The Negative Keyword List Every Service Business Needs (And How to Build Your Own)

The short answer

Keywords tell Google what you want to be found for. Negative keywords are the only instrument in the account for telling it what your business does not do — and that is a genuinely different statement, not the inverse of the first one. There is no way to express an exclusion using a positive keyword. You can bid narrowly, you can use exact match, you can write ad copy that repels the wrong reader, and none of those actually prevent the impression, the click or the charge.

The mental model that makes this click is that broad and phrase match do not buy words. They buy intent categories. When you bid on a broad match term for a service you provide, you are asking the auction to find people whose search means roughly the same thing, and "roughly the same thing" is decided by a system that knows a great deal about language and nothing at all about your business. It does not know you serve one metro area, that you sell installation rather than parts, that you are a clinic and not a school, or that the word in your keyword also means something entirely unrelated in another industry.

The negative keyword list is where that missing context gets supplied. Everything else in the account describes what you want. This is the only place you describe what you refuse.

The three kinds of waste

Most published negative keyword lists are an undifferentiated pile of terms, which is exactly why they get pasted in once and never maintained. Nobody can extend a pile. The organizing framework that makes a list maintainable is that every negative keyword you will ever add belongs to one of three categories, and knowing which one it belongs to tells you how aggressive to be with it.

• Wrong intent — the person wants information, instruction or employment rather than a service. They are researching, learning to do it themselves, or looking for a job. Nothing about your offer is wrong for them; they are simply not in a buying posture and will not be during this session.
• Wrong economics — the person wants the service but at a price point you do not and will not serve. Free, cheap, discount, secondhand, DIY-adjacent and charity-seeking queries all live here. This is the category that requires the most honesty, because "wrong economics" is a claim about your own pricing, not about the searcher.
• Wrong entity — the query is about a different thing entirely. A competitor, an adjacent industry, a product rather than a service, or a homonym where your keyword means something else in another context. This is where the largest, quietest waste usually hides, because the queries look plausible in a report until you read them one at a time.

The categories behave differently, and treating them identically is a mistake. Wrong-intent terms are usually safe to exclude broadly and permanently — a person searching for a salary figure will not become a customer. Wrong-economics terms need judgment, because "cheap" and "affordable" mean different things in different markets and a blanket exclusion of price-sensitivity language can remove a real segment. Wrong-entity terms are the most valuable to find and the most account-specific, which means no pre-made list will ever cover them for you.

How negative match types actually behave

Negative keywords have their own match types, and they do not behave like the positive ones. This is where most of the technical mistakes live, and they are invisible mistakes — a negative that does not block what you assumed it blocked produces no error, no warning and no line in any report. It simply keeps costing money.

• Negative broad match blocks a query only if every word in your negative appears somewhere in that query, in any order. Adding "free estimate" as a negative broad term blocks "estimate free" but does not block "free quote", because the word "estimate" is missing.
• Negative phrase match blocks queries containing your exact word sequence, with anything before or after. "Free estimate" as a negative phrase blocks "free estimate near me" but not "estimate for free", because the sequence is broken.
• Negative exact match blocks only that query, word for word, with nothing added. It is the surgical option and the one people reach for far too rarely — it is the right tool when a single specific query is the problem and its neighbors are fine.

Now the part that trips up nearly everyone: negative keywords do not use close variants the way positive keywords do. A positive keyword happily matches plurals, misspellings, stemmed forms and reordered phrasings. A negative keyword does not. Adding "job" does not block "jobs". Adding "salary" does not block "salaries". Adding "how to" does not block "howto". Adding "second hand" does not block "secondhand" or "second-hand" in the way you would expect from a positive keyword.

This is why competent negative lists look tediously repetitive, with singulars and plurals sitting next to each other and obvious misspellings included on purpose. If a list looks elegant, it is probably incomplete. Build the habit of adding both number forms and the two or three misspellings you actually observed in the search terms report, rather than the ones you can imagine.

