Major

Google Ads Negative Keyword Agent: Weekly Search Terms Review

Learn how to use a Google Ads negative keyword agent for weekly search terms review, decision support, and precise keyword management.

Ritvij Gautam
marketing team reviewing campaign results (Google Ads AI agents and automations)

A Google Ads negative keyword agent can streamline your weekly search terms review by identifying irrelevant queries, suggesting negative keywords with supporting spend data, and routing proposals for human approval. This process cuts wasted spend and ensures ad campaigns focus on qualified traffic.

This guide covers one Google Ads process end to end. For the Google Ads API guide, connection setup and the full list of Google Ads use cases, see More Google Ads use cases.

Why This Process Breaks Today

Search terms reviews are essential to keep your Google Ads spend targeted, but manual execution is slow, error-prone, and often ignored until a crisis. Typically, a marketer pulls the search terms report from Google Ads, filters by impression or spend, skims hundreds of entries, copies suspicious queries to a spreadsheet, justifies each, drafts negative keyword suggestions, and then (sometimes) implements changes. This workflow breaks down in a few ways:

  • Inconsistency: Different marketers may review search terms in ways that miss edge cases or generate redundant negatives.
  • Delays: Weeks may pass before unnecessary spend is caught, since the work is tedious and often deprioritized.
  • Poor Hand-off: Evidence for why a query should be a negative is rarely documented or easy to review.
  • Risk: Hasty work can mean blocking profitable terms or leaving irrelevant ones active.

The Systems and Data Involved

Several tools and specific Google Ads APIs are central to this process:

  • Google Ads Search Terms: Use the search_term_view object via GAQL (Google Ads Query Language) to retrieve queries users searched that triggered ads, along with metrics like impressions, clicks, cost, and conversions.
  • Google Sheets: For collecting candidate negative keywords and audit trails.
  • Slack: For reviewing and approving negative keyword proposals.
  • Google Analytics: To cross-reference with on-site engagement or conversion data for each query.
  • Google Ads API endpoints:
  • SearchGoogleAdsStream for retrieving search term data.
  • NegativeKeyword and MutateNegativeKeywords to apply negatives at campaign or ad group level.

Code vs Agent Judgment vs Human Approval

Each stage of the process involves different responsibilities:

  • Code:
  • Automates report extraction with GAQL.
  • Fetches the last 7 days of query metrics using the account's credentials.
  • Writes raw result rows to a Google Sheet and optionally matches with Google Analytics events using UTM parameters.
  • Tracks any actions taken (e.g., posting to Slack, applying negatives) for auditing.
  • Agent Judgment:
  • Reads each search term and associated metrics (e.g., spend, CTR, conversions).
  • Proposes negative keywords only for queries with sufficient spend and low relevance, referencing evidence (such as zero conversions or high bounce rate in Analytics).
  • Documents a rationale for each suggestion, not just a keyword list.
  • Human Approval:
  • Final review in Slack or another review queue.
  • Approves or rejects the agent’s proposals, possibly adjusting negative match types or levels.
  • Only approved negatives are sent back to Major to execute the API call.

Worked Example: Walking One Search Record Through the Process

Suppose your company sells organic dog food. Here’s how a weekly process might look for a single search term:

  • GAQL Query: Fetch search terms triggered for Campaign X over the past week:

SELECT search_term, campaign.id, ad_group.id, metrics.impressions, metrics.clicks, metrics.cost_micros, metrics.conversions
FROM search_term_view
WHERE campaign.id = 1234567890
AND segments.date DURING LAST_7_DAYS

  • Sample Row Returned:
  • search_term: goose food organic
  • campaign.id: 1234567890
  • ad_group.id: 1357
  • impressions: 405
  • clicks: 23
  • cost_micros: 5100000 (equals $5.10)
  • conversions: 0
  • Sheets Sync: This row gets pushed to Google Sheets along with calculated metrics (e.g., cost per click, conversion rate).
  • Agent Judgment:
  • Recognizes "goose food organic" is not relevant to an organic dog food offer.
  • Notes zero conversions across 23 clicks and $5.10 spent.
  • Proposes a negative keyword "goose" at the ad group level, with note: "No conversions, unrelated animal, wasted $5.10 this week."
  • Slack Message: Agent posts to the #negative-keyword-review channel:
  • Search term: "goose food organic"
  • Evidence: $5.10 spend, 23 clicks, 0 conversions
  • Recommendation: Apply negative keyword [goose] to ad group 1357
  • Rationale: Term is for a different animal, outside our product scope.
  • Human Approval: Marketer reviews and clicks Approve in the Slack workflow.
  • Major Workflow: On approval, the system uses the MutateNegativeKeywords API to apply the negative keyword. Action is logged for future audits.

