Match types: what each one actually does
Broad match shows your ad for any search Google considers related — including synonyms, implied concepts, and tangentially related queries. It reaches the widest audience but gives you the least control. Over the last few years broad match quality has improved significantly with Smart Bidding, but it still requires an aggressive negative list to avoid waste.
Phrase match shows your ad when the search contains the meaning of your keyword, in any order. It balances reach and precision well. Exact match shows your ad only for searches that match the keyword closely — it limits reach but maximizes relevance and conversion intent.
- Exact match: use for your highest-converting, most specific terms — brand, product name, direct commercial intent
- Phrase match: use for terms where the core concept must be present but word order varies
- Broad match: use only with Target CPA or Target ROAS and a well-developed negative list
- Avoid mixing match types for the same keyword in the same ad group — it creates internal competition and confuses attribution
Mapping keywords to search intent
Not all searches with your keyword in them indicate buying intent. "Google Ads tutorial" and "Google Ads agency" both contain the phrase "Google Ads" — but one is a learner and one is a potential customer. Building your keyword list around intent stages rather than just topics is what separates efficient accounts from wasteful ones.
Commercial intent keywords include price, cost, buy, hire, agency, software, tool, and near-me modifiers. Informational intent keywords include how-to, what is, guide, and example. Target commercial intent keywords with conversion-focused campaigns and either exclude informational queries or put them in separate awareness campaigns with lower bids.
Building a negative keyword list before you spend
Most accounts build negatives reactively — adding them after wasted spend shows up in the search terms report. A better approach is to build a starter negative list before the campaign launches based on known waste patterns for your vertical.
Common negative themes for B2B and SaaS: free, jobs, reviews, tutorial, how to, salary, template, definition, vs (when you do not want comparison traffic), and competitor names (unless you are bidding on competitor keywords intentionally).
- Add generic negative keywords at the account level so they apply to all campaigns
- Add vertical-specific negatives at the campaign level
- Review search terms weekly and add new negatives as the campaign accumulates data
- Ask Claude to suggest negatives from your search term report: "Show me search terms from last 30 days that are clearly informational or irrelevant and suggest negatives"
Structuring ad groups around theme tightness
Each ad group should contain keywords so closely related that every keyword can be covered by the same set of ads. Wide ad groups — ten unrelated keywords pointing to the same RSA — produce low Quality Score and poor ad relevance.
A practical rule: if you cannot write one RSA where the main keyword appears naturally in multiple headlines, the keywords do not belong together. Split the ad group.
Ongoing keyword hygiene
Keyword strategy is a continuous process. The most impactful recurring task is weekly search-term review: pull terms with significant spend and zero conversions, add negatives, and promote any high-performing search terms that are not already in your keyword list as exact match.
Monthly, review keyword-level Quality Score and CPA. Pause keywords with QS = 4 and below-average conversion rate — they are raising your costs without contributing results.
Doing this with Claude connected to the account
Keyword strategy is normally done in two places that do not talk to each other: Keyword Planner for research and the search terms report for what actually happened. With Google Ads connected through MCP Ads, both are available to Claude in the same conversation, alongside the ability to make the change.
Prompt 1: "Give me keyword ideas around commercial roofing in Texas with monthly volume, competition, and top-of-page bid, then group them by intent: informational, comparison, and ready-to-buy." Claude calls google_ads_keyword_ideas, which returns Keyword Planner data, and google_ads_keyword_forecast for the expected clicks and cost at a given bid. The grouping by intent is the part the planner will not do for you. As an illustrative example, you might get 120 ideas back, of which 25 are commercial-intent phrases worth bidding on, 40 are how-to queries that belong in a negative list, and the rest are too broad to use.
Prompt 2: "Show my current keywords by match type with spend, conversions, and CPA for the last 30 days, and list the search terms each broad match keyword is actually matching to." This uses google_ads_keyword_performance and google_ads_search_terms. What comes back is the honest picture of broad match in your account: which broad keywords are finding profitable new queries, and which are matching to nonsense. That is the evidence for tightening or loosening match types.
Prompt 3: "Add the 25 commercial-intent keywords to the Commercial Roofing ad group as phrase match, switch the two worst broad match keywords to phrase, and add the how-to cluster as campaign negatives." Claude uses google_ads_bulk_create_keywords, google_ads_update_keyword_match_type, and google_ads_add_negative_keywords. All three are writes, so Claude lists every keyword, match type, and destination and waits for approval before applying any of it. If the keywords need a new ad group, google_ads_create_adgroup builds it and it stays paused until you enable it.
The keyword research and long-tail expansion skills package the first two prompts into a repeatable planning session.
Where the manual way still wins
Keyword Planner data through the API is the same bucketed volume estimate you see in the UI, so it is directional, not precise. And the search terms report hides low-volume queries behind a privacy threshold, which means the broad match analysis is always partial in smaller accounts.
Intent classification is also easier for a human who knows the business. Claude will label "roofing contractor license" as informational, and it usually is — unless you sell exam prep. Review the groupings before approving the writes. Ad group structure is a similar case: theme tightness depends on how you want the ads to read, and that is a copy decision before it is a data decision.