One more mechanic worth stating plainly: when a query matches both a positive keyword and a negative keyword, the negative wins. There is no bidding contest, no relevance tiebreak, no "but this keyword is exact match". The exclusion is absolute, which is exactly why an over-broad negative can silently suppress an entire ad group without anything in the interface pointing at the cause.

Campaign level, ad group level and shared lists

A negative keyword can live in three places, and choosing the wrong one is the most common structural error in an otherwise well-built account. The choice is not a matter of taste. Each level answers a different question.

• Shared negative keyword list — for exclusions that are true about the business itself and will never stop being true. Employment terms, DIY intent, other industries that share your vocabulary. Maintained once at account level, attached to every relevant campaign, and updated in one place when something new turns up.
• Campaign level — for exclusions that are true about what this campaign sells but not necessarily about the whole account. A campaign selling installation should exclude repair queries if a separate repair campaign exists, and the exclusion is a campaign fact, not a business fact.
• Ad group level — for steering rather than excluding. When two ad groups in one campaign compete for overlapping queries, ad group negatives are how you decide which one wins each query. Nothing is being turned away; it is being routed.

The failure that costs the most

Here is the specific way this goes wrong, and it happens in accounts that are otherwise well managed. Someone notices junk traffic in one ad group. The fastest fix is to add the offending word to the shared list or the campaign level, because that is one action instead of several. It works. The junk stops. And somewhere else in the account, an ad group that was built specifically to capture a query containing that word goes quiet.

The canonical example in service businesses is the word "free". Excluded at account level as a broad negative, it stops "free wallpaper downloads" and it also stops "free estimate", "free quote", "free consultation", "free inspection" and "free site visit" — which, for a trade or a professional services firm, are among the highest-intent queries in the entire category. The offer literally is a free estimate. The exclusion removes the searches most likely to convert, and it does so without producing a single warning.

The rule that prevents it: exclusions go at the narrowest level that solves the problem. Only promote a negative to the shared list once you have confirmed it is true everywhere. And when you do add something to a shared list, check what it would have blocked over the last quarter before saving it — the search terms report is right there, and a search on the term takes under a minute.

Google publishes limits on how many negative keywords a shared list can hold and how many a single campaign can carry. Those limits have changed over the years and I am not going to quote a number that may be wrong by the time you read this — check the current Google Ads documentation. The practical implication is only that lists are a finite resource, so a list bloated with speculative terms nobody verified has a real cost beyond the risk of over-blocking.

The starter lists

What follows is a genuinely usable starting point for a service business, organized by category so you can take the parts that apply and leave the parts that do not. Read the reasoning above each list before copying the terms. A list you understand is a list you will maintain; a list you paste is a list that will be wrong within a quarter and nobody will notice.

Add these as phrase match negatives unless noted otherwise, and remember that plurals and misspellings need their own entries. Where a term is genuinely part of your offer — and "free" is the usual candidate — leave it out and handle it at ad group level instead.

Job seeker and employment terms

The largest single source of wrong-intent traffic for any service business that also employs tradespeople, clinicians or consultants. People searching for work use the same industry vocabulary as people buying, and the auction cannot tell the difference. This category is close to risk-free to exclude broadly, because employment language almost never appears inside a genuine purchase query.

• job, jobs, career, careers, hiring, hire near me
• vacancy, vacancies, employment, employed, recruitment, recruiter
• apprentice, apprenticeship, internship, intern, trainee, traineeship
• salary, salaries, wage, wages, pay rate, hourly rate for
• resume, cv, cover letter, job description, job application, application form
• staffing, staffing agency, temp agency, work from home, part time, night shift
• indeed, glassdoor, linkedin jobs, seek, monster, job board

DIY, tutorial and how-to terms

Someone typing "how to" has already decided not to hire anyone, at least for the next few minutes. They may hire you in three days when the DIY attempt fails, but they will not do it from this click. The nuance worth knowing: this traffic is genuinely valuable for organic content and genuinely worthless for a conversion-optimized search campaign, which is why the right response is a blog post, not a bid.

