google-ads-ppc-waste-finder

name

google-ads-ppc-waste-finder

description

Use this skill any time the user wants to find wasted PPC spend, build a negative keyword list, or audit search terms for a Google Ads / Microsoft (Bing) Ads / Yahoo Ads account. This is a specialized workflow with a built-in brand/competitor guardrail that prevents the catastrophic auto-negative mistake — recommending the advertiser's own brand or a competitor-conquesting term as a negative keyword. Without this skill, Claude often correctly identifies brand and competitor terms but still puts them on the recommended negative list anyway, which can shut down the advertiser's own brand traffic and undo a deliberate competitor strategy. Always prefer this skill over generic ad-account help when the task is search-term-level waste detection. What it does: pulls the last 7 days of search terms via the Optmyzr MCP, filters to terms that spent over a threshold (default $20) with zero conversions, classifies each by intent (Waste / Review—competitor / Protected—brand), groups Waste into themes, runs a deterministic monthly-savings projection from the bundled scripts/estimate_savings.py, and outputs a ranked table plus a paste-ready negative keyword list. Works for ONE named account per run. Lookback, spend threshold, and conversion threshold are all configurable per run. Trigger this skill on phrases like "weekly waste finder", "find my wasted spend", "search term waste report", "negatives I should add", "what queries should I block", "what's burning my [ad] budget", "where am I losing money in [account]", "build me a negative keyword list", "search terms that didn't convert", "clean up [my] search terms", "garbage queries eating my budget", "audit search terms", "run the weekly waste check", "find waste in [account]". Also triggers on recurring/scheduled requests of the same nature. Do NOT use this skill for full account audits (campaign structure, Quality Score, audiences, Performance Max, conversion tracking — defer to the sibling google-ads-audit skill), ad copywriting, auction insights, vertical benchmarking, conversion tracking setup, or multi-account performance rollups. This skill is specifically the one-account, search-term-level waste pass with the guardrail.

Google Ads PPC Waste Finder

What this does and why

Search campaigns accumulate waste quietly: a keyword matches a query that looks related but never converts, and it keeps spending week after week until someone looks. This skill does that looking on a regular cadence. It pulls the last 7 days of actual search terms, isolates the ones that spent real money and returned nothing, separates true waste from spend that only looks like waste, and hands back both a prioritized table and a list you can paste straight into Optmyzr or Google Ads as negatives.

The single most important idea: a zero-conversion search term is not automatically waste. Brand terms and deliberate competitor-conquesting terms routinely show zero conversions over a short window and are still doing their job. Recommending them as negatives is actively harmful. The guardrail step below exists to catch exactly that, and it is the part of this skill that makes it trustworthy enough to run unattended.

Inputs

Ask the user for the account name only if it isn't already in the request. Everything else has a sensible default:

• Account: the Google Ads account name (e.g., "Optmyzr Google Ads").
• Lookback: 7 days ending yesterday (default).
• Spend threshold: $20 (default). Terms at or below this aren't worth the noise.
• Conversion threshold: exactly 0 conversions (default).

If the user gives different numbers ("over $50", "last 14 days", "fewer than 2 conversions"), use theirs.

Step 1 — Resolve the account

Find the account ID with the Optmyzr ads MCP get_active_accounts tool, passing the account name as searchQuery and account_type: "adwords" (use bing or yahoo if the user names those platforms). Confirm you matched the right one if the search returns more than one similarly named account — pick the exact name match and note the ID you used.

Step 2 — Pull the wasteful search terms

Compute the date range as the 7 full days ending yesterday, formatted yyyyMMdd,yyyyMMdd. Then call the Optmyzr ads MCP get_ppc_report tool with:

• accountId: the ID from Step 1
• platform: adwords (or the platform named)
• report_name: SearchTermPerformance
• date_range: the computed range
• order_by: Cost, order: DESC
• limit: 200, page: 1
• numeric_filters: [{"field":"Cost","operator":">","value":20},{"field":"Conversions","operator":"=","value":0}] (substitute the user's thresholds if they differ)

Push the filtering into numeric_filters so the MCP returns only candidate rows — don't pull everything and filter by hand. If page 1 comes back full (200 rows), fetch page 2 to be sure you have the complete set; in practice the threshold keeps this short.

Each row gives you: Search Term, Match Type, Matched Keyword, Ad Group, Campaign, Campaign Type, Conversions, Cost, Clicks, Impressions, ROAS. Keep the Campaign name — the guardrail in Step 3 depends on it.

