Negative Keywords Hurting Performance? Six Signs

Negative keywords hurt performance when they block demand, not waste

A negative list turns harmful the moment it suppresses queries that would have converted. You will not watch it happen. Google has no report called "impressions you lost to your own negatives" — the cost arrives as an absence.

In the accounts we audit, over-exclusion almost never comes from someone deliberately blocking a good term. It comes from three habits.

Single-word broad negatives. A plumber gets tired of job seekers and adds broad negative job. It kills plumber jobs near me and plumbing apprenticeship, which was the point. It also kills small plumbing job cost, emergency plumbing job tonight and plumber for a one day job. Job-seeker traffic is worth blocking — there is a tight list of terms that does it — but the word job is not the thing to block.

Borrowed lists. A 400-entry "universal negative list" written for a B2B software account gets pasted into a discount retailer. cheap, free, discount, used, budget. Those are that retailer's buying words. cheap running shoes womens size 8 is not waste, it is the customer describing herself.

Negatives written from the wording instead of the intent. repair or replace boiler reads like a repair query to a company that only sells new boilers. It is one of their best converting terms. Add broad negative repair and it goes, and so does boiler repair cost vs new one.

The same trap sits in the mid-funnel vocabulary: review, best, vs, comparison, alternative. People type those words minutes before they buy. Blocking review to stop review-site readers also blocks best reviewed dentist near me. Deciding where that line sits is a judgement call, and it is the same judgement as whether to keep competitor searches — sometimes worth it, never automatic.

Six signs your account is over-excluded

None of these is proof on its own. Two or three together, in an account whose negative list grew fast in the last quarter, is close to it.

1. Spend sits under budget while CPCs stay flat

The cleanest signal. If impressions fell, CPCs did not rise, and search impression share lost to budget is near zero, then you are not being outbid and you are not capped. Your eligibility shrank. Something is stopping you entering auctions you used to enter, and negatives are the most common something.

Check lost impression share to rank in the same window. If that is stable too, the auctions did not get harder. You just stopped showing up in them.

2. New search terms stop appearing

A healthy Search or Shopping account throws up unfamiliar queries every week. Some are waste, some are the next keyword you should build a group around. Count the distinct search terms your account served last week and compare it to the same week three months ago. When the report returns the same forty terms month after month, discovery has stopped.

That matters most on broad match and Performance Max, which depend entirely on discovery to work. It is also why the cadence question — how often you should actually read the search terms report — is not the same as how often you should add to it.

3. Your CPA improved and your conversions fell

This is the sign people celebrate by mistake. See the arithmetic in the next section.

4. Google itself is flagging conflicts

Google surfaces two things for free. In the Recommendations tab, "remove conflicting negative keywords" appears when a negative blocks one of your own keywords. In the keyword table, individual keywords get a "negative keyword conflict" status.

Fix the real ones. But understand the limit: Google only compares negatives against keywords you have. It cannot flag the query you would have received if the negative were not there. Conflict warnings catch the loudest tenth of the problem.

5. The negative list is bigger than the keyword list

Not fatal by itself — some large ecommerce accounts legitimately carry thousands of negatives. But when an account with 300 keywords carries 4,000 negatives, most of those negatives were added to block a single query each, and nobody has ever read them back. We wrote about where the real ceiling sits in how many negative keywords are too many; the count matters far less than the match types inside it.

6. Broad negatives dominate the list

Export the list and count how many entries are broad match and one word long. In accounts we audit those entries are usually a small share of the list — often under one in ten — and responsible for most of the traffic the list removes. Broad negatives also behave unpredictably across match types, which is the whole argument in negative keyword match types don't work like positives. A broad negative is a blunt instrument being used for surgery.

A better CPA can still be a worse month

Here is why sign 3 fools people. Take an account spending $8,000 a month, producing 100 conversions, so an $80 CPA. Someone spends a Friday adding negatives. Next month:

• Spend: $6,000
• Conversions: 80
• CPA: $75

Every number on the dashboard improved. Spend down $2,000, CPA down $5. The report writes itself.

Now put a value on a conversion. Say gross profit per conversion is $200. Twenty conversions gone is $4,000 of gross profit gone, to save $2,000 of media. The month is $2,000 worse than the one before it, and the account looks tidier than it has all year.

The number that would have caught it is conversion value, or at minimum conversion count, tracked beside cost. CPA alone cannot distinguish "we stopped paying for rubbish" from "we stopped paying for customers". If you are going to argue about negatives with a client or a boss, argue in profit — the method is in how to calculate wasted ad spend honestly.

One caution before you blame the list. Volume drops have many parents: a tCPA or tROAS change, a feed disapproval, seasonality, a competitor entering, a landing page slowing down. If your negatives went in the same week as a bid strategy change, you have two suspects and no evidence. Work through the causes in order rather than assuming the most recent edit is the guilty one.

You cannot see what you blocked, so plan around it

This is the honest limitation at the centre of the whole topic, and most advice skips it.

There is no report — not in the interface, not in the API — that lists queries your negative keywords suppressed. campaign_search_term_view returns terms that were served, for Search, Shopping and Performance Max. Display, Video and App return none at all. A blocked query produces no row anywhere. It simply never existed as far as your account is concerned.

Two further gaps narrow the evidence:

• Low-volume queries never reach the search terms report in the first place, because of Google's privacy threshold. So even historical data is partial — more on what goes missing in the search terms report and what it hides.
• The query you see is not always the query typed, and the match-source column tells you less than it appears to. That is covered in search term match source and its limits.

