SEO Keyword Research Guide: From Discovery to Page Ownership

Updated: V1.3 (2026-11-03, adding evidence labels, data-quality controls, page-owner decisions, and an anonymized B2B case study)
Intended use: A reusable keyword research guide for B2B, cross-border e-commerce, SaaS, and independent websites
Scope: Research methodology and decision rules, not a ranking guarantee, permission to redesign a site, or a manual for any single tool

1. Objective and Definition of Done

Keyword research is not about exporting as many keywords as possible. It is about establishing a decision chain: business problem → user search needs → keyword clusters → one primary page owner → actionable work → performance validation.

Before starting, establish five things:

1. What products or services the business actually offers, and what it cannot provide.
2. The intended buyers and meaningful conversions (for example, qualified B2B inquiries rather than traffic alone).
3. Target countries, languages, and devices.
4. Existing URLs, page types, and commercial journeys.
5. Data sources, spending limits, and whether GSC or GA4 data is already available.

Report completion in distinct stages:

Stage	What qualifies as completed	What it does not mean
A. Seed terms and candidates	Main business lines, product synonyms, applications, purchasing terminology, and question terms are covered	Every possible user query has been found
B. Metrics and SERPs	Key markets have actual tool metrics with retrieval dates and representative SERP checks	Search volume equals actual clicks; SERPs remain unchanged
C. Page mapping	Retained keywords are clustered and assigned an existing/proposed URL, or a documented HOLD reason	Pages have already changed or rankings improved
D. Page implementation	Actual pages were checked, updated, and verified	Search engines have indexed them
E. Business outcomes	First-party GSC/GA4/lead data can support an evidence-based assessment	Causality has been established

It is possible to finish A–C as research and planning while D/E remain unfinished. State this explicitly. A tool returning no rows, an unconnected account, or an exhausted quota must never be reported as a zero value or a PASS.

2. Audit the Website Before Buying Keyword Data

For an existing website, export or list canonical URLs, page titles and H1s, product categories, product detail pages, solutions, technical resources, applications, and blog pages. Record whether each page still exists, is indexable, has real product or technical evidence, and offers a working inquiry or conversion path.

Follow this order: existing pages → target buyer tasks → seed terms. This prevents creating thin pages for every new keyword variant.

Distinguish overall brand positioning from short-term keyword priorities. The homepage should communicate the complete business, while a flagship product line can receive dedicated acquisition priority. Do not allow a single high-volume term to redefine the entire site.

3. Generate Seed Terms from Real Business Capabilities

Gather different kinds of terms for each business line:

Angle	Examples (illustrative, not live metrics)
Product names and synonyms	LED shelf lighting / shelf lights
Technology and construction	COB LED strip / magnetic rigid bar / 24V
Applications	gondola shelf lighting / supermarket display
Purchasing modifiers	manufacturer / supplier / OEM / custom
Selection questions	how to install / wiring / vs / compatibility
Customer vocabulary	Wording taken from real inquiries, sales emails, specifications, and support records

Verify that the product or service can genuinely be supplied. Do not treat nonexistent complete shelving systems, inductive wireless power, or universal cross-series compatibility as valid seed concepts.

Do not mass-translate English seed terms and treat the translations as local SEO decisions. Validate professional terminology and Google SERPs separately in each target market.

4. Collect Data Without Rebuying the Same Research

Recommended order:

1. Existing research: Check whether the same market, keywords, and tool already have data from the last 30 days, with dates and limitations recorded. Reuse valid work rather than paying again.
2. First-party data: If GSC exists, collect Query, Page, Country, Clicks, Impressions, and Position. Use GA4 to examine organic landing pages and meaningful conversions. Query only the site properties the owner has authorized.
3. Keyword tools: Use Semrush Keyword Magic or Keyword Gap, Ahrefs, DataForSEO, or another suitable provider to gather core seeds, synonyms, and competitor-page ranking terms by market.
4. Targeted SERP checks: Inspect live result types and ranking URLs for high-value or ambiguous keywords.
5. Sales and product evidence: Confirm customer language, real models, supply boundaries, and the technical evidence available to support pages.

First test 5–10 representative terms to confirm the tool returns useful data and the budget is sufficient. Then expand. When an external service explicitly reports insufficient credits, stop billable calls, reuse existing results, and document the blocker. Do not automatically recharge or switch services to keep spending.

