How can you collect and initially screen keywords and AI prompts worth considering?

Start by clarifying the topics relevant to your business, your target audience, and how you intend to use the research. Then gather candidate keywords and prompts from customer questions, search results, existing visibility, and competitors. For the initial screening, prioritize relevance and demand, then consider search volume, keyword difficulty, and search intent. Here, AI prompts are questions that your audience asks AI. They help you identify content worth creating; they are not instruction templates for having AI write articles. Candidates worth keeping should have the potential to support at least one of the following: brand awareness, AI visibility, lead generation, or purchases.

Define both the topic and the context

Build a list of core terms based on the products and services your brand actually offers, then expand it using your audience’s language. A 2018 awoo article uses “SEO” as an example, expanding it into “What is SEO?”, “SEO tutorials,” and “SEO case studies.” Although these terms share a topic, they do not necessarily call for the same type of content.

For prompts, retain the specific context. Compared with “best CRM,” “Which CRM is best for a two-person real estate team?” includes team size, industry, and purchasing needs. When collecting prompts, do not keep only the product name while removing the conditions that meaningfully affect the answer.

Expand the list using real questions and search clues

• Customer records: Review sales calls, support records, and onboarding questions. Record recurring questions and objections, keeping customers’ exact wording rather than rewriting it as marketing jargon.
• Public discussions: Look through relevant forums, social media communities, and video comments for unanswered questions and pain points. AI can help compile these, after which you can select those relevant to your business.
• Search results and keyword databases: Start with broad core terms to explore related topics, related searches, and “People also ask.” Questions in “People also ask” can offer clues to conversational phrasing, but their presence does not prove that anyone has entered them into AI. Expanding more questions can also take you further off topic.

For example, if support records repeatedly include “How do I export data from the tool to Excel?”, it is worth adding as a candidate. This question may serve both prospective buyers checking the tool’s features and existing customers who need instructions.

Procrustes also turns the questions customers genuinely care about into articles as part of its content services, with ongoing planning, writing, and updates.

Fill gaps using existing visibility and competitors

Websites that already appear in search results can use Google Search Console to identify relevant queries with many impressions but few clicks, looking for topics they have not adequately addressed. New sites or those without rankings can skip this step. You can also compare terms for which competitors rank but your site does not, or ranks lower, as well as prompts where competitors appear in AI answers but your brand is absent. However, competitor visibility only signals something worth examining—not something you should automatically adopt.

Check relevance and demand, then narrow the list using data

First, remove candidates unrelated to your services that also cannot support the goals outlined above. Then consider search volume, search intent, and keyword difficulty. Long-tail keywords are generally more specific and may have lower search volume and less competition, but they do not guarantee rankings. Terms drawn from real customer questions should not be discarded simply because a tool reports zero search volume.

When using prompt research tools, check whether the figures estimate query volume for an entire topic or describe individual prompts. Do not treat a topic-level estimate as the actual number of times each question has been used. The outcome of this round is a shortlist worth evaluating further, not a promise of results.