How to Mine Search Terms and Cut Wasted Google Ads Spend with Claude (2027)

Connect Google Ads to Claude via MCP and GAQL to isolate zero-conversion queries and AI Max expansions, cluster negative keywords, and stage landing-page fixes.

As seen in

  • Fox News
  • ABC News
  • CBS
  • NBC
  • Associated Press
  • USA Today
  • Bloomberg
  • HuffPost
Otto the beaver next to the Google Ads logo
On this page7

Connect Google Ads to Claude through Otto's hosted MCP server or Google's read-only Python google-ads-mcp server to query search_term_view via GAQL, isolate zero-conversion terms and AI Max expansions (AI_MAX_KEYWORDLESS and AI_MAX_BROAD_MATCH), and group wasted queries into campaign or account negative keyword lists. Querying filtered rows via Otto MCP's 227 tools across 15 integrations cuts Claude token usage by 80–90% compared to 3,000-row CSV uploads, while keeping Google Ads read-only and gating site or ad edits behind human approval.

Copy-paste Claude prompts to mine Google Ads search terms (start here)

Paste these five prompts directly into Claude once your Google Ads account is connected via MCP:


Prompt 1: Zero-conversion search-term waste audit
"Query search_term_view for customer ID [YOUR_CUSTOMER_ID] over the last 30 days. Return the top 30 search terms where metrics.conversions = 0 and metrics.cost_micros > 50000000, ordered by metrics.cost_micros DESC. Include search_term_view.search_term, metrics.cost_micros, metrics.clicks, and metrics.impressions."

Prompt 2: AI Max expansion audit
"Run a GAQL query on search_term_view for customer ID [YOUR_CUSTOMER_ID] filtering on segments.search_term_match_source IN ('AI_MAX_KEYWORDLESS', 'AI_MAX_BROAD_MATCH') over the last 30 days. Group the terms by match source, list the spend and conversions for each query, and flag terms with zero conversions."

Prompt 3: Root n-gram negative clustering
"Take the zero-conversion search terms from the previous queries and extract recurring 1-word and 2-word root n-grams. Group the queries under these root tokens. Then format two separate negative keyword lists: one for campaign-level NEGATIVE_KEYWORDS shared sets, and one for account-level ACCOUNT_LEVEL_NEGATIVE_KEYWORDS shared sets."

Prompt 4: Paid vs. organic search cannibalization check
"For the top 10 search terms by spend in my Google Ads query, check Google Search Console searchAnalytics.query over the same 30 days. Identify any queries where my site already ranks in organic position 1.0 to 3.0 with an organic CTR above 20%, and flag where paid spend may be duplicating high-intent organic clicks."

Prompt 5: Landing-page GA4, speed, and conversion diagnostic
"For the three highest-cost search terms with below-average conversion rates, extract their landing page URLs. Query read-only GA4 for sessions and key-event or conversion rates on each URL over the last 30 days. Then run psi_analyze_url on each destination to test mobile Core Web Vitals (LCP, CLS, INP) and Lighthouse scores. Tell me whether the waste comes from poor query intent, weak on-page conversion, or slow page performance."

Manual Search Terms CSV exports vs. Google's Python MCP vs. Claude + Otto MCP

The table compares three ways to pull Google Ads search terms into Claude.

MethodToken consumption & performanceChannel coveragePost-analysis landing page execution
Manual Search Terms CSV exportBurns 40,000 to 80,000 tokens per turn when uploading 3,000-row filesDisconnected Google Ads data only; no live Search Console, GA4, or site metricsManual; you open WordPress or GitHub separately to edit copy
Google google-ads-mcp (Python)Aggregated GAQL queries reduce token usage; requires local Python setupStrictly read-only Google Ads reporting across 3 tools according to the Google Ads MCP Server documentationNone; limited to Google Ads data retrieval
Claude + Otto MCP (https://otto.islaintel.com/api/mcp)80–90% lower token usage via filtered GAQL queries over HTTP227 tools across 15 integrations, including Google Ads, GA4, Search Console, and PageSpeed InsightsDirect landing page drafts via WordPress (29 tools) or GitHub pull requests with preview

To bypass manual CSV exports and query live reporting directly, connect Claude to Otto MCP using your organization's bearer token.