Semrush is not mandatory. Automated clustering produces suggestions, not an approved page plan. Choose competitors for relevance; collecting dozens of domains is not a target in itself.

4.1 Minimum Data Fields

Keyword research master sheet (CSV/Sheet):

keyword,market,language,volume,kd,intent,source,captured_at,metric_status,relevance,cluster_id,notes

• Volume, KD, and CPC are tool estimates. Label the platform, country, collection date, and methodology.
• Distinguish RETURNED, NULL, NOT_QUERIED, and ERROR in metric_status; a missing value is not zero.
• Similar keyword variants may share underlying search volume. Do not simply add their volumes together.
• Paid search competition, organic KD, CPC, and the business value of a deal measure different things.
• Keep the same keyword in separate rows for different countries. Do not average metrics across tools without a defensible reason.

4.2 Research Coverage and Data Quality

Every research batch should answer: what was researched, in which market, when it was collected, through which tool, and what the tool actually returned. Preserve at least the following:

Record	Minimum requirement
Query	Keep the exact submitted query, not only a translated or normalized version
Market and language	Record country/region and search language separately; English data does not represent other language markets
Date and source	Record collection date, tool/data provider, and query batch
Metric status	Distinguish returned values, returned nulls, not queried, and request errors; null is not zero
SERP evidence	Record check date, location/language, and actual result types; label old captures as historical snapshots
Coverage limits	List unresearched markets, incomplete queries, and quota or access blockers rather than filling gaps with assumptions

Older data does not become invalid automatically after a fixed number of days. However, recheck key-market SERPs and data freshness before making high-impact page decisions. If a provider reports insufficient credits or INSUFFICIENT_CREDITS, stop billable requests, preserve the missing-data status, and document the blocker. Do not automatically recharge, invent metrics, or mark failed requests as completed.

5. Clean Keywords Without Losing Valuable Long-Tail Terms

Normalize spaces, letter case, and exact duplicates. Do not merge voltages, dimensions, processes, materials, screen technologies, or distinct product models merely because their wording looks similar.

Use four keyword-handling statuses: KEEP (relevant), REVIEW (needs verification), EXCLUDE (confirmed mismatch), and BRAND (own or competitor brands analyzed separately). These statuses manage keywords; they do not prove that product facts are verified.

Label the evidence behind product and page claims separately:

Evidence label	Meaning	Limitation
CONFIRMED_CURRENT	Current supply or service capability has been confirmed by the responsible owner	Still specify applicable models, versions, and conditions; do not extend the claim to an entire family automatically
DOCUMENTED_HISTORICAL	Historical models, specifications, or archived documents are on record	Does not prove current production, stock, certification, warranty, or cross-model compatibility
SERP_OBSERVED	Search results were observed in a specified market, language, and date	Proves only what appeared in that search snapshot, not product availability or stable rankings
HYPOTHESIS	A judgment that has not yet been supported by sufficient evidence	Keep it explicitly unverified; do not present it as fact

These labels describe different kinds of evidence, not a single low-to-high score. In particular, SERP_OBSERVED cannot substitute for proof of supply, inventory, certification, or technical specifications.

Retain low-volume, high-purchasing-intent terms as REVIEW or KEEP rather than deleting them mechanically. Put broad, consumer-focused terms or terms without actual product evidence on HOLD when appropriate. Never force optimization solely because a keyword has volume.

6. Determine Search Intent from the Target Market’s SERP

Common intent types are informational, navigational, commercial investigation, and transactional. A single query can mix several intents.

For important terms, record:

• Target country, language, and date checked.
• Whether results predominantly show product detail pages, categories, manufacturers, technical guides, DIY home content, or marketplaces.
• Key ranking URLs and the tasks those pages answer.
• Which existing page type on your site can satisfy the same task.
• Whether you have a historical snapshot rather than a current SERP measurement.

Typical pitfalls: “shelf lighting” may return kitchen or furniture lights rather than retail shelf systems; “power track” may refer to wall-mounted socket rails rather than low-voltage retail power tracks; “transparent screen” may refer to OLED rather than transparent LED film. Similar words do not necessarily reflect the same customer need.

7. Cluster Around One Primary User Task and Page Owner

Evaluate clusters in this order: actual product category → buyer task → SERP page type and URL overlap → whether an existing page can fully satisfy the need.