• diy, do it yourself, do-it-yourself, myself, on my own, by myself
• how to, how do i, how can i, how does, what is, what are
• tutorial, tutorials, guide, step by step, instructions, manual, pdf
• youtube, video, watch, reddit, forum, quora
• homemade, home remedy, build your own, make your own, fix it yourself
• self install, install myself, repair myself, replace myself, unclog myself
• tips, tricks, for beginners, for dummies, cheat sheet, checklist template

Free, cheap and discount-seeker terms

This is the category that requires the most honesty about your own business. If you sell premium and your close rate on price-led inquiries is poor, these terms are pure waste. If you compete on price, several of them are your best keywords. And if your offer includes a free estimate or free consultation, excluding "free" at a broad level will do more damage than every other mistake in this post combined.

• free, for free, freeware, no cost, zero cost, at no charge
• cheap, cheaper, cheapest, low cost, lowest price, budget
• affordable, inexpensive, bargain, economical, value for money
• discount, discounted, coupon, coupons, promo code, voucher
• deal, deals, offer code, clearance, sale, closing down
• free trial, freebie, giveaway, sample, free sample, no obligation download
• pro bono, charity, free of charge, subsidized, government grant, government scheme

Research and information-seeking terms

Distinct from DIY: this person may well hire someone, but not today and not from this session. They are building a mental model of the category — what it costs, how it works, what the options are. The honest read is that a conversion-optimized search campaign is the wrong instrument for them, and content plus remarketing is the right one.

• meaning, definition, define, what does it mean, explained, explanation
• vs, versus, comparison, compare, alternatives, alternative to
• reviews, review of, ratings, rated, best of, top 10 list
• statistics, report, study, research, white paper, case study
• history of, types of, list of, examples of, difference between
• pros and cons, is it worth it, should i, do i need, when to
• wikipedia, wiki, blog, article, news, near me open now hours

Education and training terms

If you sell a service and not a course, every training query is waste — and the overlap is severe in trades, healthcare, finance and marketing, where the same words describe both the service and the qualification to perform it. This category also catches a specific trap: students searching assignment language will match professional-services keywords almost perfectly.

• course, courses, class, classes, training, trainings
• certification, certificate, certified course, diploma, degree, qualification
• school, college, university, institute, academy, bootcamp
• exam, exams, test prep, practice test, syllabus, curriculum
• learn, learning, teach, teaching, lesson, lessons
• student, students, assignment, homework, thesis, dissertation
• workshop, seminar, webinar, masterclass, online course, udemy, coursera

Used, second-hand and rental terms

Relevant wherever your service involves equipment, property or physical goods that people also buy used or rent. A person searching for a used unit wants a marketplace listing, and a person searching to rent wants a rental company. Neither wants an installation quote. Skip this block entirely if you actually offer rentals — several of these are then your money terms.

• used, second hand, secondhand, second-hand, pre owned, preowned
• refurbished, reconditioned, salvage, scrap, parts only, for parts
• rent, rental, rentals, hire, for hire, lease, leasing
• craigslist, ebay, marketplace, olx, classifieds, gumtree
• buy, for sale, price list, wholesale price, supplier, distributor
• diy kit, kit, spare parts, replacement parts, accessories, components

Legal, complaint and scam terms

Two different problems in one category. The first is reputational: queries pairing your industry with "scam", "complaint" or "lawsuit" are people investigating a bad experience, and a paid ad in that context is money spent to appear next to a negative frame. The second is regulatory: queries about suing a provider in your industry are not looking for a provider.

