English Keyword Research for Cross-Border Ecommerce: From Seed Questions to a Schedulable Map
Summary: The most common mistake in cross-border English keyword research is translating a Chinese keyword list into English. Translated terms are often not what locals actually search, so volume, intent and competition all fail to line up. The correct method starts from questions rather than words: build a seed question bank, expand, filter, group, and land on a table you can schedule from. This article gives the pass/fail test at each step.
Key Takeaways
- Do not translate a keyword list; rebuild a question bank. The same need is phrased differently in English, and literal translations usually have no search demand.
- Volume is a range, not a number. Without active ad spend, Keyword Planner shows ranges only; treating a range as precise wrecks the decision.
- Intent and funnel are two separate labels. Intent answers "what does the user want"; funnel answers "where in the decision are they". Mixing them removes the basis for scheduling.
- Filtering matters more than expansion. Expansion yields thousands of candidates; maybe two hundred are usable. Set the filter rules before expanding.
- Language and market are configured separately. The same English term differs across the US, UK and Australia; one list covering three markets means three judgments were never made.
1. Three premises: language, market, intent
The first lesson in cross-border research is not tooling but three premises.
Language is not market. English spans the US, UK, Australia, Canada, Singapore and India, and the same term carries different habits and competition in each. Treat "language + market" as one unit of analysis rather than building an "English list" and splitting it later.
Needs are phrased differently. Chinese says "which cross-border logistics provider is good"; English might say "best freight forwarder for small business" or "how to ship pallets to Europe". A literal rendering like "cross-border logistics which is good" has close to zero volume. A quick test for a translation term: put it in the search box and see whether the results are written by locals. If most results are Chinese sites or machine-translated pages, nobody uses that term.
Intent precedes form. The same root with "best", "cheap", "how to" or "vs" lands in entirely different intents, and intent determines what page type you need. Get this wrong and every subsequent page is mismatched.
Want to know how much of the ChatGPT and Perplexity answers your brand owns?
Explore TENGENCE GEO2. Four steps: seed, expand, filter, group
We do not start from a keyword tool. We start from questions.
Step one: build a seed question bank. Four sources: support and pre-sales records, internal search logs, competitor category and article titles, and questions in marketplace review threads. Record the exact wording; do not normalise it into "keyword form", because normalisation destroys the intent signal.
Step two: expand. Use tools to expand each seed into synonyms and variants. This step aims for coverage, not accuracy — nine in ten expansions will be discarded, and that is fine.
Step three: filter. The critical step, with rules fixed in advance. Usually four: drop geo-mismatches, drop intents that do not fit the page type, archive obvious brand terms separately, drop anything unrelated to the business. After filtering, typically less than a tenth of the candidates survive.
Step four: group and label. Group by topic cluster, pick one primary keyword per cluster with the rest as support, and label intent, funnel stage, target page type and priority.
| Step | Input | Output | Test |
|---|---|---|---|
| Seed | Real questions | 50–200 verbatim questions | Every entry traceable to a source |
| Expand | Seed questions | 1,000+ candidates | Synonyms and variants covered; accuracy not required |
| Filter | Candidates | 100–300 usable terms | Four rules, each auditable |
| Group | Usable terms | 20–40 topic clusters | Each cluster has a primary and support terms |
The most skipped item in that table is "rules fixed before expanding". Expand first and the team gravitates toward high-volume terms, ending with a list full of irrelevant traffic that produces content no customer reads.
One practical addition: slice the map by season as well. Cross-border seasonality differs from the domestic calendar — Western retail peaks at Black Friday and Christmas, B2B buying peaks at fiscal-year start and trade-show season. The same term can differ threefold between peak and off-season, so an annual average has no decision value. We tag each cluster with a publishing window and go live six to eight weeks ahead so the page has time to be indexed and accumulate signals.
3. Two label sets: do not mix intent with funnel
Intent answers what the user wants: informational, navigational, commercial investigation, transactional.
Funnel answers where they are in the decision: awareness, consideration, decision.
They are not the same. A commercial-investigation query can appear at awareness ("just starting to compare this category") or at decision ("choosing between two brands"). An informational query can appear at decision ("checking the returns policy before buying").