Examples:

• LED Shelf Lights / LED Shelf Lighting: possible shared cluster, subject to market-specific validation.
• LED Shelf Lighting Manufacturer: supplier-capability intent; first evaluate whether an existing category or manufacturing-capability page handles it.
• How to Install LED Shelf Lighting: installation task; perhaps a guide or technical section on an existing product page.
• 24V Power Track: separate low-voltage retail tracks from wall-mounted outlet rails and inductive wireless power.
• Electronic Shelf Labels / Digital Shelf Labels: possibly the same cluster, but not equivalent to animated full-color LED/LCD displays.

SERP overlap is supporting evidence, not a universal fixed-percentage threshold. Label final groups CONFIRMED_SERP, HISTORICAL_SERP, or HYPOTHESIS. Automated grouping does not replace human verification.

8. Map Clusters to Pages, Reusing Existing URLs First

Page planning sheet (CSV/Sheet):

cluster_id,primary_keyword,secondary_keywords,search_intent,market,target_url,page_type,action,priority,evidence_status,notes

Page role	User task it should address	What to avoid
Homepage	Brand, business navigation, overall positioning	Trying to own every product term
System or product overview	Range of offerings within a business line and procurement entry point	Mixing unrelated systems and keywords
Product family or category	Selecting and comparing a product type	Repeating every model detail page
Product detail (PDP)	Real model, specifications, fit, and inquiry	Presenting one model’s specifications as valid for the entire family
Solution / Application	Solutions, application constraints, and integration tasks	Claiming integration capabilities that have not been verified
Technical / Blog	Installation, comparisons, selection, and technical questions	Producing duplicate articles for every variant
Projects / Resources	Verifiable projects, drawings, and catalogs	Pretending an evidence page is a transactional product page

Record an explicit decision for each cluster:

• KEEP: the existing page already satisfies the primary search task; no change is needed.
• OPTIMIZE: the existing URL has a specific, evidenced content or conversion gap; improve that page first.
• NEW: the search task is genuinely distinct, existing pages cannot satisfy it, and sufficient product/case evidence supports independent content.
• VERIFY: supply, model, specification, compatibility, or target-market intent still needs validation; make no page claims until checked.
• HOLD: defer while market, resource, or business evidence is unavailable.
• EXCLUDE: confirmed mismatch; do not target it.

VERIFY is an evidence-check status, not authorization to create a URL or edit a page. Do not turn VERIFY into OPTIMIZE or NEW merely to make every row appear actionable.

Multiple URLs ranking for one term does not automatically prove keyword cannibalization. Check buyer tasks, actual content, and GSC Query × Page data before bulk merging, deletion, or 301 redirects.

A keyword map does not authorize changes to titles, H1s, canonicals, routes, crawl permissions, or production deployment.

9. Prioritize Business Fit Over the Illusion of Traffic

First ask: “Can we actually supply it, and is this a target buyer?” Then consider search volume, competition, CTR/CPC, existing page readiness, and production cost.

Priority	Criteria
P1	Strong business relevance, credible supply evidence, clear purchasing path, and an existing URL that can be reused
P2	Relevant but needs technical proof or has meaningful competition/content costs
P3	Longer-term content opportunity with unverified near-term business value
HOLD	Search intent, metrics, availability, certification, or software compatibility cannot yet be verified

When resources are limited, select 1–3 strongest page owners per core business line before creating dozens of pages. For international sites, separate shared page architecture from market-specific keyword research.

10. Implementation and Verification

For each high-priority page, prepare a short content brief: user task, primary and secondary terms, existing URL, missing purchasing information, real product/model evidence, key internal links, CTA, country and language, and expected metrics.

At the page level, verify the actual URL, HTML title/H1, canonical, main content, product evidence, functioning links, and inquiry path. Google does not require a fixed title length, keyword density, or universal word count; identify internal editorial constraints separately.

At the outcome level, use GSC queries and page-level impressions/clicks/CTR/position, GA4 organic landing sessions and real conversion events, plus CRM or human lead qualification if needed. Account for attribution differences, observation windows, seasonality, market differences, and data-provider methodologies.

When translating page ownership into a real procurement journey, inspect three links in sequence: business entry point → relevant product family/solution → evidence-backed model or technical resource → correct inquiry path. Check that links work, match the surrounding context, and lead to the right business. Do not add links merely to meet a fixed count.

Page implementation, code merge, production deployment, search-engine crawling, and business outcomes are separate acceptance stages. A passing PR or CI run does not prove that the production page has been checked; a working production page does not prove indexing or qualified inquiries.