• scam, scams, fraud, fraudulent, ripoff, rip off
• complaint, complaints, complain, report a, bbb, better business bureau
• lawsuit, sue, suing, legal action, class action, settlement
• compensation, claim against, negligence, malpractice, liability claim
• refund policy, cancel my, cancellation policy, chargeback, dispute charge
• license lookup, licensed check, verify license, is it legal, illegal, regulations

Wholesale and B2B terms for a B2C advertiser

If you serve homeowners and consumers, trade and procurement language is a clean signal that this query is not yours. The economics are also wrong even when the fit looks plausible — a wholesale inquiry priced against your retail service will not close, and will consume a sales conversation on the way to not closing.

• wholesale, bulk, bulk order, trade price, trade only, trade account
• supplier, suppliers, manufacturer, manufacturers, distributor, distributors
• oem, odm, white label, private label, reseller, dropship
• b2b, procurement, tender, rfp, rfq, request for proposal
• contract manufacturer, factory, wholesaler, import, export, moq
• franchise, franchising, become a dealer, dealer application, partner program

Consumer terms for a B2B advertiser

The reverse case, and the one people forget to build. If your service is sold to companies, consumer-scale language in a query is a reliable indicator that the searcher is an individual with a domestic problem, and your minimum engagement size makes that conversation impossible.

• home, house, apartment, flat, residential, for my home
• personal, personal use, individual, single user, one person, family
• diy, small job, one off, quick fix, handyman, odd job
• student discount, senior discount, for kids, for my dad, gift, present
• hobby, hobbyist, amateur, non commercial, side project, weekend

Image, video and download terms

Queries seeking a file rather than a service. They look industry-relevant and convert at effectively zero. This category is small, stable and worth adding once and forgetting — it is the closest thing to a genuinely universal negative block.

• image, images, photo, photos, picture, pictures
• clipart, stock photo, wallpaper, png, jpg, svg
• video, videos, mp4, download, downloads, downloadable
• pdf, ebook, e book, template download, free template, printable
• logo, icon, icons, font, mockup, psd
• app, apk, software download, crack, torrent, install file

Adult, gambling and sensitive terms as a safety net

You will rarely need this block and you will be glad it is there the one time you do. Certain everyday service words have adult or gambling homonyms, and broad match plus an unfamiliar market is enough to surface them. Treat this as insurance rather than optimization — it costs nothing to carry and prevents a category of embarrassing placement.

• porn, xxx, nude, adult entertainment, escort, massage parlor adult
• casino, betting, gambling, poker, slots, lottery
• crypto, forex signals, get rich, make money fast, mlm, pyramid scheme
• weapon, gun, ammo, drug, pharmacy online, prescription without

Competitor terms — and why not to exclude them automatically

Most negative keyword articles tell you to exclude every competitor name. I disagree, and the reasoning matters more than the conclusion. Someone searching a competitor by name is, definitionally, in the market right now for exactly what you sell. That is not waste — it is the most commercially advanced audience you will encounter all week. Excluding them removes a category of buyer, not a category of noise.

What actually deserves exclusion is not the brand name but the navigational pattern around it. Somebody searching for a competitor plus "login", "phone number", "tracking" or "cancel" is an existing customer of theirs performing an account task. That click cannot convert, and it is where competitor campaigns bleed money.

• login, log in, sign in, my account, account login, portal
• phone number, contact number, customer service, support number, help desk
• tracking, track my order, order status, invoice, bill payment, pay my bill
• cancel, cancel subscription, unsubscribe, close account, refund status
• careers at, jobs at, salary at, reviews of employees, glassdoor
• is down, outage, not working, complaints about, scam or legit

The exceptions where excluding the brand outright is correct: the competitor sells something you genuinely do not, so the query is wrong-entity rather than competitive; or your landing page cannot make an honest, specific comparison, in which case you are buying an expensive click for a page that does not answer the question. If you do run competitor terms, isolate them in their own campaign with their own budget so the economics stay visible rather than being averaged into your brand or generic campaigns.