Step 3 — Apply the intent guardrail (do this before grouping)

For every flagged term, classify it by the intent of the campaign it spent in. Read references/grouping-and-guardrails.md for the full logic and edge cases. The short version:

• Brand campaigns (campaign or ad group name signals brand — e.g. contains "Brand", "BR", or the advertiser's own name, and the matched keyword is the brand or a misspelling of it): never recommend as a negative. This is the advertiser's own name. Zero conversions here is almost always a tracking or short-window artifact. Mark as Protected — brand.
• Competitor / conquesting campaigns (name contains "Competitor", "Competitors", "Conquest", or the term is a known rival brand and it's matching deliberately): flag for human review, don't auto-negative. Paying for competitor traffic is often a deliberate, long-payback play. Mark as Review — competitor.
• Everything else (generic, product, category, research-tool, or clearly off-topic queries in non-brand/non-competitor campaigns): genuine waste candidate. These are what the negative list is built from. Mark as Waste.

Only Waste rows flow into the paste-ready negative list. Protected and Review rows still appear in the ranked table (so the user sees the full picture) but are labeled and excluded from the negatives.

Step 4 — Group the waste into themes

Cluster the Waste rows into a handful of human-readable themes by what the searcher actually wanted, not by string matching. Typical themes: "free / DIY tool seekers", "job seekers", "wrong product or category", "informational / how- to", "geographic mismatch". A theme can be a single high-spend term if it doesn't fit with others. For each theme, identify the shared token(s) that would make a good phrase negative (e.g., the theme "free keyword tools" shares "keyword planner").

Step 5 — Estimate monthly savings

Run the bundled helper to project savings deterministically (so the math is consistent every run):

python3 scripts/estimate_savings.py <rows.json>

Write the flagged rows to a small JSON file first (the script's --help documents the shape). It returns weekly and projected monthly waste per theme and in total, using monthly = weekly_spend * 30.4 / lookback_days. Projected savings assume the negative fully eliminates the term's spend going forward — a reasonable upper bound; say so in the output. Only Waste rows count toward projected savings; Review/Protected spend is reported separately as "spend to review", never as savings.

Step 6 — Output

Produce two things, in this order.

A. Ranked waste table, sorted by 7-day cost descending, one row per flagged term:

| Search term | Theme | Campaign | Match type | Clicks | 7-day cost | Est. monthly | Status |

Status is one of Waste, Review — competitor, Protected — brand. Below the table, give a one-line total for projected monthly savings (Waste only) and a separate line for monthly spend flagged for review.

B. Ready-to-paste negative keyword list — Waste rows only. Default to exact-match negatives of the exact search term (safest, won't over-block), and where a theme has a clear shared token, also offer the phrase-match negative for the theme.

Critical formatting rule — keep the code blocks paste-ready. Each fenced code block in this section must contain ONLY the negatives, one per line, with nothing else: no inline # comments, no rationale, no per-line annotations, no trailing notes inside the fence. A user must be able to triple-click → copy → paste the entire block into Optmyzr or Google Ads without first stripping out commentary. All explanation goes in prose outside the code blocks. Use two separate blocks (one for exact, one for phrase), not one combined block.

Use this exact structure:

Exact negatives (one per term):
[google keyword planner]

Phrase negatives (theme-level, broader — review before applying):
"keyword planner"

If you want to flag a specific phrase negative as needing extra caution (e.g., "free" could over-block a legitimate "free trial" query), say so in a sentence after the code block, not inside it. One short sentence per caveat is plenty.

Close with a one or two sentence summary: total projected monthly savings, how many terms are clean waste vs. need review, and a reminder that brand/competitor terms were deliberately excluded.

Before the user applies anything — remind them to dedupe against existing negatives

After producing the paste-ready list, include this single line:

Before pasting, take 30 seconds to check the account's existing shared negative lists (Google Ads → Tools → Shared library → Negative keyword lists; or Optmyzr → PPC Solutions → Negative Keywords). Any term already on a list applied to the same campaign can be skipped.

This is currently a manual pre-step. Once the Optmyzr MCP exposes a negative-keyword listing (a NegativeKeywordPerformance report option or a dedicated endpoint), the skill should fetch the current negatives and dedupe automatically. See the skill's v0.2 roadmap in Notes.

Running on a schedule

If the user wants this weekly, offer to set up a scheduled task (e.g. Monday morning) that runs this skill on the named account and delivers the table + negative list. Keep state lightweight — each run is independent and just looks at the trailing 7 days.

Notes

• Spend and conversions are in the account's currency and conversion settings; don't convert.
• If zero rows clear the threshold, say so plainly — that's a good week, not an error.
• Never apply negatives automatically. This skill recommends; a human applies.

v0.2 roadmap

• Auto-dedupe against existing negatives. When the Optmyzr MCP exposes a way to list current negatives (shared lists, campaign-level negatives, ad-group negatives), fetch them in a new Step 1.5 and filter the recommended list to drop any term already negated for the same campaign. Until then, the skill includes a single-line manual-check reminder in Step 6.