If robots.txt still blocks the whole site or GSC is not connected, you can deliver strategy and page planning, but cannot claim search performance has been validated. Enabling crawling, changing indexability, or publishing production changes needs separate authorization.

11. Failure and Blocker Handling: Lessons from Practice

1. Valid older research with a dated snapshot: Reuse it with its date; recheck critical SERPs before new high-impact decisions. Do not automatically repurchase.
2. Keyword tool credits exhausted: Stop paid calls and preserve missing-data statuses. Never invent volume or KD.
3. A GSC property exists but is not connected to the project: Distinguish account access from a property that this research can actually query. The same applies to GA4: an account or property existing does not prove that the project is connected or that conversion events are readable.
4. An expert category page already exists: Reuse its page owner rather than creating a new URL for every keyword.
5. Historical model specifications: They may support technical provenance but do not prove current production, stock, certification, or cross-model compatibility.
6. Multilingual sites: Publishing many language versions does not mean completing separate keyword research for each market. State countries, priorities, and gaps.
7. Research is not publication: A–C can be delivered as recommendations; D implementation and E conversion measurement are independent subsequent stages.
8. Research coverage is incomplete: Keep not returned, not queried, request errors, and true zero values separate. Never fill gaps with data from another country, a similar query, or a different language.
9. A competitor has a more elaborate page structure: First confirm that it serves the same buyer task, then check whether your site already has equivalent content. A competitor having a page is not a reason by itself to create one.
10. PR status is confused with production status: Record code checks, CI, merge, production HTML/HTTP/canonical acceptance, and subsequent search performance separately. A commit or merge is not proof of SEO results.

12. Minimum Deliverables Checklist

• Product scope, target markets, and meaningful conversions defined.
• Dated keyword master sheet with documented sources; missing metrics distinguished from actual zero.
• Traceable SERP assessments or explicit unverified status for ambiguous P1 terms.
• Cluster-to-owner mapping, with existing URLs verified against source files or a page inventory.
• KEEP / OPTIMIZE / NEW / VERIFY / HOLD / EXCLUDE decisions supported by business reasons.
• Metric freshness, unresearched markets, tool costs, GSC/GA4/indexing blockers clearly recorded.
• Actionable next steps without treating a research document as permission to edit the site; VERIFY, HOLD, and EXCLUDE decisions have explicit reasons.
• Key product facts have evidence labels, applicable model/version scope, and limitations; historical documents are not presented as current supply claims.
• Business-entry, product/resource, and inquiry links for high-priority pages have been checked.
• Research, implementation, production acceptance, indexing observation, and business outcomes are recorded as separate stages.

References

• Google Search Essentials
• Google SEO Starter Guide
• Google People-first Content
• Google Search Console Performance
• Semrush Keyword Magic Tool
• Semrush Keyword Gap
• Semrush Keyword Strategy Builder

This guide provides general operating principles. Store each company’s actual search volumes, model numbers, URL owners, and evidence in its own research report rather than mixing them into a general guide.

13. Competitor Page Layout Comparisons: From Keyword Clusters to Content Gaps

Introduced in V1.2 and expanded in V1.3. Comparisons should rely on independently verifiable public pages and authorized site materials. Observing page structure does not prove Google ranking or conversion performance.

The goal is not to collect competitors’ page titles. It is to determine how a page answers a specific search task completely, then check whether your own site already offers equivalent content.

13.1 Compare Pages Serving the Same User Task

Do not use a competitor’s product detail page as proof that you need a new category page. Nor should different business models, regions, or sales channels automatically be considered SEO gaps.

For the same keyword cluster, compare equivalent page types: business hubs, product categories, individual PDPs, solutions, specialist guides, and question-based FAQs. If an external page targets different products or buyers, label it as a reference rather than a direct competitor.

