Say you have two darts. You give me money for throwing them and win the competition. I will throw them for you. The first: I throw at the board eyes closed. It hits, but pretty random. I throw the second but never tell you where it hit. Wanna play again?

This is basically the story of search terms.

Open the search terms report for any non-brand search campaign, add up the cost of every query listed, and compare it to the campaign spend. I have seen campaigns where listed queries account for roughly half. The other half is a single row at the bottom labelled “Other search terms”, and Google will not tell you what is in it.

Since September 2020, when Google decided that search queries which did not reach an undisclosed volume threshold would no longer be shown to the advertiser who paid for them. The stated reason was privacy (no joke). The practical effect is that a business buying clicks on Google can see what it bought for the part of the budget that is cheap and repetitive, and cannot see what it bought for the part that is rare and, as I will argue, disproportionately expensive and irrelevant.

The mechanism that decides which queries you may see is described, when Google describes it at all, in terms of statistical significance. That framing is doing a lot of work. Significance is a property of estimates. Spend is not an estimate. A click that cost €18 cost €18 whether the query was searched once or a thousand times, and the advertiser’s interest in knowing whether that click was for “sales management software” or “sales manager at software company” does not diminish with rarity. If anything, rarity is where the waste concentrates, and the only queries you are allowed to clean up are the common ones.

What did Google change in 2020, and how much did it hide?

Until September 2020 the search terms report did what its name says. Every query that triggered an ad and received a click appeared as a row, with cost, clicks and conversions. It was the single most used report in the product, because it is where negative keywords come from and where new keywords are discovered.

On 1 September 2020 a notice appeared in every account: from now on the report would only include terms that a significant number of users searched for, even if a term received a click. Google’s statement to Search Engine Land explained the change as being made to maintain our standards of privacy and strengthen our protections around user data. The word “significant” was not defined then and has not been defined since.

The industry measured the damage within weeks:

  • Seer Interactive, 5.1M data points, 30+ businesses — 28% of paid search budget and 20% of clicks lost visibility
  • Metric Theory, client portfolio — 19% of clicks and 31% of cost missing from the report
  • AgencyPPC, standard search campaigns — Visible queries fell from 75–85% of clicks to 40–60%
  • Teach Traffic, client accounts — Click visibility fell from 95% to 33%; conversion visibility down 60%
  • Practical Ecommerce, five accounts tracked by script — Hidden-term clicks went from under 6% to over 30% in three of five accounts

One agency’s pre-change audit of a $120,000, 26,000-term account showed why the threshold matters so much: 44% of the budget went to terms with five clicks or fewer, and 30% to terms with a single click. Whatever “significant” means, it is above one click, and it is probably above five.

What did Google's 2021 walk-back actually say?

A year later, on 9 September 2021, Google expanded the report to show more queries that met its privacy standards, backdated to February 2021. Agencies reported real improvements; JumpFly saw one account’s visible share go from 13% to 57%. Two details in that announcement are more important than the headline.

The thresholds, Google wrote, ensure anonymity by only reporting on terms that have seen sufficient search volume across all Google searches. Not in your account. Across Google. A query can be searched by 40 different people in your account over a quarter and still be hidden, because 40 is not significant against the whole of Google Search. For a B2B SaaS advertiser whose entire category is a rounding error in global query volume, that is the difference between “some long-tail terms are hidden” and “most of my category is hidden”.

Where do hidden search terms stand today (2026)?

Five years on, the hidden share has not gone back to pre-2020 levels; it has settled at a level the industry has simply learned to live with. A recent agency write-up on the “Other search terms” row reports accounts with 57% and 67% of search-term-report spend hidden, and cites a documented all-exact-match campaign with 80% of spend in “Other”, because close variants are matched and then not reported. The same analysis found visible terms returning 5.6 times spend against 2.6 times for hidden ones. Nils Rooijmans, the Google Ads scripts specialist, answers the question of whether a script can recover the hidden data in one word: no. His coping list is Search Term Insights (Google’s own category-level summary of the queries it will not show you), Keyword Planner, running the same keywords on BING (but who does that honestly?) to read the queries there, and third-party long-tail tools.

Why is “statistical significance” the wrong test for money?

The word Google chose is doing a lot of work, so take it seriously for a moment. Significance is a concept for estimates: a sample is significant when it is large enough that the average you compute from it is trustworthy. It is the right test when the question is “does this query convert?” A query with one click and no conversion tells you nothing about its conversion rate, and Google is correct that you should not draw conclusions from it.