Industry-specific negatives

The lists above are general. What actually separates a well-run account from a copied one is the industry-specific block — the terms that only make sense once you know what the business does and what else the words mean. These are starting points; the real ones come from your own search terms report.

Home and trade services

Plumbing, electrical, HVAC, roofing, cleaning, pest control, landscaping. The dominant leaks are employment queries, DIY instruction, parts and equipment retail, and training or licensing. Note that "estimate", "quote" and "inspection" often pair with "free" here and must be protected at ad group level rather than excluded. Pruning matters more in this category than in most, because the genuinely transactional half of the query space is small to begin with — the argument for that, and for how emergency and planned intent should split the account around it, is in the guide to Google Ads for local service businesses.

• parts, part number, replacement part, spare, filter replacement, fuse
• home depot, lowes, hardware store, tool hire, tool rental, machine rental
• license, licensing, licence, permit, code requirements, building code
• union, apprentice program, trade school, night class, ncca, city and guilds
• wholesale supply, plumbing supply, electrical supply, hvac supply, equipment for sale
• insurance claim, warranty claim, home warranty, landlord obligation, tenant rights
• salary, day rate, hourly charge for, how much do plumbers make, is it a good career

Healthcare and clinics

The hardest category, because symptom searches look enormously relevant and are overwhelmingly informational. People search symptoms to self-diagnose at a volume that dwarfs appointment-seeking, and a symptom query converts at a fraction of a booking query. Also strip medication research, insurance policy questions and the training queries that shadow every clinical term. Negatives only govern which searches you buy, though; what a clinic is permitted to promise once those searches arrive is a separate constraint, and that half of the problem is covered in the post on advertising healthcare without overpromising.

• symptoms, symptom checker, causes of, is it serious, contagious, side effects
• medication, dosage, mg, pill, prescription, over the counter
• home treatment, natural remedy, natural cure, herbal, does it go away, will it heal
• nhs, medicare, medicaid, insurance coverage, does insurance cover, copay
• nursing course, medical school, cpd, residency, become a, assistant course
• icd code, cpt code, billing code, medical coding, transcription, records request
• free clinic, charity care, low income, sliding scale, government hospital

Education businesses and consultancies

If you sell consulting or done-for-you services, student and self-learner traffic is the constant drain, because coursework language and professional service language are nearly identical. If you sell courses, invert this list and use the done-for-you terms as your negatives instead. For a consultancy placing students abroad, this list is doing something more specific than removing waste — it is separating parents and applicants with money to spend from students looking for free coursework help, which is the same separation worked through in the post on winning students without buying unqualified enquiries.

• assignment, homework, essay, dissertation, thesis, project report
• sample answer, model answer, past paper, question paper, solutions manual
• free course, free certification, scholarship, financial aid, fee structure
• internship, entry level, fresher, career change, portfolio examples
• ppt, slides, presentation download, notes pdf, syllabus pdf, book pdf
• salary after, job prospects, is it worth studying, best university for

Hospitality and hotels

Two structural leaks: employment queries in a sector that hires constantly, and aggregator or navigational traffic where the searcher wants a booking site rather than a property. Event and venue words also carry heavy homonym risk in this category. The aggregator terms are a decision rather than an automatic exclusion, because a property chasing direct bookings is competing with those same platforms for its own name — that trade-off gets its own treatment in the post on competing with the OTAs for direct bookings.

• jobs, hiring, housekeeping jobs, chef jobs, front desk jobs, hospitality course
• booking com, expedia, agoda, airbnb, tripadvisor, trivago
• cancel booking, change reservation, refund policy, no show policy, loyalty points
• hotel management, hotel school, hospitality degree, barista course, bartending course
• for sale, hotel for sale, franchise opportunity, investment property, hotel investment
• free breakfast recipe, buffet menu pdf, room service menu pdf, wallpaper, photos of

Real estate

Real estate keywords collide with rentals, careers, investing education and portal traffic more than almost any other category. The single biggest decision is whether you serve buyers, sellers, renters or investors, because each of the other three becomes a negative list for the one you chose. The harder half of that decision is not excluding the wrong audience but separating people who are browsing from people who are ready to transact, which is the subject of the post on buying intent in real estate campaigns.

