What the guides printimpressions over eligible impressions
impressions ÷ eligible impressions
All eleven print it and so does Google. The eligible count is never shown to you, so the formula only runs the other way round.
What Google returnsany share under 10% reads 0.0999
below 10% → 0.0999
A campaign under the line looks bigger than it is, so a total rebuilt from campaigns comes out high. None of the eleven guides mentions the floor.
Google reports the share, not the eligible count. For the account or a date range it gives you the combined figure itself. For a grouping of your own you run the formula backwards, and the 10 per cent floor gets in the way.
What I checked
I opened thirteen guides on 21 September 2026. Six are the advertiser guides on Google's first page for search impression share formula, as I listed them when choosing the keyword; I did not save that results page. Keyword Planner puts that phrase at 1,600 US searches a month, against 110 for the next one, impression share. Seven more came from a second search engine for that phrase and for google ads impression share. WordStream refused an automated fetch, and so did Search Engine Land's piece on the Target Impression Share bid strategy, so eleven were read, from ten publishers: MetricHQ belongs to Klipfolio. I set six questions before reading and added three during it. Three of the first six carry the article:
The first row is no fault: an example needs a number. Several guides say plainly that Google estimates the count. Growth Minded Marketing writes "Google Ads estimates the total impressions a keyword or search term was eligible to receive", PPCexpo divides by "the estimated number of impressions you were entitled to receive", and HubSpot quotes Google's "Eligible impressions are estimated using many factors". HubSpot and Growth Minded Marketing also turn the formula round to get the count back. PPCexpo warns that shares are "not additive", and MetricHQ writes the aggregate as a sum over a sum. None of the eleven says what the floor does to any of that.
Where the eligible count is, and where it is not
The Ads API has a service that lists every field it can return. Querying it takes API credentials, but it describes fields and touches no account. I asked it for anything with eligible in the name. Two of the results are metrics, one for Hotel campaigns and one for store reach from location assets. Search has no eligible impressions metric. The fourteen fields in the impression share family are all ratios, and Google's list of columns for custom column formulas has no eligible count either. Google's API team said as much in 2020: "It isn't possible to pull the number of potential impressions directly".
Three guides write as if you are handed the count. HubSpot's example, a few lines under its Google quote, opens "Let's say we created an ad and Google says there are 5,000 potential impressions available." Count says platforms "calculate this behind the scenes" and that the figure "is usually provided in your campaign reports as" a column called Eligible Impressions, "or used automatically in impression share calculations". Cometly makes finding it step two of its method.
What Google does give you is the combined share. Its field reference lists search impression share, and both lost columns, as selectable for the customer, which is the whole account, and the interface shows one figure for the date range you pick. Google works those out from its own estimates rather than from the rows you see, so the account figure sits at the floor only when the account itself is under 10 per cent. The two can disagree: in 2021 a developer found an account total higher than the campaign rows allowed, and Google's API team wrote "I replicated your issue in the UI". For the account, or a date range, take Google's figure.
For a grouping Google does not report, the formula runs one way:
Eligible impressions = impressions ÷ impression share.
Back out each row's eligible impressions, total both columns and divide. Google's API team suggested the same method in 2020, saying it "should give a good estimate assuming all of your impressions are on the search network", and described the result: "You should think of it as a weighted average." Averaging the shares goes wrong even with no floor involved, because a campaign with ten times the eligible volume counts the same as a small one.
Google floors impression share at 10 per cent
The API's field reference, on search impression share: "Any value below 0.1 is reported as 0.0999." The lost columns have the matching ceiling: "Any value above 0.9 is reported as 0.9001." All fourteen fields in the family carry one or the other, six floors and eight ceilings, and the wording is the same in v24 and v25. Google's help page on custom columns gives the rule for formulas you build in the interface: the share column "will be 0.0999 or empty". In 2017, when the old API returned the floor as text, the AdWords API team wrote that "both the web interface and the API will have same special value" and that "there is no workaround to get the actual percentage via API in those cases". I found no current Google page describing what the interface table shows.
