Search Term Mining and Negatives Where Spend Justifies Them

Google has withheld search terms below a privacy threshold since September 2020, so the search terms report does not reconcile to campaign spend and has not for years. The first number this engagement produces is that gap on your account: the share of spend with no term attached to it. It sets the ceiling on what mining can achieve, and it decides whether sculpting is the right move at all or whether the answer is a structural change that reduces how much unseen traffic you are exposed to in the first place. How the mining is done: Terms come through the Google Ads API rather than the interface, because this is n-gram work and the interface cannot do it. Every term is broken into unigrams and bigrams, and those fragments are aggregated by spend, conversions and cost per conversion across the whole account rather than campaign by campaign. The pattern that matters is almost never one expensive term. It is a two-word fragment appearing across four hundred terms, each individually too small to notice, together carrying a fifth of the waste. That is the finding you are paying for, and it does not appear in a sorted table however long you stare at one. The threshold, and why most negatives should not exist: A negative earns its place when blocking it moves a number you would report. Everything below that floor is noise, and the floor is set against your account's own spend rather than a house rule. A report announcing 240 negatives added last month describes activity, not value, and each of those entries is a permanent constraint somebody will have to reason about in two years when performance dips and nobody remembers why. I add few, I write down why, and I remove the ones that have stopped earning their place. Over-negation is the damage nobody measures: Before a list is applied I check candidate terms against historical conversions, because the expensive mistake here is not the waste left in. It is the converting query quietly blocked by a fragment that looked obviously irrelevant. Generic words are the usual offenders. In Hebrew the risk runs higher, since a fragment can sit inside a legitimately different word once a prefix particle attaches to it. Negatives go in tranches with the conversion effect read between them, never in one push. Performance Max, honestly: Account-level negative keywords apply to Performance Max and are capped at 1,000, which makes that list a budget to spend rather than a bin to fill. Brand exclusion lists handle the brand question separately. Search themes are inputs, not keywords, and the search category report is aggregated to a level where mining is coarse. You will get a plain account of what can and cannot be seen there rather than a confident chart built on a report that does not support one. What I will not do: Sell a wasted spend report. Apply a template negative list assembled for somebody else's account, which is the fastest way to block converting traffic on day one. Sculpt below the spend floor to make a deliverable look thicker. Take on search term work while conversion tracking is broken, because mining is downstream and inherits every measurement error above it. Not for you if: Your monthly search spend is small enough that the gap this closes will not cover the fee. That arithmetic is simple and I will do it before you buy rather than after. You want a fixed weekly count of negative additions as evidence of effort. You need a document to show a client rather than a change to an account.