• real estate license, realtor course, become a realtor, agent commission, agent salary
• zillow, rightmove, redfin, realtor com, property portal, listings site
• for rent, rental, sublet, roommate, room for rent, tenant rights
• foreclosure, auction, repossessed, distressed, tax lien, wholesaling course
• property tax, stamp duty calculator, mortgage calculator, affordability calculator
• floor plan template, house plan pdf, virtual tour software, photography service

Professional services — legal, accounting, marketing

Professional services keywords are dominated by people seeking free advice and by students. Both look like qualified queries in aggregate. The other pattern worth catching is form-seeking: someone looking to download a template or a filing form is trying to avoid hiring anyone.

• free advice, free consultation online, ask a lawyer free, free legal aid, legal aid
• template, form, forms, download form, sample contract, sample agreement
• do it myself, without a lawyer, without an accountant, self file, diy filing
• salary, billable hour rate, how much do they earn, is it a good career, exam
• software, best software for, quickbooks, xero, tax software, filing portal
• definition, meaning of, act, statute, section, case law, judgment text

The homonym trap

This is the waste category that survives longest in an account, because the queries do not look wrong. They contain your keyword. They read as plausible. And they belong to a different industry entirely. Homonyms are also the strongest argument for reading the search terms report as text rather than scanning it as a table — you catch these by reading, not by sorting.

• "Crown" — a dental restoration, a piece of molding, a brand of beverage, and a monarchy. A dental practice bidding on crown terms without disambiguation buys carpentry traffic.
• "Bridge" — dental again, and also civil engineering, dentistry's opposite in scale, a card game, and a section of a song.
• "Seal" — driveway and roof sealing, gasket seals, a marine mammal, and an official stamp. Sealing contractors reliably pay for the animal.
• "Blinds" — window coverings and hunting blinds, plus poker terminology. Three unrelated industries, one word.
• "Pitch" — roof pitch, a sales pitch, a sports field, and musical pitch. Roofers bidding broadly on pitch terms buy sales training traffic.
• "Cast" — a fracture cast, metal casting, fishing, and a film cast. Orthopedic clinics and foundries compete for the same word without knowing it.
• "Panel" — solar panels, fence panels, electrical panels, and a panel discussion. The last one is a conference query and converts at nothing.
• "Tint" — window tinting and hair tinting. Automotive and salon budgets quietly fund each other.
• "Screen" — window screens, screen printing, medical screening, and a display. Four industries, and all four run search ads.
• "Will" and "trust" — estate planning terms that are also two of the most common ordinary English words, which makes broad match on them close to unusable without heavy negation.

The fix is not to abandon the keyword. It is to disambiguate at the phrase level rather than the word level. Do not add "crown" as a negative if crowns are your service — add the specific disambiguating pairs you observed: "crown molding", "crown royal", "crown court". This is why homonym negatives have to come from your own report and cannot be pre-supplied by anyone, including this post.

Building your own list from the search terms report

Everything above is a starting point. This section is the actual deliverable, and it is worth more than any pre-made list because it produces the negatives specific to your account, your market and the way the auction has decided to interpret your keywords this month.

One prerequisite first, and it is the unglamorous one nobody wants to hear: if your conversion tracking is wrong, this process will produce confident, wrong decisions. You will exclude queries that converted through an untracked path and keep queries credited with conversions that never happened. Fix the measurement before you touch the negatives — the sequence for that is in the conversion tracking audit, and it takes an afternoon that will save you a quarter of misdirected exclusions.