How to Build It on Major

The process runs best as three Major objects working together:

App: Fetch and Store Search Terms Data

  • Connectors: Google Ads (service account or user OAuth), Google Sheets.
  • Scheduled fetch: Once per week, run a GAQL query targeting search_term_view for selected campaigns or ad groups and output to Google Sheets.
  • Optionally match: Pull engagement metrics from Google Analytics for queries with UTM tracking.

Agent: Judge Irrelevance and Draft Suggestion

  • Reads data: From the same Google Sheet.
  • Business logic:
  • Flag any query above a spend or click threshold with low/no conversions or high bounce.
  • Categorize the type of mismatch (e.g., wrong animal, off-topic intent, etc).
  • Draft negative keyword suggestions at ad group or campaign level, matching to context and noting spend evidence.
  • Compose review: Formats evidence and recommendation for human review.

Workflow: Approval and Execution

  • Queue: New suggestions sent as messages or tasks to the reviewer on Slack.
  • Human step: Marketer approves, rejects, or modifies each suggestion directly in Slack.
  • On approval: Workflow triggers Major to call MutateNegativeKeywords on Google Ads.
  • Logging: Every action (agent recommendation, human approval, final API call) is written to a summary sheet and audit log connector.

What to Measure

  • Wasted Spend Prevented: Track cumulative spend on queries that become negatives after the process.
  • Change Lead Time: Time from when a suspect search term appears to when it is negated.
  • Rejection Rate: How many agent suggestions are not approved by humans.
  • Search Term Volume: Queries reviewed each cycle, to tune thresholds for future automation.
  • Conversion Rate Uplift: Shift in campaign metrics after implementing new negatives.

Guardrails: What the Agent Must Never Do

A dependable agent needs hard boundaries to avoid risky automation. Guidelines include:

  • Never apply, edit, or remove negative keywords without explicit human approval. All execution happens only after a marketer signs off.
  • Never suggest negatives for queries with conversions or those with ambiguous, possible alternative relevance. Safety over breadth.
  • Never suggest phrase or broad match negatives when an exact match is needed. Always pair match type logic with recommendations.
  • Never communicate confidential information about ad accounts, budgets, or private notes outside of internal review and audit tools.
  • Never retry failed changes without logging and alerting a human. All errors require manual intervention.

Frequently asked questions

What data does the agent use from Google Ads for its negative keyword suggestions?

The agent analyzes fields from the Google Ads search_term_view such as search_term, impressions, clicks, cost (micros), and conversions. It uses these metrics to assess relevance and identify costly, low-performing queries.

How does the agent ensure negative keywords are only applied after approval?

The workflow sends all negative keyword proposals with evidence to a human reviewer, usually via Slack or a review queue. Only after the marketer approves does the system call the Google Ads API to apply the negative keywords.

Can the process incorporate Google Analytics data for better judgment?

Yes, by matching search terms with Analytics session data (using UTMs, for example), the agent can also reference bounce rates or on-site conversion metrics for each query, leading to more informed suggestions.

Is auditing and error-handling built into the process?

Yes, all actions from data pulls to agent suggestions, approvals, and API calls are logged in connected audit systems. Any API error or failed change is flagged for a human to review before retrying.

Start building on Major

Connect Google Ads, describe the process, and Major builds the app that holds its state, the agent that makes the judgment call and the workflow that runs it, with scoped access and an audit log from the first deploy.

Frequently asked questions

What can Google Ads AI agents do with my ad data in Major?
They can ingest, analyze, and act on your Google Ads data using the official API. You can use them to automate reporting, flag anomalies, or trigger follow-up across your other systems.
Does Major replace my Google Ads account or campaigns?
No, Major orchestrates your Google Ads data and operations. It never replaces your Google Ads account or built-in automation features.
What permissions do I need to connect Google Ads to Major?
You need a Google Ads developer token and an OAuth client for secure API access. Your access level determines which operations you can perform.
How does automation with Major differ from Google Ads's built-in rules?
Built-in rules automate actions within Google Ads, while Major lets you build workflows that use Google Ads data alongside other tools for reporting, escalation, and human approvals.
Can agents in Major update or pause my ads automatically?
Yes, agents can programmatically update, pause, or review ads based on your criteria and workflow definitions, using the Google Ads API and with proper approvals if required.