for whoever reads the number and has to decide whether to use it.
A concrete case: the page in simplified Chinese
Suppose you see a page in simplified Chinese among the results and conclude that RankAI is measuring 豆包, 元宝, Kimi or 千问. It is not: a page in simplified Chinese does not mean RankAI measures 豆包, 元宝, Kimi or 千问. The language of the page is not proof of the engine being observed. If the table shows an N/A for that row, the correct reading is that the measurement is unavailable, not that the engine was measured and returned zero. Here too the rule holds: absence does not convert into a value.
The limit to keep in mind
This approach applies to the measurements RankAI shows as N/A: we cannot say anything about how other tools handle missing measurements, or about engines that fall outside what is observed. During pre-launch, checkout can stay off: the invitation is to create an account or open the relevant page, not an immediate trial. For the full picture on how to read unavailable measurements, see the in-depth page.