1. Open the search terms report for a window long enough to contain real volume — a month for a busy account, a quarter for a quiet one. Too short a window and you are reading noise; too long and you are reacting to a query mix that no longer exists.
2. Sort by cost, descending. Not by impressions, not by clicks. You are looking for where money went, and a query with many impressions and no spend has cost you nothing.
3. Read the top block line by line, as sentences rather than as rows. This is the step people skip and it is the step that finds homonyms and wrong-entity queries, because those are invisible to sorting and obvious to reading.
4. For each query that is wrong, ask which of the three wastes it is — wrong intent, wrong economics, or wrong entity. Write the category next to it. This forces the judgment that determines match type and placement level.
5. Identify the shortest word or phrase that distinguishes the bad query from the good ones around it. This is the negative. If you cannot find one that does not also catch a good query, the answer is an exact-match negative on that single query, not a broad one on a word.
6. Choose the level. Is it wrong for the business, for this campaign, or only for this ad group? Default to the narrowest and promote it later if the same term reappears in other campaigns.
7. Check what the proposed negative would have blocked historically. Filter the same report by the term and read the matched queries. If a converting query appears in that filter, the negative is too broad and needs to be tightened before you save it.
8. Add the plural, the singular, and any misspellings you actually observed. Negatives do not expand to close variants, so anything you did not type is not blocked.
9. Log the addition with the date and the query that prompted it. Six months later, when someone asks why a campaign stopped matching a term, this log is the only artifact that can answer.
10. Do the inverse pass. Scan the same report for queries that converted and are not covered by any keyword you own, and add those as positive keywords. Negative work that never produces a new positive keyword is only half the job.

That last step is the one that separates this from a cleanup exercise. The search terms report is a two-way instrument: it tells you what to stop buying and what you should have been buying all along, and most people only ever read it in one direction.

A maintenance cadence

Negative lists decay. Query language shifts, seasonal intent changes, the auction reinterprets your keywords as it accumulates data, and every new campaign or landing page changes the surface you are matched against. A list built once and never revisited is a list that was correct in a market that no longer exists.

• Weekly — read the search terms report for every campaign spending meaningfully. Sort by cost, read the top block, add what is obviously wrong. This is a short task when it is done weekly and a miserable one when it is done quarterly.
• Monthly — review low-volume campaigns where a week produces too few queries to judge, and do the inverse pass looking for converting queries that deserve their own keyword or ad group.
• Quarterly — audit the shared lists themselves. Read every term and ask whether it is still true about the business. Services change; a term excluded because you did not offer something two years ago is now blocking a service line you sell.
• After every structural change — new campaign, new match type, new landing page, new service, new geography. Each of these changes what the auction will match you to, and the first two weeks after the change are when the new junk appears.
• After any noticeable drop in impression volume — check whether a recently added negative caused it before investigating anything else. It is the cheapest hypothesis to test and it is right more often than it has any business being.

The cadence itself should be written down and treated as a governance rule rather than a habit, for the same reason bid changes should be: undocumented, reactive account work produces changes nobody can explain later. The argument for that discipline is made in full in the bid governance framework, and it applies to negatives with equal force — every exclusion is a change to what the account is allowed to see.

When negatives go too far

This is the part most negative keyword articles omit, because "add more negatives" is a simple recommendation and "it depends how many" is not. Over-negating is a genuine failure mode with its own symptoms, and it is harder to diagnose than the waste it was meant to fix — waste shows up as a query you can read, while over-negation shows up as an absence you cannot.

• It starves Smart Bidding of data. Automated bidding models improve by observing outcomes across a range of queries. A campaign that has been aggressively pruned presents the model with a narrow, homogeneous slice of traffic and gives it very little to learn from, which is the opposite of what an automated strategy needs to work.
• It blocks discovery. A meaningful share of the best-performing queries in a mature account are ones nobody would have predicted at launch. Every speculative negative added before it was observed in the report is a query category that will now never be discovered, and you will never know what you excluded.
• It compounds invisibly. Negatives interact. A term added to a shared list, plus a phrase negative at campaign level, plus an ad group exclusion, can combine to suppress an entire theme, and no single one of the three looks wrong on its own.
• It shrinks campaigns into irrelevance. Past a certain point, a campaign narrowed by negatives has too little volume to produce a statistically meaningful read on anything, so every subsequent decision about it is made on noise.

