For most of the paid search era, the B2B playbook was addition. You added keywords, added audiences, added budget, and trusted the platform to sort out who deserved to see your ad. That worked when Google's match types did what their names implied and a "phrase match" keyword actually required the phrase. It does not work now. Google's own documentation describes broad match, the default, as the type where ads can show on searches related to your keyword, including searches that do not contain the keyword's direct meaning. Phrase match has since absorbed most of what the old broad match modifier used to do, and exact match now includes close variants. The names still say "exact" and "phrase." The behavior is looser than either word suggests.
The result is a discipline that has quietly inverted. The lever that decides whether B2B ad spend is efficient is no longer only what you target. It is what you exclude. In an EMARKETER survey conducted in partnership with Demandbase, 58% of B2B marketers agreed that ad waste is a significant issue, and 52.4% estimated that between 16% and 45% of their total ad budget is wasted on irrelevant accounts. "Spray and pray," in other words, is not a strategy anymore. It is a leak, and by the industry's own estimate it is draining a meaningful slice of every B2B budget.
This article is a technical teardown of the two mechanisms that seal it: negative keyword constraints and audience exclusions. We will get into match-type behavior, account structure, the real limits Google enforces, how to automate the discovery of waste with n-gram analysis instead of eyeballing a report, and how B2B teams extend the same logic to LinkedIn and Meta where a wasted impression costs far more. This is the "powered by AI, led by humans" version of budget defense: let the machine surface the waste, keep a human deciding what is actually waste.
Why B2B Bleeds Worse Than B2C
Before the mechanics, it helps to understand why B2B is the harder case. The math is unforgiving. B2B search terms are ambiguous by nature, because the same words that describe your product also describe things that will never buy it. "CRM" pulls in students writing papers, job seekers, and people who want a definition. "Project management software" pulls in the person comparing free personal to-do apps. Every one of those is a real click you paid for, and in B2B the click is expensive: business and industrial keywords sit among the higher average cost-per-click categories in Google Ads benchmark data, while B2B click-through rates tend to run below flashier consumer verticals. You pay more per click and fewer of them are qualified.
Then there is the funnel. A B2C conversion often happens in the same session as the click. A B2B conversion happens weeks later, across a buying committee, through a form fill that a human then has to qualify. That lag breaks the feedback loop that Google's automated bidding depends on. Google advises that Smart Bidding performs best with sufficient conversion data, and for value-based strategies specifically it recommends at least 15 conversions in the last 30 days, because thinner data makes for noisier optimization. Many B2B accounts never generate that volume on genuine sales-qualified events. So the algorithm optimizes toward the cheap, plentiful conversions it can see (newsletter signups, ebook downloads, low-intent demo requests from students) and away from the rare, expensive ones that actually become pipeline. Left alone, the system gets more efficient at buying the wrong outcome.
Negative constraints are how you correct that bias from the outside. You cannot make Google's model care about your pipeline. You can make it stop spending on the queries and audiences you already know are dead ends, which forces whatever budget remains toward the searches that matter.
Negative Keywords Are Not the Inverse of Keywords
The first thing to internalize is that negative match types do not behave like positive ones, and the differences are the whole game. Positive match types have loosened relentlessly over the past several years. Negative match types have not. Critically, Google confirms that negative keywords do not match close variants, so a negative for "flight" will not automatically block "flights." That literalness is deliberate, and it is what makes negatives predictable. Understanding the three types is the difference between sealing a leak and thinking you sealed it.
Negative broad match blocks a search only when it contains all of your negative terms, in any order, but Google notes your ad may still show if the search contains only some of those terms. Unlike positive broad match, it does not expand to synonyms or close variants. If you add "free" as a negative broad, you block queries containing "free," but that reach is exactly why it is dangerous for a word that appears in both wanted and unwanted contexts, since it can catch "free trial" and "free consultation" along with the junk.
Negative phrase match blocks searches that contain your full term in the order you specified, and per Google the search can include additional words on either side and the ad still will not show, as long as all the terms appear in that order. This is the workhorse for theme cleanup: broad enough to catch variations, narrow enough not to nuke adjacent intent.
Negative exact match blocks only the precise query, in the exact order, with no extra words. It is for surgical removal of a single query without disturbing everything around it. Useful when one specific term is bleeding and the variations near it convert.
Here is the operational nuance that costs B2B accounts real money. When you add negatives directly in Google Ads, the interface lists broad as the default match type, alongside exact and phrase. But because negatives are literal and never expand to variants, a too-narrow choice means the same irrelevant query can reappear tomorrow with one extra word attached, while a too-broad choice can silently block searches you actually wanted. Most advertisers pick without thinking about which shape of waste they are dealing with, so they either under-block and keep paying, or over-block and throttle good traffic. The rule of thumb for B2B, where intent is muddy, is to reach for phrase on theme-level cleanup and reserve exact for precision cuts. In essence: match the constraint to the shape of the waste, and remember that Google will not expand a negative for you the way it expands a positive.