A reading of 0.0999 means anywhere from zero to ten per cent. Three consequences follow, and none of the eleven guides mentions them:
- A chart goes flat where it matters. A campaign sliding from 9 per cent to 2 per cent draws a level line at 9.99 from API data. Google's own Account Summary script, which writes daily account figures into a spreadsheet for charting, turns any value at or below 0.0999 into the text <10% first, so those days reach any chart built on that column as words.
- A spreadsheet average drops the worst days. The Google Sheets page for AVERAGE says "Any text encountered in the value arguments will be ignored." Average a column holding <10% and the rows under the floor leave the calculation, which pushes the result up.
- A rebuilt total overshoots. Divide a campaign's impressions by 0.0999 when its real share was 2 per cent, and you credit it with a fifth of the eligible impressions it had. The group's share rises with it.
None of this is news to people who build on the API. Developers raised the floor on Google's API forum in 2017, 2018 and 2019. My claim is narrower: none of the eleven guides I read passes it on.
Impression share for a custom group, four ways
The numbers below are invented to show the arithmetic, and they are the only campaign figures in this article. Say A, B and C are three campaigns under one label, a grouping Google does not report. Google keeps its estimate of each share; the next column is what you see.
| Impr. | Google's estimate | What you see | Eligible, backed out | |
|---|---|---|---|---|
| A | 9,000 | 60% | 60% | 15,000 |
| B | 2,000 | 40% | 40% | 5,000 |
| C | 400 | 2% | < 10% (0.0999) | 4,004 |
Campaign C was eligible for 20,000 impressions. Backing out from 0.0999 credits it with 4,004. The second column below moves C to 8 per cent, eligible for 5,000; what you see stays the same.
C at 2 per centMethodC at 8 per cent
28.5%
Impressions over the three campaigns' eligible estimates
45.6%
36.7%
Plain average of the shares you see
36.7%
47.5%
Impressions over eligible backed out at 0.0999
47.5%
50.0%
Spreadsheet AVERAGE with <10% stored as text
50.0%
With C at 2 per cent it holds half the group's eligible volume, and the rebuilt figure lands further from the estimate than the plain average does. With C at 8 per cent the rebuilt figure is about 2 points high and the plain average about 9 points low. The direction depends on how much volume sits under the floor. The rebuilt figure cannot fall below the estimate, so with one floored row it is an upper limit, 47.5 per cent at most here, and there is no lower one. The first row is the sum of the three campaign estimates, the most you can rebuild; Google's own figure for a grouping it does report can differ from such a sum, as the 2021 case shows.
Lost impression share, and whether the columns add to 100
DashThis says share and lost share "account for a total of 100%". Trustworthy Digital's third bidding scenario, "When Budget and Rank Are Both Limiting You", wants lost to budget and lost to rank both above 50 per cent, which passes 100 before you count the impressions the campaign did get. None of the Google pages I saved says the three columns add up. The closest is a footnote on its page about top metrics, attached to a different formula: "Eligible impressions are the number of impressions you would receive if you weren't budget-constrained, and if you won all of the auctions you participated in." Read that way, budget and rank are the two ways to miss, and the columns share one total. That is my reading, and I have not tested it on account data.
You can test it on your own account. At campaign level, where Google shows lost to budget, take rows with a share above 10 per cent and add the three columns. If they come to 100, you can try a way under the floor. Say campaign C above shows lost to budget 30 per cent and lost to rank 68 per cent: its three reported columns add to 107.99, which marks it as floored, and 100 minus 30 minus 68 gives back its 2 per cent. This is my workaround, not Google's. It rests on the untested reading above, the lost columns are rounded too, and if either reads 90.01 it tells you nothing.
Three guides also put budget among the things that set eligibility: Klipfolio, Count and MetricHQ. Google says eligible impressions are estimated from "many factors, including targeting settings, approval statuses, and quality", with no budget among them, and the footnote above describes them as what you would get "if you weren't budget-constrained". If budget shrank the denominator, lost to budget would have nothing left to measure.
