How to Use Exact, Phrase, and Broad Match in Google Ads

I learned paid search when exact match felt exact, phrase match followed a recognizable phrase, and broad match was the setting you watched with both hands on the controls.

One early broad-match lesson was painfully simple: an MBA program could show for people searching for NBA scores because Google decided NBA might be a typo for MBA. The platform’s confidence did not make the traffic useful.

That mental model is obsolete.

Google now matches meaning and intent—according to them, not only literal strings. Exact match can include close variants, synonyms, paraphrases, reordered words, and searches Google considers to have the same intent. Phrase match reaches searches that include the meaning of the keyword. Broad match can reach searches related to the keyword while using additional signals that the keyword text cannot express by itself.

In one current account I manage, broad match is producing a substantially lower cost per lead than exact match ever did. I would not turn that observation into “broad is always better.” The broad-match campaign sits inside an unusually governed system: extensive negative-keyword processing, dozens of automated scripts, and an AI-assisted search-term sweep across the accounts I watch at least weekly.

Broad match did not remove the need for control. It changed where the control lives.

The Difference Between Exact, Phrase, and Broad Match

Google’s current keyword matching documentation describes the match types as overlapping layers. Phrase match can reach everything exact match can reach plus additional searches. Broad match can reach the searches available to phrase and exact plus a wider set of related queries.

Microsoft Advertising uses a similar spectrum, but its documentation makes the operational risk easier to say out loud. Microsoft describes broad match as eligible for related searches, synonyms, spelling errors, and other semantic variations, and its API documentation notes that query expansion can sometimes match irrelevant queries. That is the practical point: both platforms sell broader matching as machine-assisted reach, but the advertiser still owns the waste, routing, compliance risk, and cleanup.

Exact Match

Syntax: [keyword]

Exact match gives the most steering and the narrowest reach of the three positive match types. An ad may show when the query has the same meaning or intent as the keyword.

That is narrower than broad match. It is not a literal character-for-character lock.

Google’s close-variant documentation says exact match can include misspellings, singular and plural forms, reordered words, implied words, synonyms, paraphrases, and searches with the same intent. You cannot opt out of close variants.

Use exact match when:

the intent is valuable and already understood;
the budget or conversion volume requires tighter steering;
the query needs a specific ad, page, location, or business rule;
you want a controlled starting point before testing broader discovery;
an identical query should receive keyword prioritization inside the account.

Exact match can still produce a surprising search-term report. Review it.

Phrase Match

Syntax: “keyword”

Phrase match has changed enough that I do not treat it as the old middle ground anymore. Historically, phrase match meant the words had to appear together in a recognizable order. Today, it may show ads on searches that include the meaning of the keyword. The search can be more specific and does not need to repeat the words exactly as entered.

That makes phrase match much less useful as a control mechanism than it used to be. If the platform can decide that a different wording carries the same meaning, phrase match is no longer “exact plus modifiers.” It is a softer intent-matching setting with a narrower ceiling than broad match.

I still use it sometimes, but I do not trust it by default. Phrase match can be useful when the core product or service meaning must remain present but the searcher can add meaningful context such as price, location, audience, feature, or problem.

Use phrase match when:

exact match is too restrictive but broad match is not yet supported by enough conversion evidence;
modifiers help reveal useful demand around a stable service or product concept;
the account needs a transitional test between narrower control and broader discovery;
the team can review new query patterns frequently.

Phrase is not “exact with extra words.” It also uses meaning, which is why it is close to useless if the account owner expects it to preserve literal wording.

Broad Match

Syntax: keyword with no brackets or quotation marks

Broad match may show ads for searches related to the keyword, including searches that do not contain its direct wording. Google can consider landing pages, other keywords in the ad group, the user’s previous searches, location, and other signals.

Google recommends pairing broad match with conversion-based Smart Bidding and responsive search ads. The bidding system can then evaluate more auctions and adjust the bid using auction-time signals. That wider opportunity set is why broad match can sometimes produce more conversions at a lower CPL than an exact-only structure.

Use broad match when:

conversion tracking represents the outcome the business wants;
the campaign has enough trustworthy signal for the bidding strategy;
the offer and landing-page scope are clear;
negatives and routing rules protect known boundaries;
search terms are reviewed on a real cadence;
budget can absorb exploration without creating a business emergency.