The Structure That Makes Constraints Scale
Adding negatives one at a time does not scale to a real B2B account. The architecture matters, and Google gives you several places to hang exclusions, each with a different scope and a hard limit worth committing to memory.
At the widest level, an account-level negative keyword list is a single global filter that, in Google's words, automatically applies to all Search and Shopping inventory in relevant campaign types, including Performance Max. Google caps this at 1,000 negative keywords per account, so it is not where you put everything. It is where you put universal junk: the terms that signal a user wants information rather than a purchase, or a job rather than a vendor, things like "definition," "login," "wikipedia," and "jobs." Every B2B account should have this filter populated before it spends a dollar, because it is one of the few ways to apply bulk exclusions to Performance Max at the account level.
Below that sit shared negative keyword lists, built in the Shared Library. Unlike the account-level filter, these do not apply automatically. You attach them to the specific campaigns you choose, which is exactly what you want for exclusions relevant to some campaigns but not others. Google's documented limits are up to 5,000 negative keywords per list and up to 20 lists per account. For a multi-product B2B company, this is the backbone: a "competitor names" list, a "wrong industry" list, a "job seeker" list, each mapped to the campaigns where it belongs.
Then there are campaign-level and ad-group-level negatives for the surgical work: excluding "blue widget" queries from your "red widget" campaign, or keeping an informational term out of one tightly themed ad group while allowing it elsewhere.
Performance Max deserves its own note because it was, for a long time, the black box where budget went to disappear. That changed. Google now documents that you can add negative keywords directly to Performance Max campaigns, or apply them at the account level, to keep those campaigns from showing on queries with those terms across Search and Shopping inventory. The negatives you add there primarily affect Search and Shopping, so YouTube and Display waste is controlled through placement and content exclusions rather than keywords. The important shift is philosophical: PMax is no longer a channel you simply trust. It is one you constrain like any other.
Automate the Discovery, Not the Judgment
The manual version of this work is a person opening the Search Terms report every week, reading rows, and adding negatives by hand. It does not scale, and it is reactive by design: you can only block a wasteful query after it has already spent your money, and you are blocking exact strings that recur forever in new variations. There is a better way, and it is the part most B2B teams skip.
The foundational technique borrows from computational linguistics: n-gram analysis. An n-gram is simply a contiguous sequence of n words, a concept drawn from natural language processing and formalized in the standard academic reference on the subject, Jurafsky and Martin's Speech and Language Processing, and taught as a foundational NLP tool by outlets like the ACM. Instead of treating each search term as a unique string, you break every query into its component word sequences (single words, two-word pairs, three-word phrases) and analyze performance at the pattern level. The payoff is coverage. If your report shows "cheap CRM tool," "cheap CRM software," and "cheap CRM app" all spending with zero conversions, n-gram analysis surfaces "cheap" as the common problem, and one carefully scoped negative eliminates dozens of variations at once, including future ones you have not seen yet. It turns negative keyword management from whack-a-mole into pattern defense.
You do not have to run this by hand. Google Ads scripts, which are written in JavaScript, can automate reporting and repetitive account tasks on a schedule, and the platform's Search Terms report is the documented starting point for finding queries to exclude. A scheduled routine can scan that report, flag terms with meaningful spend and no conversions, and surface prioritized candidates for review, which is a far better use of an analyst's time than manual row-reading.
Here is the reassuring part, and the reason we say automate the discovery, not the judgment. Automation is superb at surfacing candidates. It is bad at knowing that a recurring non-converting term is not actually waste, but a signal that you need a different landing page or a separate, better-targeted campaign for that intent. Not every underperforming n-gram deserves to be negated. A script cannot make that call. A marketer who understands the business can. This is the whole "powered by AI, led by humans" thesis in one workflow: the model does the reading, the human does the deciding.
One more guardrail belongs in every B2B account: watch for negative keyword conflicts. It is entirely possible to add a negative that silently blocks a keyword you are actively bidding on. If you bid on "free project management solutions" but have "free" sitting in a negative list, your ad simply will not show, which is why Google publishes guidance on fixing issues with negative keywords, including the reminder that negatives do not match variants, so you must exclude the exact terms you mean. Aggressive exclusion without checking for conflicts is how teams accidentally throttle their own best terms.
The Other Half: Excluding People, Not Just Queries
Negative keywords solve the search half of the problem. They do nothing for the enormous share of B2B budget that flows through audience-based platforms, and this is where the "spray and pray" instinct does the most damage, because on these platforms you are not paying per relevant click. You are paying per impression to a targeting bucket, and the buckets are full of people who will never buy.
