A band rather than a number, whole weeks only, and nothing at all about anybody else. What it reports, measured rather than assumed — and the two readings that call for opposite actions.
Impression share answers one question: of all the times somebody searched a term and an ad appeared, how often was the ad yours? It is the only figure Apple publishes that is about the auction rather than about your campaign, and it is the difference between "nobody is searching this" and "plenty are, and you are losing".
It is also hedged, delayed and narrower than most people expect. Everything below was measured against the live API rather than read in a blog post, because Apple's own documentation is thin and the gaps are where the mistakes live.
A share arrives as a low and a high. Under one per cent it says so rather than giving you a figure. Above ninety it stops being specific and says 91–100%. In between you usually get a single value. If you see a range, that is Apple's precision and not a rounding by whoever built your dashboard — printing a midpoint would be inventing the digit Apple deliberately withheld.
There is no daily impression share. The only granularity Apple reports is a full week running Sunday to Saturday, which means today's searches are not in it and the current week appears once it has finished. A tool showing you a daily impression share line is showing you something Apple did not give it.
The report is keyed by app, storefront and search term. It is not keyed by campaign: ask Apple to filter by campaign and it answers "Field 'campaignId' is not a recognized field". That is not an oversight in the API. The share belongs to the app — if two of your campaigns bid on the same word in the same storefront, they produce one row between them, because to the person searching there was one ad.
Apple withholds any term with fewer than ten impressions in the week, and reports nothing at all for a campaign running in more than one country. On one live account it reported thirty-six of forty-two terms, and one of the six it skipped had over a thousand impressions. A missing row does not mean you got none of that search.
You cannot see anybody else's. Ask the report for a competitor's app and it returns zero rows — tested with two competitor identifiers in a storefront where our own app returned eight. Nobody outside Apple can see who bid, what they paid, or what share they took.
Tools that sell "auction insights" are showing you something else, and the honest version of it is worth having: the App Store search results for that word. That is who somebody typing it actually chooses between, whether or not any of them bought an ad. It is a different fact from the auction, and a screen that blurs the two is lying to you gently.
If a term converts for you and you are taking a fifth of it, the rest is available and somebody else is buying it. This is the clearest buy signal Apple gives.
Taking ninety per cent of a search that rarely installs means you are winning an argument you should not be having. Either the term is wrong for your app, or your App Store page is losing people who were interested enough to tap.
Each row carries your position among the advertisers bidding on that search. Note what Apple does not say: how many advertisers there are. "#2" means one advertiser took more of that search than you; it does not mean second of a known field, and a dashboard presenting it as a score out of something has added the something itself.
Read it weekly, per term, alongside what that term converts at. A rising cost per install with a falling share is the market getting more expensive. A rising cost with a rising share is you outbidding yourself. They look identical on a spend chart and they call for opposite actions.