What is Impression share?

A search campaign records 84,000 impressions and the platform estimates it was eligible for 240,000. Impression share is 84,000 ÷ 240,000 = 35%. The lost columns split the missing 65%: if 40 points were lost to budget and 25 to rank, the largest single available gain is a budget increase, and doubling the bid would compete for a smaller share of the shortfall than adding money would.

Almost every advertising metric describes what happened: impressions delivered, clicks received, conversions produced. Impression share is unusual because it describes what did not happen.

It divides the impressions an account received by the impressions the platform estimates it was eligible to receive. The column is a search-advertising one: Google Ads and Microsoft Advertising both report it, alongside the two lost columns and the top-of-page variants. A share of 35% says the account won roughly a third of the auctions it could have entered and missed two thirds.

That is information no performance metric contains. A campaign can have an excellent cost per acquisition and be capturing a small fraction of the demand available to it, and nothing in the cost metrics will say so. This is the number that does.

The numerator is counted directly. The denominator is not.

The platform models how many auctions the account was eligible for, based on its targeting, its budget, whether its ads were approved and serving, and the query or audience volume it observed. That model is proprietary, it involves inference, and its output carries an error band.

The practical consequence is to treat impression share as directionally reliable and precisely unreliable. A move from 30% to 55% is meaningful. A move from 52% to 54% is not, and building a report that tracks it to a decimal place gives spurious precision to a modelled figure.

The headline share says how much was missed. The lost columns say why, and they are the part that produces a decision.

Lost to budget counts eligible impressions the account did not receive because it ran out of money, either because the daily budget was exhausted or because pacing limited delivery. The remedy is money, and the size of the column is a direct estimate of how much more volume is available at roughly current efficiency.

Lost to rank counts eligible impressions lost because the ad was not competitive enough to be shown. Rank combines the bid with the platform's quality estimate, so the remedies are a higher bid, better expected click-through, more relevant ad copy, or a better landing experience.

A campaign losing forty points to budget and twenty-five to rank has a spending problem. One losing five to budget and sixty to rank has a competitiveness problem, and adding budget to it will change almost nothing. Reading the headline figure without the split leads to whichever intervention the reader was already inclined toward.

The denominator is the volatile half, and it moves for reasons that have nothing to do with the account.

A competitor entering the market raises the number of auctions and lowers your share without changing your delivery at all. A seasonal spike in query volume does the same. So does anything on your side that widens eligibility: adding keywords, loosening match types, expanding geography or extending schedules all grow the pool you are being measured against.

This makes period-to-period comparisons treacherous unless targeting has been stable. An account that broadened its keyword set and saw impression share fall has not got worse; it has started measuring itself against a larger opportunity.

The discipline is to compare impression share only across periods with consistent targeting, and to note structural changes in the report so the next reader does not misread the step.

A high share is not automatically good, because the eligible pool is defined by the account's own targeting rather than by what is worth buying.

An account with broad match types and no negative keywords will be eligible for a great deal of poorly qualified traffic. Winning a large share of it is expensive and unproductive. The correct response to a low share in that situation is to narrow the eligibility, which paradoxically raises the share by shrinking the denominator — and genuinely improves the account.

Conversely, an account with a tightly qualified keyword set and a low share is leaving real money on the table, and every missing point is demand going to a competitor.

The metric cannot distinguish these cases. Reading it against the conversion rate and value of the eligible traffic is what does.

Alongside the overall figure, platforms report shares for the top of the page and for the absolute top position, plus the equivalent lost columns.

These answer a different question: not whether the ad appeared, but where. Position matters a great deal on search, where the first result captures a disproportionate share of clicks, and it matters for brand defence, where appearing below a competitor on your own name is a specific and visible problem.

Using the position variants for position questions, rather than inferring position from the overall share, avoids a common confusion — an account can hold a high overall share while rarely appearing at the top, which is a completely different situation from a low overall share.

Full coverage is rarely the right goal. The final increments of any competitive auction set are the most expensive, because winning them means beating everyone in the hardest cases, and the marginal return on that money is usually lower than its return elsewhere.

The defensible pattern is to set targets by segment rather than for the account: near-full coverage on exact brand terms where the cost is low and the defence matters, a deliberate mid-range target on the core commercial terms, and no target at all on exploratory or broad segments where the share is a byproduct rather than an objective.

Underneath all of it sits the same ceiling as everywhere else: impressions are worth winning up to the point where the conversions they produce still cost less than the contribution they generate.

• Read the two lost columns before acting on the headline share. They say whether the constraint is money or competitiveness, and the remedies are different.
• Treat it as an estimate with an error band, not a count. The denominator is modelled from auctions the platform believes you were eligible for.
• Compare it only within a consistent targeting definition. Broadening keywords or geography grows the denominator and lowers the share without anything getting worse.
• Chase it only where the eligible traffic is worth winning. A high share of a poorly qualified query set is an expensive achievement.
• Use the absolute top and top-of-page variants for position questions rather than inferring position from the overall figure.