I negated a query at 1am that had bounced 34 times in a row. Two weeks later, the client asked why Tuesday phone bookings had dropped by a third. That query was the Tuesday query.

The query had looked like textbook waste: same landing page, zero form fills, engagement rate in the single digits. I had treated a bounce as a vote on intent. It wasn’t. The useful decision is whether the query is wrong or the page is wrong. To make it, you need the search term, its landing page, GA4 engagement and the conversions that don’t necessarily happen on that page. Negate the first problem. Fix the second. Leave a query tied to calls or pipeline alone.

What does bounce rate actually tell you?

In Universal Analytics, a bounce meant a one-page session with no further interaction. GA4 defines it differently: bounce rate is the share of sessions that were not engaged. An engaged session lasts longer than 10 seconds, has a key event, or has two or more page views. Engagement rate is the share of sessions that clear one of those bars; bounce rate is the remainder.

That makes bounce useful, but narrow. A prospect can read your price list quickly, take the phone number and leave without GA4 recording an engaged session. Another visitor can keep a tab open long enough to count as engaged without getting any closer to buying. The engagement definition measures what happened in the session, not what the visitor intended or what happened afterward.

I treat bounce as a flag for where to look, never proof of what to cut. That matters most when calls close the deal. One agency audit found Google Ads recorded only one-third to one-half of real sales on phone-close accounts until offline conversions were imported. If your measurement misses the sale, a bounce-rate purge can look efficient right up until the bookings stop.

Step 1: Which search terms are worth investigating?

Start in the search terms report, not the negative-keyword list. Export the last 30 days and sort by cost. I don’t make a routine exclusion decision under 20 clicks or 1x target CPA in spend, unless the intent is plainly wrong. A query for jobs, DIY instructions or something free does not need to consume a full CPA before you recognise that you don’t sell it.

That floor is a filter, not a law of statistics. Practitioners looking for negative-keyword problems warn that 10 to 50 clicks with zero conversions can still be too little to negate when the intent is plausible. Adalysis uses a stricter starting filter for this sort of review: roughly 25 clicks, zero conversions and very high bounce, then a check for patterns across queries. I use the floor to keep the review focused. I don’t use it to turn every eligible term into a negative.

Carry the expensive, high-bounce terms into diagnosis. Keep their cost and conversion data beside them. A term that spent money and produced no form fills is a candidate for investigation, not yet a verdict.

Step 2: Did the query fail, or did the landing page fail?

A search term never bounces. The session on the landing page bounces. Before blaming the query, find out which page the click reached. For AI Max, the Search terms and landing pages view shows query-to-page pairings in one report. Elsewhere, use the account’s search-term and landing-page information, then inspect the relevant landing page’s behaviour in GA4.

Be precise about that join. A GA4 report filtered to a landing page and session source/medium shows behaviour for traffic to that page; it does not, by itself, assign an engagement rate to each search term that landed there. Use it to spot a page pattern. Use search-term-level Google Ads conversion data, calls and imported offline conversions to judge a particular query. If several terms share a page, don’t pretend the page’s bounce rate belongs to each of them equally.

Then ask the cheaper question first: would I want this visitor if the page were perfect? A bad query often answers itself. Look for jobs, hiring, salary, course, training, DIY, free, cheap, used, repair when you sell new, and locations you do not serve. That check comes before bids or page work because no landing-page rewrite makes an instruction-seeker into a buyer of new equipment.

I once found commercial ice machine repair manual eating $400 in a month for a client who installed new units. Bounce rate was 94%. The word manual settled the intent question; the bounce rate did not. I negated the term that day.

A page problem has a different shape. Five or six relevant queries all reach the same URL and all struggle there. The ad promises same-day garage door repair in Austin; the click reaches a generic homepage with no price, no service area and a contact form buried below three stock photos. That homepage-for-everything pattern can lose buyers you already paid to reach. So can a message mismatch: the query is monthly bookkeeping for agencies, but the page opens with annual tax returns.

Read the query before the percentage, then read the page through the searcher’s eyes. If the query describes a service you sell and the page answers another question, fix the pairing, not the keyword.

Step 3: Should you negate, reroute or leave it alone?

There are three decisions, and only one belongs in a negative list:

  1. Negate the wrong query. Job seekers, students, DIYers, freebie hunters, the wrong product and the wrong city belong here when those intents do not fit your business.
  2. Reroute the right query on the wrong page. Give that intent an ad and page that answer it directly.
  3. Leave a valuable query alone. Check calls, assisted conversions and open opportunities before deciding that a bounced session produced nothing.

When you negate, match type and level matter. Negative keywords block differently from positive keywords: a broad negative requires all its terms, in any order; a phrase negative requires the same order; an exact negative targets that query without extra words. Plurals and synonyms need separate attention. My default is a phrase negative at campaign level for a clearly bad phrase that should never serve there, and an exact negative at ad group level when only one variant is bad. If the placement is uncertain, start narrow. The complete guide to negative keywords is worth keeping open when you make the upload.

