How to find Google Ads negative keywords without losing two hours

Most accounts leak money to a small set of irrelevant queries, and most account managers do not have a repeatable way to find them. This is the working version of that process, in the order you actually do it.

The instruction "review your search terms" is doled out like medicine that everyone agrees is good for you. The instruction is fine. The procedure underneath it is what people skip, and skipping the procedure is why the work stretches into a two-hour afternoon and then never gets done again.

A real sweep takes twenty minutes on a small account, an hour on a large one, and only one of those numbers includes inventing a process from scratch. The other number assumes you have a process. This is the process.

What you are looking for, before you start

The point of the sweep is not to read every search term. It is to find the queries that ate the most spend without paying you back. That distinction matters because it determines the sort order.

You are looking for terms with three properties:

1. High cost. They consumed enough of the budget to be worth blocking.
2. Zero or low conversions. They did not pay you back.
3. No buyer intent. A reasonable person reading the term would not expect a buyer behind it.

All three matter. A high-cost, zero-conversion term that says "best emergency plumber phoenix" is not a negative; it is a real query that just had a slow week. A low-cost, zero-conversion term that says "free plumber" is a negative even though the cost was small, because every dollar it does spend is wasted by definition.

The third property is the only one that requires judgment. The first two are filters. Use the filters to shrink the list to a manageable size, then apply judgment on what is left.

The setup, before you read anything

Three settings, all on the search terms report page.

The clicks-greater-than-one filter is worth a moment. Without it, the report fills up with one-click curiosities, every one of which steals attention from the real high-cost rows. With it, you only see queries that have happened at least twice, which is the point at which a pattern starts to mean something.

The single decision per row

For every row that survives the filters, the question is the same. Did the searcher likely intend to buy what you sell? That is the only question. Everything else is downstream of it.

The mistake that takes the longest to undo is over-blocking. New account owners see a row with no conversions and add it as a negative without checking the intent. Six months later the campaign has been quietly starved of good traffic by a list of two hundred negatives, half of which were applied to terms that simply had a slow month.

If a term feels like a buyer who just did not convert, leave it alone. The rule is conservative on purpose. Negatives compound in both directions; a wrong negative keeps blocking real buyers forever, just like a right negative keeps saving real money forever.

A worked example, top to bottom

Imagine the following first-page output on a residential plumbing campaign in the Phoenix metro. Last thirty days, sorted by cost, filtered to clicks greater than one.

Walk through the eight rows.

emergency plumber phoenix. Clear buyer intent, four conversions, keep.

free ac repair near me. "Free" plus a service category that is not even the campaign's category. Two negatives' worth of mismatch in one row. Add a single-word block for "free" at the shared-list level, and a phrase block for "ac repair" at the campaign level if the campaign is plumbing-only.

24 hour plumber tempe. Buyer intent, three conversions, keep.

how to fix a sink. Information seeker. Add the phrase "how to" as a campaign-level negative. That phrase block alone catches dozens of similar queries.

best plumber scottsdale. Buyer intent, two conversions, keep.

plumber salary 2026. Job seeker. Add "salary" as a single-word block at the shared-list level.

plumbing supply wholesale phoenix. This is the row to be careful with. It looks like a wholesale buyer, who a residential plumber does not serve. But "supply" and "wholesale" are both real words your business might use in some other ad copy. Add the phrase "supply wholesale" as a phrase-match negative at the campaign level, which catches the specific overlap without blocking innocent terms like "plumbing supply phoenix" if you ever expand to commercial.

diy garbage disposal replacement. "DIY" is the tell. Add "diy" as a single-word block at the shared-list level. "Replacement" is fine on its own; do not block it.

Eight rows produced four negatives and one phrase block. About four minutes of judgment, applied once, which keeps applying for the life of the account.

Where the time actually goes

The sweep takes longer than it needs to for two reasons.

The first is decision fatigue. After about thirty rows, the brain starts auto-flagging things as either "obviously fine" or "obviously negative" without re-examining the intent. That is when over-blocking starts. The fix is to take a one-minute break at row thirty, even if you don't feel like you need one, and then keep going. The break is uncomfortable; the decisions after the break are sharper.

The second is mid-sweep tangents. A row reads "law firm associate hours 2026" and you suddenly want to look at the law firm campaign instead. Or a row makes you want to rewrite the ad copy. Or you remember that the landing page for emergency plumber has a spelling typo. None of these are wrong observations. They are wrong work to do during a sweep.

The notes file does not need to be elaborate. A scratchpad with three or four lines is enough. The point is that the thought is captured so it stops competing for attention while you finish reading the report.

The shape of a typical first sweep

After about an hour on a previously un-swept account, the typical output looks like this.

That is twelve hundred dollars a month of leakage cleared by a one-hour pass through a UI screen. The savings compound monthly because the blocks are persistent. The ROI on a single sweep is roughly the rest of the year of the campaign, which is why the work pays back even when the account owner is paying themselves.

When to use a tool and when not to

A spreadsheet plus the search terms report is fine for one-account, monthly-cadence work. The decisions are not hard; they just repeat. A human can read a row, judge it, and apply a negative in about thirty to ninety seconds, depending on how often they have to switch between the report and the negatives screen.

A tool helps when the per-row time matters. We built Sensei Ads so that the read-judge-apply loop drops from sixty seconds per row to about ten. The CSV plus a paragraph of business context goes in, the categorized list of negatives comes out with reasons attached, and the human reviews instead of inventing.

The honest comparison is this: if you sweep one account a quarter, the tool is overkill. If you sweep three or four accounts every two weeks, the tool earns the subscription back inside a single sweep. The decision point is not the size of the account; it is the cadence and the count.

The next sweep, and the one after

The second sweep on the same account will go faster than the first because most of the obvious blocks are already in place. It also tends to surface a different kind of leakage: the queries that survived because they looked ambiguous in sweep one and that the data has now disambiguated.

By the third or fourth sweep, the report becomes quiet. The top fifty rows are mostly real buyer queries with reasonable conversion rates, and the work shifts from finding leaks to maintaining the list. That is the point at which the cadence can drop, and the account owner can stop dreading the sweep.

If the report still looks chaotic at sweep four, the cause is usually further upstream: match types that are too broad, ad copy that misrepresents the service, or a campaign structure that mixes intents. The negatives can keep up with the symptoms, but the underlying issue needs its own work. The match types piece is the right next read for the broad-match part of that.