In Google Ads, the primary tool is Customer Match. You export your customer list from your CRM and upload it, and Google's documentation confirms Customer Match lets you use your online and offline data to reach and re-engage customers across Search, the Shopping tab, YouTube, Gmail, and Display, where the same list can be applied as an exclusion. For B2B, this extends naturally to excluding closed-lost accounts, active opportunities that sales is already working, and anyone who has churned. Two practical constraints matter here. Match is never perfect: Google reports most advertisers see match rates between 29% and 62%, because not every uploaded record can be tied to a logged-in user. And the lists decay: Customer Match lists have a maximum membership duration of 540 days, and Google recommends refreshing them regularly, including through continuous syncing with third-party CRMs. A customer list uploaded once and forgotten quietly stops protecting you.
LinkedIn is where the economics make exclusion non-negotiable. The platform offers the most precise professional targeting available, but it charges for it. Independent 2026 benchmark data puts B2B CPMs in North America and Western Europe in the range of roughly $55 to $110, with highly specialized C-suite audiences pushing far higher. At those rates, every impression served to the wrong person is money you will not get back. LinkedIn's own Campaign Manager is built for this: its documentation describes an Exclude audience step that lets you narrow reach by excluding audience attributes or Matched Audience lists, and trade coverage notes that by controlling which lists you include and exclude, you can even stagger exposure across the funnel.
The exclusion targets on LinkedIn are specific and worth naming. Common practice is to exclude students, interns, and unrelated roles that click but do not buy, job titles below your buyer's seniority, and whole departments that do not influence the deal, such as HR titles when you sell to engineering. Competitors are the easy win: LinkedIn ad specialist AJ Wilcox notes that the simplest way to keep competitors from seeing your ads is to add their company names as an exclusion, either individually or as an uploaded company list applied across all campaigns. The most valuable habit is to review the demographics data monthly and exclude any job title or company segment carrying real spend and zero conversions. This is the audience-side twin of n-gram analysis: find the segment that spends without converting, then cut it. The point is not that exclusions lift your click-through rate in isolation. They remove whole categories of spend that should never have happened, which makes everything else in the account more efficient.
The connective tissue that makes all of this durable is CRM integration. A customer list exported once and forgotten decays fast in B2B, where accounts close and churn every week, and as noted above Google's own lists expire on a rolling 540-day basis. The teams that get this right keep exclusion lists automatically synced from the CRM across Google, Meta, and LinkedIn, so a newly closed deal stops seeing acquisition ads without anyone touching a campaign. Static exclusions are a snapshot. Synced exclusions are a system.
Key Takeaways
- The lever moved from targeting to exclusion. Positive match types have loosened to the point that Google's own docs describe broad match, the default, as showing on searches that may not contain your keyword's direct meaning. What you exclude now determines B2B efficiency as much as what you target.
- Negative match types are literal on purpose. They do not match close variants, so they behave predictably. Default to phrase for theme cleanup, exact for surgical cuts, and be careful with broad-match negatives on mixed-intent words.
- Structure the constraints by scope. Account-level lists cap at 1,000 keywords for universal junk; shared lists give you up to 20 lists of 5,000 each for themed exclusions; and Performance Max now accepts negatives directly at the campaign and account level.
- Automate discovery with n-grams, keep judgment human. N-gram analysis, a foundational NLP technique, blocks patterns instead of strings and covers variations you have not seen yet. Scripts can surface candidates from the Search Terms report, but some non-converting terms are a signal to restructure, not exclude, and only a human should make that call.
- Watch for negative keyword conflicts. An over-eager exclusion can block a keyword you are paying to rank on, which is why Google publishes guidance on fixing negative-keyword issues to catch it.
- Exclude people, not just queries, and sync from the CRM. At LinkedIn's roughly $55 to $110 B2B CPMs, every wasted impression is expensive. Use Customer Match and LinkedIn's exclusion audiences to cut students, junior titles, wrong departments, competitors, and current customers, and keep those lists synced, since Google's Customer Match lists expire on a rolling 540-day basis.
Adapt the Account or Keep Funding the Leak
The uncomfortable truth of B2B paid media in 2026 is that the platforms are optimized for the platforms. Google's match types get looser, Performance Max wants your trust, and every default is tuned to spend more of your budget rather than less. None of that is malicious. It is just the incentive structure you are operating inside, and it means efficiency is no longer something you can buy by targeting well. It is something you have to engineer by excluding well.
That is the difference between "spray and pray" and a disciplined paid growth system. One adds and hopes. The other builds a set of automated constraints (n-gram-driven negative keyword lists, structured exclusions across the right scopes, CRM-synced audience exclusions on every platform) and then puts a human in the loop to make sure the machine is cutting waste and not pipeline. The accounts that do this do not just spend less. They redirect the same budget toward the searches and the buyers that actually convert, which is what makes every other optimization in the account work harder. Solving for cheap clicks is solving for 2020. The goal is qualified pipeline, and the fastest path to it is to stop paying for everything else.
If your B2B account is still built on addition, that is the leak, and it is fixable. Deviate Labs builds paid growth systems that combine disciplined targeting, automated negative-keyword constraints, and CRM-fed audience exclusion to turn ad spend into a clear path toward scalable acquisition. If you want a team to audit where your budget is bleeding and engineer the constraints that stop it, see our paid media services.