Don’t promote a bad pairing in one ad group into a campaign-wide ban without checking the others. A term that fails against one page may convert against another. Root-word bans deserve the same suspicion: excluding job to stop employment searches can also stop relevant queries such as job site delivery or job lot pricing. Check the terms the negative would catch, and check for conflicts before uploading it. An industry list of 700+ negative-keyword ideas can help you find candidates; it is not a list to paste into a live account unexamined.

Rerouting takes more work, but it preserves demand. Say emergency plumber cost bounces at 88% on your homepage. That searcher wants a number and a phone button, not your origin story. I would split the intent into its own ad group, repeat the price range in the ad copy and send the click to a page that opens with that range and a click-to-call button. Same query, better pairing. Where I run this with groas, a landing page built for every ad group comes with the Search Ads service, so separate intents do not have to share one generic page.

Leaving a term alone can be the hardest call when the dashboard says zero. Local services often get quick phone calls. B2B prospects may leave after a first visit, return later and move into an opportunity that your web-session report does not explain. Count real enquiries as primary conversions and import offline conversions before judging those terms. Give a plausible query its sales cycle. My rule: if it has a call, an assisted conversion or a linked open opportunity, I don’t negate it for bouncing. Bounce rate never outranks money.

Step 4: What can you automate without wrecking the account?

Automate the sorting and the alerts. Keep judgment in the decision. A free auto-negative script connected to a Sheet of positive words illustrates the trade-off: it can add queries that lack those words after an impression minimum, run on an hourly schedule, and offer a preview and email alert. It enforces a word rule. It cannot tell whether a caller booked or whether the ad sent a buyer to the wrong page.

I will let a script surface candidates with spend over 1x target CPA, zero recorded conversions and engagement under 10%. I won’t let bounce alone trigger an exclusion. I also use preview mode for the first two weeks and inspect what the rule proposes. Rules that pile up hundreds of exact negatives can run into account limits; shared lists help with that operational problem, not with bad judgment.

Most AI tools have a similar boundary. Flagging a high-bounce query and drafting a negative is useful work. It isn’t the same as checking a phone call, a later return or five relevant queries stranded on one weak page. The buyer’s guide to AI Google Ads tools is useful here for one question: what will the tool exclude automatically, and what evidence will it check first? Keep a human on the negate button for anything with assisted value.

Where I want execution handled rather than another queue of recommendations, I use groas. It works on search terms, pages and bids around the clock, while a strategist owns the decision to negate or reroute. Cause first, exclusion second. Never the reverse.

Which mistakes cost the most?

Three mistakes account for most of the waste I see in this review:

  • Negating on bounce alone. Twelve clicks and 100% bounce tell you little. Even at 40 clicks, a tracking gap can look like an intent gap until you check calls and assisted conversions.
  • Banning a lazy root word. The relevant queries caught by a negative can cost more than the irrelevant ones it removes. Check the surrounding terms first. An n-gram review can reveal a genuinely bad shared phrase, such as custom koozies spread across 38 separate queries, without assuming every term containing one word is waste.
  • Applying the negative too high in the account. A campaign negative based on one ad group’s bad pairing can block a term converting elsewhere. I check conflicts before every upload.

Skip those three errors and the job gets simpler. You are no longer asking, “Which percentage looks ugly?” You are asking what the visitor wanted, where the ad sent them and whether the business got value.

What should you do today?

My Friday routine takes 30 minutes. It replaced the monthly bounce-rate purge I used to do at 1am. I sort the last 30 days of search terms by cost, hide terms below my review floor and investigate the top 20 that remain. I put landing pages and GA4 engagement beside the queries, then check key events, calls and imported offline conversions. Page-level behaviour tells me where to inspect. Query intent and conversion evidence tell me what to change.

  1. Sort by cost. Set aside terms under 20 clicks or 1x target CPA in spend, unless plainly wrong intent is already visible.
  2. Check the page and the outcome. Find the query-to-page pairing, inspect GA4 engagement for that page and check real enquiries, not form fills alone.
  3. Make one of three calls. Negate the wrong query at the narrowest useful level, reroute the right query on the wrong page, or keep a term tied to calls or pipeline.
  4. Log and check. Record the change and check negative-keyword conflicts before the next upload.

Do the first pass today: pull your 10 most expensive search terms from the last 30 days and put each beside its landing page and its calls. You may find a term to negate, a page to fix or a high-bounce query that earns its keep by phone. I keep that Tuesday query taped to my monitor as a reminder. Bounce rate starts the search. Revenue makes the call. If you want the join, page build and weekly decision handled around the clock with a strategist accountable for the numbers, that is what groas runs as Search Ads for a flat fee.