Broad match is not a substitute for knowing what the company sells. It gives the system more auctions in which to make a prediction. If the conversion event is wrong, it can find more of the wrong outcome efficiently.

What Happened in My Current Account

The account had historically relied on narrower matching for control. Broad match now produces a lower cost per lead than exact match did.

That result became possible only after control moved beyond the positive keyword list.

The operating system around the account includes:

dozens of scripts that monitor or enforce defined query and account rules;
extensive negative-keyword libraries and historical exclusion records;
a weekly AI-assisted sweep across every account I watch;
recurring review of what broad match admitted;
protected terms and conflict checks intended to prevent useful demand from being blocked.

Across the working data, review history, and exclusion queues supporting this work, I have processed more than one million negative-keyword records and decisions. That does not mean a single Google Ads campaign holds one million active negatives. Google currently documents limits of 10,000 negative keywords per campaign, 5,000 per shared negative list, 20 shared lists per account, and 1,000 account-level negatives. The larger number describes the governance workload and historical records around watched accounts, not one active list.

This distinction belongs in the article because scale can sound impressive while hiding the real question: are the active exclusions still necessary, correctly scoped, and safe?

Broad Match Did Not Win by Itself

My observation supports a narrower conclusion than the usual platform claim:

Broad match can outperform exact match when it receives trustworthy conversion signals and operates inside a disciplined query-governance system.

Several parts of that statement need to remain visible.

First, cost per lead is not automatically cost per qualified lead. The broad-match result should be reconciled with CRM outcomes, sales acceptance, revenue, customer type, and lag.

Second, the negative system changes the available query population. This is not a clean comparison between a default broad keyword and a default exact keyword.

Third, the reporting window changes which exclusions appear safe. In one cross-platform review, a short lookback produced a large negative queue. Extending the history and grouping reordered query phrases into normalized intent families preserved dozens of families that had meaningful older conversion evidence. After both the historical check and a human intent review, only a small minority of the original candidate families remained defensible exclusions.

That finding is not an argument for keeping every old query. It is an argument against allowing a recent zero to become a permanent rule without checking conversion lag, seasonality, sparse volume, and reordered variants. The negative keyword sweep documents the full review boundary.

Third, Smart Bidding, ads, pages, budgets, targets, competitive conditions, and historical learning can all change the result. Match type is one part of the causal system.

I would not copy the keyword into another account, switch it to broad, and expect the same CPL.

How Negative Keywords Work

Negative keywords prevent an ad from serving when the search matches the exclusion rule. For Search campaigns, Google supports negative broad, phrase, and exact match—but negative matching behaves differently from positive matching.

Negative keywords do not expand through close variants the same way positive keywords do. Google states that singulars, plurals, synonyms, and related forms may need separate exclusions. Casing and misspellings are handled, but a negative for flower does not necessarily block flowers.

This helps explain why mature negative libraries can become large. It also makes careless automation dangerous.

Negative Exact Match

Syntax: [negative keyword]

The ad is blocked when the search is the exact negative phrase without additional words. This is the narrowest and safest starting point for one observed irrelevant query.

If [free marketing course] is negative exact, that precise query is blocked, but free marketing course for nonprofits may remain eligible.

Negative Phrase Match

Syntax: “negative keyword”

The ad is blocked when the search contains those terms in the same order, even if additional words appear before or after them.

Use negative phrase when the ordered concept is consistently irrelevant and the surrounding modifiers do not make it valuable.

Negative Broad Match

Syntax: negative keyword with no brackets or quotes

The ad is blocked when every word in the negative appears in the search, in any order. Some of the words can appear without triggering the full exclusion.

Negative broad does not mean “block everything related to this concept.” It still follows the negative words. That difference is easy to misunderstand if someone assumes positive and negative broad match expand in the same way.

How to Add Negative Keywords in Google Ads

The current Google Ads workflow is:

open Campaigns;
open Audiences, keywords, and content;
select Search keywords;
open the Negative search keywords tab;
choose the campaign or ad group scope;
add one negative per line using brackets, quotes, or no formatting for the intended match type;
optionally save the entries to a new or existing negative keyword list;
review overlap warnings and save.