Step 1: Query search_term_view and isolate AI Max expansions with GAQL

Use Google Ads Query Language (GAQL) on the MCP search endpoint instead of pasting thousands of rows into the context window. The Google Ads API Reporting Overview documents GAQL against resources like search_term_view via GoogleAdsService.SearchStream or GoogleAdsService.Search.

To get started, follow the Connect Google Ads to Otto setup guide to link your account once via OAuth. Once connected, instruct Claude to send a GAQL string that pulls only non-converting search terms above a minimum spend threshold:

SELECT
  search_term_view.search_term,
  search_term_view.status,
  segments.search_term_match_source,
  metrics.impressions,
  metrics.clicks,
  metrics.cost_micros,
  metrics.conversions
FROM search_term_view
WHERE segments.date DURING LAST_30_DAYS
  AND metrics.conversions = 0
  AND metrics.cost_micros > 25000000
ORDER BY metrics.cost_micros DESC

This request pulls exactly the rows you need, avoiding the 40,000 to 80,000 tokens consumed by pasting a raw 3,000-row spreadsheet.

Next, isolate queries generated by automated expansion. Advertisers can isolate AI Max for Search campaigns in search_term_view by filtering on segments.search_term_match_source, as detailed in the Google Ads API AI Max reporting guide. The match source returns two critical values:

  • AI_MAX_KEYWORDLESS: Matches derived from your landing page, assets, or site content without an exact keyword match.
  • AI_MAX_BROAD_MATCH: Matches expanded through broad-match algorithms.

If irrelevant page-derived matches keep appearing, review locking Google Ads AI Max landing pages.

Step 2: Group wasted queries by root n-grams into campaign and account negative lists

Adding hundreds of individual exact negative keywords bloats account architecture without preventing related variations from triggering ads. Instead, instruct Claude to parse the non-converting search terms and group them into shared 1-word and 2-word root tokens (n-grams).

Say your software campaign wastes spend on "free billing template", "billing courses online", and "how to learn billing". Claude can surface shared roots like "free", "courses", and "learn" for broad-match negatives instead of one exact negative per query.

Once root terms are isolated, structure them into Google Ads API shared sets. According to the Google Ads API Shared Sets guide, shared negative keyword lists use a SharedSet resource with two distinct configurations:

  1. Campaign-level negative sets: SharedSetType NEGATIVE_KEYWORDS, attached via CampaignSharedSet on Search, Display, Shopping, MULTI_CHANNEL App, and LOCAL campaigns.
  2. Account-level negative sets: ACCOUNT_LEVEL_NEGATIVE_KEYWORDS via CustomerNegativeCriterion (at most one list per account) for universal disqualifiers such as "jobs" or "free".

Step 3: Cross-check search terms against Search Console, GA4, and PageSpeed Insights, then stage the fix

Zero conversions do not always mean bad query intent. The destination may cannibalize organic rankings, fail on-page conversion, or load slowly on mobile.

First, check organic cannibalization. The Google Search Console searchAnalytics API guide documents searchAnalytics.query by query, page, device, and country. With Connect Google Search Console to Otto, tell Claude to run gsc_search_analytics on your highest-spend paid terms. Strong organic positions on the same query may mean paid spend duplicates free clicks—route gaps to the Otto AI SEO Agent.

Second, check post-click conversion on each landing page. GA4 is read-only in Otto MCP, with no spend or configuration changes. Tell Claude to query GA4 for sessions and key-event or conversion rates on each high-spend term's destination URL over the same 30-day window as your Google Ads pull. Strong sessions with weak conversions on a relevant query point to an on-page problem; strong GA4 conversion rates with zero Google Ads conversions suggest attribution or URL mismatch before you add negatives.

Third, test mobile destination performance with psi_analyze_url (no user Google API key required). Poor mobile Core Web Vitals on the paid URL can leak conversions before you add negatives. See how Core Web Vitals drift causes paid conversion leaks.

Fourth, stage landing-page fixes via the Otto AI Website Manager. Otto opens a WordPress draft or GitHub pull request with preview in the same workflow—not a live publish without approval:

  • WordPress: 29 MCP tools; new posts and pages default to draft.
  • GitHub and Vercel: Site changes open as pull requests with preview orchestration.

