Search-Term Miner & Negative-Keyword Generator

Your search-term report is where budget quietly leaks and your next winning keyword hides in plain sight. This prompt reads the whole thing, ranks the waste by dollars, and hands you exact additions, negatives, and a list to file them in.

What this isA prompt that turns a pasted search-term report into a sorted action list: keywords to add, terms to negate, and terms to review, organized by intent and by wasted spend.

You give itA pasted Google or Microsoft Ads search-term report and one line on what you sell.

You get backExact keyword additions with match types, a structured negative-keyword list, high-spend zero-conversion terms flagged to cut, and a weekly cadence.

Paste this into Claude, ChatGPT, or any capable model. Fill in the bracketed parts, paste your report under the prompt, and it does the rest. It works only from the numbers you paste, never invents data, and says so when a column is missing.

You are a senior paid-search analyst. I will paste a search-term report from [Google Ads / Microsoft Ads]. Your job is to decide, term by term, what to add as a keyword, what to negate, and what to leave for review. Sort the work by buyer intent and by wasted spend so I act on the dollars first.

WHAT WE SELL
- Offer in one line: [what the product or service is and the core problem it solves].
- Who buys: [ICP or audience in one line. Example: RevOps leaders at mid-market B2B SaaS].
- What is NOT us: [products, audiences, free or DIY intent, jobs, competitors, or regions we do not want. List anything you already know is off-target].
- Goal of this campaign: [leads / demo requests / purchases / app installs]. Target CPA or max acceptable cost per conversion: [number, or "unknown"].

THE REPORT
- I am pasting the search-term report below this prompt. Columns may include: search term, match type, campaign, ad group, impressions, clicks, cost, conversions, conv. value, CTR, CPC.
- The period covered is: [last 7 / 30 / 90 days, or "unknown"].

HARD RULES (follow these exactly)
1. Work ONLY from the rows I paste. Do not invent search terms, numbers, or trends. Do not pull in outside keyword ideas. If you want to suggest expansion ideas, put them in a clearly separate "Ideas (not from the data)" section at the very end and label them as guesses.
2. Never fabricate a metric. If a column I need is missing (for example no cost or no conversions column), say "not enough data" for that judgment and tell me which column to add to the export. Do not estimate it.
3. Never negate a term that has conversions, even one, without flagging it for human review first. A converting term is never an automatic cut.
4. Quote real numbers from the rows when you justify a call (cost, clicks, conversions). Do not round in a way that hides small spend.
5. If the report is mostly one or two days of data or under ~100 total clicks, open with a one-line warning that the sample is thin and that I should wait for more volume before negating aggressively.

WHAT TO PRODUCE, in this order:

A. SUMMARY (3 to 5 lines)
- Total cost, total clicks, total conversions across the pasted rows (sum only what is present).
- Total spend on terms with zero conversions (the waste pool).
- The single biggest issue you see in one sentence.

B. NEGATE NOW. Ranked by wasted spend, highest cost first
For each: the term, its cost, clicks, conversions (should be 0), why it is off-target, the negative I recommend, the match type for that negative ([exact] / "phrase" / broad), and the level to add it at (account / campaign / ad group / shared list). Prioritize high-spend zero-conversion terms. Group obvious throwaways (clearly free, DIY, jobs, wrong product, wrong region) so I can bulk-add them.

C. REVIEW. Terms that need a human call
Terms that converted but look loosely related, terms with spend but too few clicks to judge, ambiguous intent, or anything close to but not clearly inside our offer. For each, give the term, its numbers, the specific question I should answer, and the two possible actions depending on my answer.

D. ADD AS KEYWORDS, by intent tier
Split into:
   - High intent (ready to buy / our exact product): propose each as [exact] match, with the converting or strong term it came from and its numbers.
   - Mid intent (problem-aware, comparison, category): propose as "phrase" match, with a note on which ad group it belongs in.
For each addition, name the existing ad group it fits or say "new ad group: [suggested name]" if it deserves its own. Only propose adds that are supported by clicks or conversions in the pasted data.

E. STRUCTURED NEGATIVE-KEYWORD LIST
Design a reusable negative structure, not just a flat dump. Output as:
   - Shared list "[Brand / Universal Junk]". Terms that should be blocked across every search campaign forever (free, torrent, jobs, salary, login, reviews of competitors, etc. that apply to us). Give the list name and the negatives with match types.
   - Campaign-level negatives. Terms wrong for this campaign but maybe fine elsewhere.
   - Ad-group-level negatives. Terms that should be steered to a different ad group rather than killed (note which ad group they belong in instead).
Explain in one line why each tier exists so I can keep maintaining it.

F. CONFLICT CHECK
Flag any term you are about to negate that also appears as a winner I want to keep, so I do not block my own converting traffic. If you cannot tell from the data, say so.

G. WEEKLY CADENCE (4 to 6 bullets)
A repeatable routine I can run every week: how often to pull the report, what date range to use, the order to work the sections (waste first), the spend or click threshold below which a zero-conversion term is not worth negating yet, and when to graduate a converting search term into its own keyword. Make it specific to the volume you saw in this report.

Be direct and specific. Quote the numbers. Where the data cannot answer, say "not enough data" and name the missing column instead of guessing.

PASTE THE SEARCH-TERM REPORT BELOW THIS LINE:

When to use it

Run it on every active search campaign on a weekly rhythm, or any time CPA creeps up and you suspect the query mix has drifted. It is built for Google and Microsoft Ads search-term reports, not display or shopping.

How to use it

Export the report with cost and conversion columns included, then paste it under the prompt so the model can rank by real dollars. Treat section B as ready to apply, but always sanity-check section C and the conflict flags before you bulk-negate, because a single converting term cut by accident costs more than the waste you saved.