B2B keyword research prompts

These prompts help turn research into decisions. Open the one that fits your current step, copy it and replace the square brackets with your product details and data. Pass reviewed output to the next step; you do not need to run every prompt each time.

The prompts do not supply search data by themselves. Provide exports or sources you have checked, and preserve missing data as unknown. Before adopting a recommendation, inspect its source rows and product fit. For this article’s example, you can give prompt 3 the two lists in the comparison file; the historical dates should remain in the answer.

1. Turn customer questions into seed queries
Use when you have product context and customer questions, but no research list yet.

Help me create a seed-query list for a B2B product.

Product and verified capabilities: [details]
Audience and the decision they need to make: [details]
Target language and country: [details]
Anonymized customer questions: [paste; state if unavailable]

Suggest queries expressing this audience’s problems and decisions. Separate wording from the supplied customer questions from your own suggestions. Do not invent product capabilities, search volumes, competition or questions that customers actually asked. If essential product or audience information is missing, ask before suggesting queries.

Return a table: query | reader need | supplied source or illustrative idea | product fit | required validation.
End with the queries you would investigate first and why. Each idea is a research candidate, not an instruction to write an article.

2. Clean an export without losing context
Use when you have keyword-tool data that needs to be prepared for comparison.

Clean the following keyword export for content planning.

Data: [paste a table or attach a file]
Tool, country, date range and metric definitions: [known details]

Preserve each original query and source row ID. Add a normalized query column for review only. Flag duplicates and spelling variants, but do not automatically delete or combine rows with different countries, dates or metric definitions. Do not add variant volumes and present the sum as a unique audience.

Distinguish zero from missing data and a yearly average from the latest-month estimate. Do not alter numbers or fill missing values. Ask for the definition of any ambiguous column.

Return a clean table retaining source row ID, query, provenance, measurement conditions and notes, followed by issues requiring my decision. Do not infer search intent from the numbers alone.

3. Decide whether two queries need one page
Use with two result lists and content from the relevant pages.

Help me assess whether two queries can be served by the same page.

Query A and its results: [full URLs, titles and result types]
Query B and its results: [full URLs, titles and result types]
Country, language, device if known, source and date for each list: [details]
Relevant excerpts from pages that were opened: [content and source URL]
Our product and audience: [details]

Compare organic results only. Separate ads and AI citations. Report shared URLs separately from shared domains and explain any normalization. State how many results each list contains; do not assume ten.

Compare the reader’s task, page type and audience. Where page content was not opened, label the classification as title-based only. Do not describe old results as current or infer causation from overlap.

Return: observations linked to source rows | possible interpretation | recommendation to group, separate or investigate | evidence that would change the recommendation. Do not substitute a fixed overlap threshold for judgment.

4. Turn research into a page decision
Use to choose an update, a new page or a hold, then hand the team a task.

Turn the attached research into a page-decision map.

Query groups, needs and evidence: [paste checked research]
Existing pages and their roles: [URL, audience, content and desired action]
Product capabilities and limitations: [details]

For each group, first assess whether an existing page can serve the need. Recommend a new page only for a distinct reader task with useful content we can provide. Do not invent existing URLs, performance metrics or capabilities. Label proposed URLs as new. Hold decisions with missing or contradictory evidence.

Return a table: group | reader need | existing or proposed page | update, new or hold | reasoning and source | missing information | next task | responsible role.

For the best-fitting group, add a mini-brief: the reader’s decision, the explanation currently missing, evidence to collect and page acceptance criteria. Do not draft the full article yet or propose a page for every wording variant.

5. Prioritize for fit and ability to compete
Use when the page map exceeds the team’s available time and resources.

Prioritize this content map by business fit, evidence and ability to execute.

Page map: [paste]
Site maturity, relevant pages and known performance: [data and sources; mark gaps]
Competing results and available backlink or authority data: [data, tool and date]
Team capacity and planning period: [details]

Do not select a target solely for high search volume. For a new or low-authority site, assess whether a competitive target should wait. Compare actual pages and evidence; do not turn DR or KD into an automatic cutoff or ranking guarantee. Without competition or authority data, mark ranking feasibility as unassessed.

Return three groups: work now, investigate before deciding, and hold. For each item, include reasoning, explicitly estimated effort, a dependency and evidence that would change its priority. Distinguish research tasks from writing assignments.

End with one task achievable this week within the supplied capacity. Do not forecast traffic, leads or revenue without suitable data and a model.