The test I apply before adding any negative: can I name the query I actually saw that prompted this? If the answer is "no, but it seemed like a good idea", the term does not go in. Speculative negatives are how lists get to a size nobody can review and a shape nobody can explain.

The metric that lies

Here is what happens after a heavy negation pass, and it is worth knowing in advance because it feels exactly like success. Cost per click falls. Click-through rate rises. Impression share looks healthier. The share of spend on queries you would describe as relevant climbs sharply. Every dashboard number a manager or a client looks at improves at once.

And total conversions fall.

The mechanism is straightforward once you see it. Aggressive negation removes the expensive, competitive, high-intent queries alongside the junk — because the expensive queries are expensive precisely for the reason they convert, and they are often the ones that contain a word your negative list caught. What survives the pruning is cheap traffic, and cheap traffic is cheap because fewer advertisers want it.

Cheaper traffic is not the goal. The goal is more revenue at an acceptable cost, and the two come apart routinely. The second number that catches it: total conversions and cost per conversion, tracked before and after every negation pass, on a fixed window. If conversions fell while CPC improved, the pass went too far, and the correct response is to reverse the most recently added negatives one at a time rather than to celebrate the CPC.

This is the same distortion that makes Quality Score a poor optimization target — a metric that measures the shape of your traffic rather than the value of it will always be improvable in ways that make the account worse. The four Quality Score myths cover the mechanics of that specific trap, and the underlying error is identical: optimizing a proxy that correlates with performance until the moment you start managing it directly.

What negatives cannot fix

Negative keywords solve exactly one problem: buying the wrong queries. They do nothing about the other three reasons a search campaign underperforms, and every hour spent building lists to fix those is an hour spent on the wrong thing.

• They do not fix a landing page that does not answer the query. A perfectly-targeted click on a page that fails to convert is still a wasted click, and no exclusion will improve it.
• They do not fix a conversion signal that measures the wrong event. A campaign optimizing toward an action the business does not value will find efficient ways to produce more of it, and negatives will only make it faster.
• They do not fix an offer that is not competitive in the market you are bidding into. This is the answer nobody wants and it is sometimes the true one.
• They do not fix what happens after the lead arrives. A qualified inquiry that sits unactioned for two days is indistinguishable, in the revenue report, from a query you should have excluded.

That last one is worth taking seriously, because it is the most common place good ad spend goes to die. Negatives filter waste before the click; qualification and routing filter it after. The two are complementary and the second is usually weaker — the process for building the after-the-click half is covered in building lead scoring automation, and an account with immaculate negatives feeding a pipeline that drops leads has simply moved the waste somewhere harder to see.

Where to start

If you have never built a negative list: take the employment block, the DIY block and the image and download block from above, put them in a shared list, attach it to every search campaign, and stop. Those three are close to universally true for a service business and carry almost no risk of blocking something you wanted. Do not add the free and cheap block until you have checked whether your own offer uses the word "free".

Then do one search terms report pass, properly — sorted by cost, read line by line, categorized into the three wastes before anything is excluded. That single hour will teach you more about what your account is actually buying than any list somebody else wrote, including this one. Put the weekly review in the calendar before you close the tab, because the version of this that fails is always the version that was going to be done later.

I run the ads side of Arcetis on this exact sequence, and the pattern that holds across every account is the one this post opened with: the list is not the work. Knowing which intent each term excludes, and being able to defend it out loud, is the work — because that is the only version of the list that survives contact with next quarter.

Related reading