For a reusable list, open Tools, then Shared library, then Exclusion lists and the Negative keyword lists tab. Create or update the list and apply it only to campaigns that share the same exclusion boundary.

Account-level negatives apply across eligible Search and Shopping inventory. Reserve that scope for terms that are genuinely unwanted everywhere. A query that is wrong for one campaign may be valuable to another.

Google’s instructions for adding negatives also warn that a negative overlapping a positive keyword can prevent the ad from serving. Preview the blast radius before applying a phrase or broad rule.

Where Should a Negative Live?

Use the narrowest scope that represents the business decision.

Ad group: route a query away from the wrong message while preserving it elsewhere.
Campaign: exclude an intent, product, geography, audience, or business rule from one campaign.
Shared list: apply a stable theme across selected campaigns.
Account level: block an intent that is invalid across eligible inventory for the entire account.

Do not put every discovery into a global list. Routing and rejection are different actions.

A search for a supported service entering the wrong ad group should usually be rerouted. A search for a service the company never provides may deserve a wider exclusion.

How I Use AI in the Weekly Sweep

An AI agent can read far more search-term rows than a person can review manually. It can group recurring concepts, normalize variants, find cross-account repetition, flag positive-negative conflicts, and prepare proposed text, match type, and scope.

The agent should not receive unlimited authority to invent and apply exclusions.

My preferred workflow is:

preserve the raw search-term export;
attach the triggering keyword, source match type, campaign, ad group, page, cost, conversions, and value;
check conversion lag and downstream outcome;
classify the query as irrelevant, relevant, uncertain, or misrouted;
propose the smallest negative text and supported match type;
compare the proposal with positive keywords, existing negatives, and protected terms;
route uncertain or high-blast-radius candidates to human review;
retain the decision, rationale, scope, and date.

Automation can safely enforce known rules. For example, a reviewed global exclusion dictionary or a deterministic prohibited-service rule can be applied consistently. Open-ended semantic judgment needs a review boundary because one plausible-looking exclusion can quietly remove valuable demand.

The more negatives the system has processed, the more important removal and consolidation become. A weekly sweep should look for obsolete and conflicting exclusions as well as new ones.

Negative Keywords Do Not Directly Raise Quality Score

I originally described the negative system as helping keep Quality Score up. The practical relationship needs more precise wording.

Negatives can remove irrelevant searches, improve the query mix, and create a better connection among intent, ad, and landing page. Those changes can improve performance and may eventually affect the historical signals reflected in Quality Score.

The exclusion itself is not a direct Quality Score control. Quality Score is calculated from expected CTR, ad relevance, and landing page experience for exact-search history. Diagnose those components directly instead of crediting every score movement to the negative list.

A Better Match-Type Test

Do not compare broad and exact using two unrelated date ranges and call the difference causal.

Define:

the same business conversion and value rules;
the same customer and geography boundaries;
a conversion-based bidding strategy appropriate to the test;
a stable budget and target;
expected learning and conversion lag;
primary CPL, qualified CPL, revenue, or profit outcome;
guardrails for irrelevant-query cost and lead quality;
a search-term review cadence;
which negative rules both variants share.

Use a campaign experiment when feasible, and avoid changing ads, pages, tracking, budgets, and match type simultaneously. Google recommends experiments to isolate automated bidding and targeting changes from seasonality and other account movement.

Even with an experiment, inspect the queries. Two variants can report similar CPL while acquiring very different customers.

My Current Rule for Match Types

Exact match is a steering tool. Phrase match is a controlled expansion tool. Broad match is a prediction and discovery tool that needs trustworthy measurement and governance.

None is automatically the “advanced” choice.

I use the match type that creates the best business outcome within the control system the account can actually maintain. In the current account, that is broad match—and the lower CPL is real enough to change how I operate. The lesson is not to abandon exact match. The lesson is that exact syntax is no longer the only place where control can live.

Continue with the full negative keyword sweep process for the evidence fields, cross-platform differences, and review queue. The paid search systems trail connects matching with Quality Score, conversion governance, ROAS, and downstream outcomes.