Orphex Search Term Cleaning

Orphex Search Term Cleaning

Use this skill when someone asks which Search queries should be excluded, whether paid-search spend is being wasted on irrelevant demand, or how to draft a negative-keyword review. Work from user-supplied exports and documents; an already-authorized read-only data source is optional. Return a proposal, never an account edit.

Establish what the rows mean

Ask for or extract the reporting window, account timezone, currency, campaign and ad-group IDs or names, search term, matched keyword, reported match type, impressions, clicks, cost, conversions, and conversion value if available. Also request the business definition of brand terms, protected products or services, target customer, conversion action, attribution window, current negative keywords and their scopes. Preserve the export's date and segment grain. If scope, currency, or brand rules are missing, label those fields unknown and do not widen a proposal to account level.

Check whether the period is complete for conversion lag and whether the report includes only visible query rows. Search term reports omit low-activity queries for privacy; an empty or short report does not prove that no other queries served. The report's match-type column describes the query's relation to the triggering keyword and can differ from that keyword's configured match type. Keep those fields separate. Use Google's Search terms report guide for current column semantics.

Classify before proposing exclusions

For each query, classify it as clearly irrelevant, relevant/protected, or uncertain using supplied product, audience, brand, and conversion evidence. Distinguish branded from nonbrand only with the provided brand dictionary or explicit user confirmation; mark ambiguous brand variants for review. Protect any query with a conversion, meaningful value, or an explicit business protection. A zero-conversion row is not proof of irrelevance, especially with few clicks, delayed outcomes, or an incomplete attribution window.

Calculate historical exposure from the selected rows only: sum cost, clicks, and conversions over the stated period. Show the query count and period beside the total. Call this “observed spend exposure,” not recoverable savings. A negative may reduce future cost, delivery, and conversions; it cannot guarantee savings or preserve every useful variant. Do not invent a universal click, spend, or conversion threshold.

Choose match type and scope cautiously

Propose the narrowest negative that addresses the evidence. For Search negatives, broad excludes when all terms appear in any order, phrase excludes the same ordered phrase with possible surrounding words, and exact excludes only the same query without extra words. These behave differently from positive keyword matching. Negatives do not expand to close variants; singular/plural, synonyms, or other variants need separate review if they should also be blocked. See Google's negative keyword guide.

Name a campaign or ad group only when the export proves that scope. Before suggesting a shared list or account-level exclusion, inspect every supplied campaign and protected/converting query it could affect. Check existing negatives, overlapping campaigns, brand/nonbrand intent, and product names that also appear in the proposed phrase. If those checks are unavailable, label the scope “unverified” and ask for review rather than recommending a wider block.

Deliver an auditable proposal

Lead with the period, currency, campaign scope, data completeness, and any brand-definition gaps. Give one row per proposed term with: search term; class and rationale; proposed negative text and match type; proposed scope; impressions, clicks, cost, conversions/value; observed spend exposure; relevant or protected overlaps checked; blocking risks; confidence; and reviewer decision. Add a separate “keep” or “investigate” list for converting and ambiguous terms. Confidence describes evidence quality, not certainty of future impact. Tie every row to its source file, row identifier, or query and date.

Fictional example: a 30-day USD export shows “free repair manual” with $42 cost and no conversions, while “repair manual subscription” has two conversions. Propose reviewing the exact negative [free repair manual] at the demonstrated campaign scope; preserve the converting query. The $42 is past exposure, not a savings forecast, and unreported low-volume queries remain unknown.

Any later account mutation requires explicit user authorization naming the account, campaign/list scope, exact terms and match types, and action. This skill and an installed copy do not authorize changes.