Three things to do
Ask Google for the total first
For the account, or for a date range, use the share Google reports at that level rather than rebuilding it. Rebuild only for groupings Google does not report, within one campaign type, and with impressions on Google Search alone, because Search partners "won't be included in impression share data".
Label a floored total as a ceiling
If any row in a rebuilt total reads 0.0999, print the result as a maximum, at most 47.5 per cent in the example. Or recover the floored rows from the lost columns first, once your own rows show the three adding to 100.
Draw a floored point as a band
On a chart, show anything at the floor as a shaded band under 10 per cent or as a marker. A level line at 9.99 reads as stable while the campaign behind it may be falling.
Where this sits in our own tool
Google Ads Vitals does not report impression share. It reads a campaign export broken down by date and maps impressions, clicks, cost, conversions and conversion value. With plain Impr. and Clicks columns in the export, any share columns are ignored. Without them, a looser second pass takes a header such as Impr. (Top) % or Impression share as impressions, and Click share as clicks, and every rate built on them is wrong. The load message lists the headers it matched without saying what each was taken for. I confirmed this by running the file's own matcher, and it is mine to fix.
Our auction insights article, which sets your impression share beside other advertisers', recommends the same rebuild for an account figure. For the account, Google's own figure is the better source, and any rebuild needs the note about the floor. That piece also says the three columns "add up to a complete account of your own eligibility", which on the evidence above is a reading rather than a documented rule.
Related reading
- The Quality Score discount table is one division, another Google number read as more than it is
- Performance Max reports arrived without their labels, on rates you cannot compare across campaign types
- The search terms report that does not add up
Questions people actually ask
Impressions divided by eligible impressions, which is how Google writes it on its help page About impression share. Google estimates the eligible number and does not report it. For the account or a date range, Google reports the combined share itself; for a grouping of your own you use the formula the other way round: eligible impressions equal impressions divided by the share Google shows you.
If they make up the account, or you want a date range, take the share Google reports at that level. For a grouping Google does not report, back out each campaign's eligible impressions by dividing its impressions by its share, add those up, and divide total impressions by that total. Do not average the shares. If any campaign in the group reads below 10 per cent, the result is an upper limit rather than a value.
Google reports any share below 10 per cent as 0.0999 in the API, and as 0.0999 or empty in custom columns. Lost impression share above 90 per cent comes back as 0.9001. In 2017 Google's API team said the web interface showed the same special value, <10%, and that "there is no workaround to get the actual percentage via API in those cases". I found no reason for the cut in Google's documentation.
The Google pages I read do not say so, though one footnote defines eligible impressions in a way that implies it. Test it on your own campaigns: take rows above 10 per cent and add the share, lost to budget and lost to rank. A row at the floor will not add to 100 even if the rule holds, because its share is reported too high.
Google's help page says eligible impressions are estimated from "many factors, including targeting settings, approval statuses, and quality". Budget is not named, and a footnote on another Google page describes eligible impressions as what you would get "if you weren't budget-constrained". What a low budget costs you shows up in lost impression share (budget). Three of the eleven guides I read put budget into the denominator, which would leave that column with nothing to measure.
It can drop without anything getting worse. Google estimates the eligible set, and its help page says "Changes to your bids, quality, or Google's ad systems may change the set of auctions in which the system estimates you were competitive." Read impressions next to the share before you act on a drop.
WRITTEN BY
Olha · clinic data analyst
I build the reporting our managers open every morning at a multi-branch medical clinic — and package it so other practices don't have to start from scratch.
Published on 22 September 2026; the pages were read on 21 September 2026. Three limits worth stating outside the body text. The sample is eleven guides from one day, and two publishers refused the fetch, so rerun the method rather than trust the counts. The worked example is invented, and the size and even the direction of the error depend on how much eligible volume sits under the floor. And whether the three columns add to 100 is a reading here, untested on purpose, because testing it would mean publishing an account's numbers.