13.2 Record Nine Fields per Page

Field	Purpose
Competitor URL	A directly inspectable page, not merely a domain
Target task / cluster	What buyer problem is addressed; do not call these “ranking keywords” without ranking data
Title / H1	Verified visible or inspectable page text; otherwise mark unknown
Main content	Product counts, specific model specifications, selection comparisons, installation help, FAQs, and documentation
Internal links	Actual category → product, guide → product/resources, and product → purchasing paths
Our owner URL	The existing page and business line; distinguish source code from production
Already exists?	Whether corresponding content is already there; do not report existing work as a gap
Evidence / limitation	Live observation, code, historical files, inference, and data date
Action / priority	KEEP, OPTIMIZE, VERIFY, or HOLD, with the smallest justified change

13.3 Do Not Infer Competitor Sales, Rankings, or Technical Claims

• A complete-looking competitor page does not prove it ranks better organically. Ranking, organic traffic, and keyword share require market-specific data and suitable sources.
• A competitor’s specifications, certifications, inventory, integrations, and compatibility apply only to the exact products verified; they cannot be transferred to your own.
• More precise page wording does not automatically mean the URL should change. Consider a new page only when a distinct search task, independent content value, and insufficiency of existing pages are all established.
• Generic technical guides from other sites may target end consumers. B2B sites should prioritize selection, specifications, installation, and quotation questions that support procurement.

13.4 Audit Internal Links Alongside Page Ownership

For every high-priority cluster, check whether real pages enable users to:

1. Reach the relevant product family or solution from the main business/topic entry point.
2. Reach valid individual models and technical documents from that family or solution.
3. Navigate back to the right business area from a model or guide and reach a real, working inquiry or purchase path.

This is a relevance audit, not a requirement to add a fixed number of links to each page or build another technical service. Fix broken links and links leading to the wrong intent before unnecessary expansion.

13.5 Practical Corrections from Comparison Work

When comparing COB LED strip category pages, one competitor may show voltage, wattage, width, and cutting interval directly in listings, while another site may have only old models and unverified specification revisions. The correct action is not to populate every card immediately. First confirm a small set of currently sold models, then decide whether a concise specification comparison belongs on the existing family page.

Similarly, a category titled “Shelf Lighting” may primarily show connectors and power-track accessories. Its title alone does not prove comprehensive coverage of retail shelf luminaires. Inspect actual products and content, not titles alone.

13.6 Know When to Stop

If comparison work shows that an existing page owner, buyer guide, model links, and inquiry path already meet the search task, mark KEEP / NO ACTION and stop. Only evidenced gaps belong in the implementation queue. Do not keep producing synonym pages or generic SEO articles merely to imitate high-performing sites.

13.7 Anonymized B2B Case Study: From Keyword Clusters to the Smallest Justified Page Action

This case omits the company name, exact URLs, country-level search metrics, SKUs, and internal execution status. It preserves only transferable research and decision logic. It is not evidence of universal ranking outcomes and makes no traffic or lead-growth claims.

Keyword/product theme	Facts to verify	More defensible page decision
Commercial display lighting / LED strips	Whether the category already contains real products; whether voltage, wattage, width, and process refer to the same current model/version	Keep the existing family page first. Consider a compact specification comparison only after model versions are verified; do not copy historical specifications into every product card
Low-voltage conductive power rails	Mechanical/electrical matching among the rail, feed-in, pickup, and load; whether “wireless” still relies on conductive contacts	Explain contact-based conductive power accurately; verify compatible components and ratings per system. Do not claim unproven wireless transmission or universal cross-family compatibility
Electronic shelf labels and dynamic displays	Whether e-paper, animated full-color screens, and LCD belong to different display/power systems	Separate the technologies and procurement tasks; do not merge specifications, prices, or compatibility because the names sound similar
Stretched commercial LCD	Whether independently sellable models exist or only a complete catalog/brochure is available	Let the existing resource catalog serve the query first. Create a standalone PDP only when a real SKU, distinct buying task, and sufficient evidence all exist
Transparent display film	Whether LED film is being confused with transparent OLED in SERPs or customer language; whether glass, control, and installation evidence is sufficient	Preserve the technology boundary and use verified film documentation; do not borrow OLED specifications or claims
Light-box manufacturing terms	Whether current sellable models, materials, manufacturing capability, and real project evidence exist	Mark HOLD when evidence is insufficient; do not create a manufacturer page merely because a keyword tool reports search volume

The lesson is not that every keyword needs its own page. First identify the real procurement task, then determine whether the existing owner can satisfy it. Choose KEEP / NO ACTION when the page is sufficient; use OPTIMIZE only for a demonstrated gap; use VERIFY when supply or technical facts remain unconfirmed instead of publishing premature commercial claims.

V1.3 principle: keyword metrics, SERP observations, product facts, and production acceptance must be evidenced separately. The value of research is more accurate and disciplined page decisions—not a larger page count.