How to Deduplicate Keywords Across Countries Without Merging Different Search Tasks
Reconcile keyword exports from one seed across markets with separate concept, exact-query and search-task decisions. Use a worksheet to preserve sources and hold uncertain intent.
When one seed produces lists for several countries, remove exact repeats inside each country-language list, but keep a separate record for the same spelling in another market. Link related records with a shared concept ID; group them into a page task only after that market's query, result pages and business fit have been reviewed. That is the central distinction: a common concept helps compare work across countries, while an intent or page decision must be made within a country.
This tutorial starts after seed collection and before multilingual keyword-to-URL mapping. It is about reconciling discovery exports, not discovering more seeds or deciding which live URL wins. The proposed IDs and gates are an editorial workflow, not a Google-required format. The worked case below is explicitly simulated: it reports no real US or UK search result, metric, or preferred wording.
Keep four identities separate
A source seed is the phrase that started a discovery run. A concept is the business problem you want to compare between markets. A market query is one exact string paired with a country and language. A task is what a person appears to want to accomplish with that query in that market. These are different objects. Treating the source seed or translated string as the task ID makes an unverified assumption look like a content decision.
- source_seed_id: preserve the original input and why it was chosen. One seed can lead to several concepts; the seed is not the final keyword.
- concept_id: relate records for the same product area or problem across markets. Shared concept does not imply shared search intent, page type or wording.
- market_query_id: give each country-language and exact-string combination its own stable ID. The same string in US English and UK English gets two IDs. A second US export of the exact US string reuses the US ID and adds a source reference.
- task_id: assign only after review shows which user task and page type a market query supports. Two queries may share one market task, but two markets should not share a task ID for URL planning.
The multi-country keyword database guide explains how to keep candidate, observation and decision records over time. Here, the additional question is whether candidates from several runs belong to the same market task. Keep raw source files and exact strings even after a task grouping is approved; grouping is a decision, not deletion of the underlying rows.
Deduplicate in this order
First, freeze each raw export and note its provider, target country, query language, retrieval date and settings. Do not combine numbers from different countries into one volume column. If Search Console is one of the sources, its country dimension identifies where the search originated; query, page and country are distinct dimensions. Preserve the export or report reference so a reviewer can reconstruct the slice.
- Pass 1 — exact repeat in one market: match country, language and exact query string. Reuse the existing market_query_id, append the new source or observation, and leave any previous decision intact. Repeated discovery does not itself validate intent.
- Pass 2 — possible variant in one market: use case folding or spacing cleanup only to *find* candidates for review. Preserve both original strings and IDs. A different modifier can imply a different job; do not collapse it with a text-normalization formula.
- Pass 3 — related query in another market: link both rows to the same concept if the underlying business idea truly matches. Keep separate country, language, query, metric, result-page and reviewer fields. Identical spelling is not an exception.
- Pass 4 — task grouping: compare the actual problem, likely page type and target-market result evidence. Group only within one market when one page could responsibly satisfy both queries. Otherwise split or hold.
This order avoids a common spreadsheet trap: deduplicating all rows on a lowercased keyword string before market and source columns are added. The resulting list can no longer show whether a phrase was found in two countries, twice in one country, or through two independent sources. A dedup key is useful for finding work; it should not erase the evidence needed to decide what that work means.
Use an intent label as a question, not a merge command
An informational, commercial or transactional tag can help prioritize review, but it cannot by itself tell you that two strings deserve one page. Ahrefs says its intent tags may be multiple and can be wrong or change. Its clustering explanation uses similarity in results as a signal and notes that ambiguous groups still require interpretation. Those are useful clues, not a universal overlap percentage or an automatic cross-country equivalence rule.
For each candidate pair, write the user's task in plain language: compare products, buy one item, obtain a template, learn a process, troubleshoot an issue, or another action supported by the evidence. Then ask what page could complete it and whether your organization can make that page. If the task is unclear, record two hypotheses rather than choosing the label that gives the tidiest spreadsheet.
Check the exact query in the intended country using a documented search setup. Record the date, device, region or location method, visible location, relevant result URLs, page types and important limits. Open the destination pages; snippets alone may hide a different audience or task. If you only have an unlocated web-index search, label it an editorial clue, not a target-country SERP observation. Use the existing SERP evidence workflow for the repeatable one-query record. A reviewer who knows the market must confirm customer-facing local phrasing before it is used in a brief.
- Group within a market: the tasks, likely page type and business offer align, and the documented result sample supports one page job. Keep both exact queries and their evidence references under the task.
- Split within a market: the same broad concept points to materially different jobs, such as choosing software versus downloading a template. Give each job a separate task ID; do not conclude automatically that each needs a new URL.
- Hold: the result sample is mixed, the location is uncertain, the language reviewer has not resolved meaning, or your offer cannot yet answer the task. Name the next check and its owner.
- Reject for this task: the observed job is outside the planned offer or page. Preserve the query and reason so it can be reconsidered for a different task later.
A simulated three-query reconciliation
Suppose the source seed is `team planning tools`, recorded as S-01. The team uses C-01 for the broad concept of team coordination resources. The following rows are invented to demonstrate record linkage. The query strings are not recommendations for either country, and the described result patterns were not observed. Every decision remains on hold until real evidence and review replace the simulation.
- Q-US-01: `team planning software`, US English, C-01. A simulated result set contains software comparison pages, so the *hypothetical* task T-US-SOFTWARE is comparing options. Record SIM-SERP-US-01 as a simulation, not market evidence. Decision: hold for an actual US capture and reviewer.
- Q-GB-01: the same exact words, UK English, C-01. It gets a new query ID because the country differs. A simulated mixed set contains vendor and template pages; leave its task unresolved. Do not copy T-US-SOFTWARE or US evidence into the UK row.
- Q-GB-02: `team planning template`, UK English, C-01. The exact string differs from Q-GB-01. A simulated template-oriented set suggests a different hypothetical task, T-GB-TEMPLATE. Keep the query separate and hold both UK tasks until real local evidence and language review exist.
- Second US export: a simulated new source repeats the exact Q-US-01 tuple. Keep Q-US-01, append SIM-EXPORT-US-02 as another source, and do not change its hold decision merely because it appeared again.
The worksheet supplies the field-by-field version of this practice case, a blank decision row and a quality check. Its useful outcome is a reproducible *hold* and a clear path to validation, not a fabricated difference between US and UK search behavior. Replace all SIM references with dated source material before using a row for content planning.
Hand off a market task, not a global keyword
Before handing work to a writer or URL owner, verify four things: each market query has its own exact string and source; each grouped task has a country, page job and reviewable result evidence; mixed cases have an owner and next check; and a local-language reviewer has approved any phrase intended for customer-facing copy. If one item is missing, keep that task on hold rather than turning a shared concept into an approved content brief.
Send validated market tasks to multilingual keyword mapping, where an existing or proposed URL can receive ownership. A shared concept may support corresponding pages in several locales, but it does not itself establish a page relationship. Google's localized-page guidance covers how to identify alternate versions once those pages exist; it is not a keyword clustering method. Keep the concept link for comparison and the market-specific task record for the publishing decision.