But that is not the question the search terms report answers for an advertiser. The question is “what did I buy?” And cost is not an estimate. A click on “housing sales management software” that cost €18 cost €18. The query’s rarity does not make the charge less certain, and it does not make the advertiser’s judgement that the query is irrelevant any less reliable. You do not need a sample of 50 to know that a query about housing is not a query about enterprise resource planning. One is enough. Google has applied a test designed for inference to a question of inventory, and the effect is that the only purchases you may audit are the ones that recur.

Why the long tail is where the waste lives

Consider what the threshold selects for. A query that is searched often enough to clear it is, almost by definition, a query that other advertisers are also bidding on, that Google’s matching has seen many times, and that you have had the chance to add as a keyword or a negative already. The queries below the threshold are the opposite: one-off phrasings, unusual combinations, mis-matches produced by broad and phrase match reaching for anything semantically adjacent. These are exactly the clicks that are most likely to be irrelevant, and in a competitive B2B category they are not cheap. The CPC on a rare query is set by the same auction as the common one; rarity discounts nothing.

What can visible search terms tell you about the hidden ones?

Here is the inference Google will not let you make directly, but which the visible data supports. Open the visible search terms for any broad or phrase match keyword and sort them into three piles: the keyword itself and its close variants; related queries you are happy to pay for; and queries that have nothing to do with your business. In a well-maintained B2B SaaS account the third pile is not small. Those irrelevant queries got through the same matching logic, from the same keywords, under the same settings, as the hidden ones. The only difference is that they happened to be searched often enough to be shown to you.

There is no mechanism by which the matching that produces 20% irrelevant queries among the visible terms would produce 0% among the hidden ones. If anything the hidden share should be worse, because the rarer a query is, the further it tends to sit from the keyword that matched it. The visible junk is your best available estimate of the hidden junk, and it is an estimate you can actually compute.

What does a hidden search term look like in practice?

Take the keyword “ERP management software” on phrase match. Among the visible search terms you find “sales management software”: a dozen clicks over a month, not your product, added as a negative. That is the system working as designed.

Now consider everything adjacent to that query that is searched a handful of times: “housing sales management software”, “sales team management software for realtors”, “used car sales management software”. Each is searched rarely, each is matched by the same logic that matched “sales management software”, each is clicked occasionally at the full B2B software CPC, and none of them clears the threshold. You will never see them. You cannot negative them individually. You are paying for them every month, and the report that would have told you so in 2019 now rolls them into one row.

The practical consequence is the opposite of what Google’s framing implies. Because the hidden terms cannot be cleaned, the visible terms have to be cleaned harder. A visible query with a single click in 30 days, which in 2019 you might have left alone because the spend was trivial, now has to be evaluated as a representative of everything like it that you cannot see. Adding “sales management” as a phrase negative blocks the visible query and, silently, every hidden variant containing that phrase. That is the only lever that reaches into the hidden half, and it only works if you treat the visible list as a sample rather than as the whole.

What you can still do about the half you cannot see

You cannot recover the hidden queries. You can shrink the hidden share, stop most of the junk inside it, and measure whether you are winning. In rough order of impact:

  • Measure your hidden share every month, per campaign. Sum the cost of the visible search terms, divide by campaign cost. Track it. A campaign at 30% hidden and a campaign at 65% hidden need different treatment, and the number tells you which keywords and match types are feeding the “Other” row. The Google Ads API returns the “Other search terms” row as its own line, so this is one query through an MCP connector or a script.
  • Treat negatives as the only tool that reaches the hidden half. Negative keywords are applied to the actual query at auction time, not to the report, so a negative blocks hidden queries containing it just as it blocks visible ones. This is the whole game. Prefer phrase-match negatives on the offending concept (“sales management”, “housing”, “free”, “jobs”) over exact negatives on the specific query, because the phrase negative generalises into the tail and the exact one does not. Remember that negatives do not match close variants: block “job” and “jobs” both.
  • Lower your bar for what counts as a visible junk query. One click in 30 days used to be ignorable. It is now a sighting of a species. If it is irrelevant, negative the concept, not the string, the same day.
  • Decide match types by hidden share, not by Google’s recommendation. The worse a keyword’s visible-junk ratio, the more of its hidden spend is junk too. Phrase and broad keywords with a high irrelevant share in the visible terms are candidates for exact match or for removal. Note that even all-exact campaigns carry hidden spend through close variants, so exact is a reduction, not a cure.
  • Read Search Term Insights for what it is. Google’s Insights page groups the queries it hides into themes with cost and conversions per theme. It will not give you the query, but a theme called “real estate software” with €2,000 of spend and no conversions is enough to write a negative list from.

If you want to hear more of me ranting on hidden search terms, and clean yours up: let’s talk.