Mixing them breaks scheduling: if you sequence content by funnel but judge competition with an intent label, you end up with pure explainers scheduled into the decision stage. We fill both columns, but scheduling uses intent plus target page type, while funnel only sets content depth.
A note on TOFU/MOFU/BOFU: that vocabulary is a funnel model from marketing and inbound practice, not a search-intent taxonomy, and navigational queries belong to no funnel stage. Treating the two frameworks as one is precisely why many keyword maps look complete and cannot be executed.
4. Tools and data caveats
Three things you must know about tooling.
Keyword Planner volumes are ranges. Google's documentation and community answers both state that without active spend, volumes appear as ranges (for example 1K–10K); finer figures require spend. A range cannot support a "is this worth doing" judgment. Use it for relative size and competition, not absolute value.
Geography and language are configured separately. Keyword Planner's targeting supports country/region and language separately, plus a "nearby" expansion. For cross-border work, run one configuration per market rather than one global pass.
First-party data beats third-party estimates. Internal search queries, Search Console query data and verbatim support tickets are more trustworthy than any estimate. Third-party tools are good at discovering terms you did not know; first-party data is good at confirming that people actually use a term with you.
Measure the gap between the two yourself: place the same terms' Keyword Planner ranges beside their Search Console impressions and you will see the range bounds differ severalfold. The reason is that the former smooths and buckets by advertising conventions while the latter counts only your own impressions. Rank with the former, validate with the latter, never compare them directly.
5. Five common errors and a two-week schedule
By frequency:
- Translating a Chinese list. Covered above; most common and most damaging.
- Treating a range as an exact number. An ROI calculation on a 1K–10K range cannot be right.
- Ignoring plurals and word order. "running shoes" and "shoes for running" can carry different intents; they are not one term.
- Head terms only. Head terms are competitive and convert poorly; long-tail is where cross-border sites actually earn traffic.
- Archiving the map when it is finished. A keyword map is not a one-off deliverable. Update it quarterly with new questions, keeping the previous version so you can see which terms are new.
Scheduling: two weeks. Week one for seeds and expansion, pushing the question bank past 100 entries; week two for filtering, grouping and labelling, producing the first 20 executable clusters. From week three, schedule by cluster rather than by keyword — keyword-level scheduling produces ten articles competing for one intent.
Two reusable heuristics. If a cluster has fewer than three support terms, demand is too thin to justify its own article; merge it into a neighbouring cluster. If a cluster's primary term ranks top twenty in three markets, it deserves localized versions; if it only has volume in one market, ship one language first.
One reminder beyond tools: cross-border competition cannot be read from ad data alone. High ad competition may only mean high order value and good paid ROI, which is unrelated to organic difficulty. A more reliable organic test is the page type of the top five results — if all five are marketplace category pages, do not fight for it; if two or three are vertical blogs, there is room.
FAQ
Q: Can we use Keyword Planner without an ads account?
Yes, after completing account setup (an account without campaigns is possible), but data comes as ranges only. Finer data requires actual spend.
Q: How many markets should one keyword map cover?
Start with one or two core markets and prove the process before expanding. A map covering five markets at once usually ends up shallow in all of them.
Q: Should British and American spellings be researched separately?
Yes. "colour" and "color" have separately counted volume and competition, and pages should use the matching spelling — otherwise users perceive the page as not written for them.
Q: Do brand terms belong in the map?
Yes, archived separately. Their value is defence — keeping your own site first — not acquisition.
Q: How often should the map be refreshed?
Quarterly for the base, plus a seasonal pass before peak periods. Cross-border seasonality follows the target market's shopping calendar, not the domestic one.
Related Reading
- Content architecture for cross-border marketing
- Multi-regional site architecture: ccTLD, subdomain or subdirectory
- Multilingual content automation done properly
Data Sources
- Use Keyword Planner — Google Ads Help — how volume ranges are produced and what they represent.
- Refine your new keywords in Keyword Planner — Google Ads Help — geography and language targeting options.
- Create a Google Ads account without a campaign — Google Ads Help — account setup without active spend.