The practical consequence: the only evidence about a negative is what the terms it blocks did before you added it. Which means the date you added a negative, and the report window that predates it, are the two most valuable things you own during a walk-back. Record the date. Keep the export.

How to walk it back without undoing everything

The instinct — delete the list, start clean — buys back the demand and the waste in the same motion. Two weeks later the plumber salary and how to fix a leak yourself clicks are back and you have no idea which removal helped. Do it in batches instead.

Step 1: freeze additions for two weeks

Stop adding anything. You cannot measure a removal while the list is still growing. Two weeks is usually enough to get a clean baseline for a mid-sized account; on low-volume accounts, take four.

Step 2: export everything and sort the suspects

Pull every negative with four columns: the text, the match type, where it lives (shared list, campaign, ad group), and the date it was added if you have it. Then sort into three piles.

• Pile A — high suspicion. Broad, one or two words, generic. job, free, cheap, repair, review, used, diy, course, template.
• Pile B — inherited. Anything from a list you did not write for this account. Whole shared lists copied in count as one suspect each.
• Pile C — leave alone. Exact and phrase negatives that match a specific observed query: plumber salary, plumbing courses near me, how to unblock a sink yourself. These almost never cause an over-exclusion problem. Do not touch them. They are the part of the list that is working.

If you want a structured pass over the rest of the account while you are in there, the one-hour audit checklist covers the neighbouring settings that also throttle volume.

Step 3: test each suspect against history

For each entry in pile A, pull the search terms report for the 90 days before the negative went in. Filter for terms containing that word. You now have a real answer to a real question: how many conversions did the queries this negative blocks actually produce?

Three outcomes. Conversions there — remove the negative, and if the waste was real, replace it later with a phrase negative on the specific bad query. Clicks but no conversions — keep it, but narrow broad to phrase. Nothing at all — the negative is doing nothing either way; it survives on a technicality.

Where this method fails: if the negative was added before your report window, or added by an automated tool months ago, there is no history to check. In that case treat it as pile A and remove it as a live test. You will have to buy the answer with a few hundred dollars of traffic. That is cheaper than guessing for another year.

Step 4: remove 20 to 40 at a time, and write the date down

One batch, one date annotation in the account. Do not spread the batch across campaigns at random — keep each batch inside a set of campaigns you will be reading together, or you will not be able to attribute anything.

Step 5: judge each batch on three to four weeks

Smart Bidding needs time to re-learn a widened query set, and conversion lag adds more. Reading a removal after five days tells you about CPCs and nothing about conversions. Judge on impressions, clicks, conversions and conversion value together. Costs went up and conversions did not? Put that batch back. Both went up? Keep going.

If a removal turns out badly, the ability to reverse it in one action matters more than the removal itself. PPCleaner keeps every block undoable for 90 days for exactly this reason — the batch you regret should cost you a click, not an afternoon of re-typing exclusions.

Performance Max and Shopping over-exclude in their own ways

Performance Max is the easiest campaign type to strangle and the hardest to diagnose. It has no keywords to conflict with, so Google's conflict warnings never fire. It accepts up to 10,000 campaign-level negative keywords and does not accept shared negative keyword lists — so every exclusion was typed into that one campaign, and removing them means editing that campaign directly rather than lifting a list off it. Copying a 2,000-entry list into six Performance Max campaigns also means six separate walk-backs. Both halves of that are laid out in negative keywords in Performance Max and the 10,000 cap.

Performance Max also has less to fall back on. A Search campaign with a strangled query set still has its keywords. A Performance Max campaign that can no longer reach its cheap search queries reallocates toward whatever inventory is left, and that inventory is usually worse. The quiet version of this failure is described in where Performance Max wasted spend hides.

Standard Shopping behaves more like Search — it takes negative keywords and shared lists — but the query patterns are different enough that a Search-style list does real harm. Model numbers, sizes and material words look generic and are not: broad negative used on a parts retailer removes used oem brake pads, which converts. The differences are worth reading before you copy a list across: Shopping campaign negative keywords work differently.

Rules that stop a list turning against you again

A list that grows for two years without a review will always end up over-excluded. These five rules keep the growth safe.

Block queries, not words. The unit of exclusion is a query you saw in the report, added as phrase or exact. plumber salary, not salary. It takes more entries to cover the same waste, and that is the price of not blocking your own customers.

Reserve broad for words that cannot appear in a buying query. For most businesses that is a very short list: salary, jobs, vacancy, wikipedia, pdf, meaning, lyrics. If you can construct one plausible buying query containing the word, it is not a broad negative.

Never inherit a list unread. Every entry moved between accounts needs the pile-A test above. A list that was correct for a law firm is dangerous in a legal software account.

Keep the date. Whatever adds your negatives, you need to know when each one went in. Without dates, the historical test in step 3 is impossible and you are down to live testing everything.

Re-read the whole list once a quarter. Twenty minutes. You are looking for two things: broad entries that crept in, and negatives whose reason has expired — the product you stopped selling, the competitor who went out of business, the seasonal exclusion you never removed.

One last note on automation, including ours. PPCleaner never blocks a term that converted in the last 90 days, only ever adds negatives, and keeps every block reversible for 90 days. That rules out the worst class of mistake but not all of them. A term with no conversions in 90 days can still be an early touch in a six-month B2B cycle, or a genuine seasonal buyer who shows up in October. If your sales cycle is longer than the window, treat the 90-day rule as a floor, not a guarantee, and read the removals on high-value campaigns yourself. Anything that adds negatives to your account — a script, a tool, a junior, you on a Friday afternoon — has to be readable and reversible, or you cannot walk it back at all.