Enrique Gomez, Owner of Enzo Maverick Pilates and Experiences Sensory, said: "Otto has been pivotal for us. The ease of making website changes on the fly lets us sharpen our brand and revise as fluidly as we need."

Read-only safety, approval gates, and how Otto compares to Opteo, Synter, Ryze AI, and Madgicx

In Otto MCP, Google Analytics 4, Google Search Console, Google Ads, Google Business Profile, YouTube, Google Sheets, Shopify, and PageSpeed Insights are strictly read-only, matching Google's read-only Google Ads MCP Server documentation. Claude cannot change bids, negatives, pauses, or budgets in Google Ads through MCP.

Where writes exist, Otto gates them: LinkedIn Ads default to DRAFT, Meta requires reviewer or admin, Facebook Pages and Instagram publish live (unlike draft-first WordPress), and site or ad changes ship only after owner approval.

  • Opteo: Google Ads monitoring and bid or negative suggestions only—no Meta, SEO, website edits, or Claude MCP server.
  • Synter: Paid media connectors across 40+ networks for Claude, ChatGPT, and Cursor—no website or technical SEO deploys.
  • Ryze AI: Paid ads and SEO workflows; Otto also manages GitHub, WordPress, and Google Business Profile updates (Ryze AI alternatives in 2027).
  • Madgicx: Meta creative analytics plus rule-based bid and budget automation for e-commerce; Otto can stage landing-page fixes in the same chat.

Otto is the Otto AI Paid Ads Manager on your team: routine site, search, and ad reporting with strategy and approval in your hands.

Plans, honest limitations, and frequently asked questions

Otto Webmaster ships a 227-tool MCP server across 15 integrations; MCP is today's supported path (Otto CLI coming soon). Pricing matches core Otto—no separate MCP seat:

  • $1 Trial (3 days): A 3-day trial of Pro that includes website updates, SEO, and read-only access to ads and socials, with a homepage redesign included built from your existing words and photos.
  • Pro Plan: $199/month (or $99/month billed annually). Includes website updates, SEO maintenance, and read-only access to paid ads and socials.
  • Master Plan: $3,500/month (or $3,000/month billed annually). Adds approval-gated write access to paid ads and socials, with human approval required on every change. Cancel anytime.

Check current Otto Trial, Pro, and Master pricing for complete tier breakdowns.

Honest limitations

  • Read-only Google Ads: Google Ads via Otto MCP is strictly read-only. You must import suggested negative keyword lists into Google Ads Manager or Google Ads Editor manually.
  • Write access tiering: Full read-and-write execution across paid ads (Meta and LinkedIn) requires the Master plan; Pro and Trial are read-only for advertising channels.
  • Integration scope: Otto supports 15 live integrations. Unannounced third-party integrations like Wix, Webflow, or TikTok are not supported in the live catalog.
  • No ranking promises: Otto structures sites, schema, and content so search engines and AI assistants can read the business clearly, and does not promise Map Pack placement, specific rankings, or ChatGPT citations.
  • Agency portals: White-label agency portals remain limited.

Frequently asked questions

How do I connect Claude to Google Ads without exposing API secrets? You connect Google Ads once inside Otto via OAuth. Your AI client holds only an organization-scoped bearer token prefixed with otto_mcp__ over HTTP at https://otto.islaintel.com/api/mcp. OAuth secrets never leave Otto's server.

Can Claude accidentally pause my Google Ads campaigns or increase my budget? No. Google Ads access in both Otto MCP and Google's open-source google-ads-mcp is strictly read-only. Write endpoints do not exist on the Google Ads integration.

Why query GAQL through MCP instead of uploading a CSV? A 3,000-row Search Terms CSV can consume 40,000 to 80,000 tokens per turn; filtered GAQL via Otto MCP cuts usage by 80–90%.

Stop wasting ad spend on irrelevant search terms and disconnected landing pages. Connect Claude to Otto MCP to audit your Google Ads queries and stage destination fixes in a single conversation.

Written by

Guillermo Marti

Creative Director & Head of Growth at IslaIntel

Leads Otto's brand, creative, and growth, from paid ads and SEO to the AI workflows behind them.