From Keywords to Prompts

Reading time

13 min

Date published

2025-09-09

From Keywords to Prompts

Does it still make sense to run a keyword analysis today? How should we actually work with data from organic search? And can prompt analysis bring us closer to the real needs of users?

The Data Remains, the Strategy Changes: The New Role of Keywords

In the era of AI-powered search, the very role of keywords is changing. They are no longer just “tags” designed to push your content into the coveted top 10 search results.

Today, keywords act more as signals of user intent. They serve as starting points from which AI-driven search engines extract related queries and piece them together into a personalized answer.

This means we can no longer look at keywords as isolated phrases. Instead, they should be seen as a map of needs and scenarios—ones that your content must address naturally and convincingly.

Organic Data Remains a Key Indicator of Potential

One thing is certain: keywords remain the most reliable measurable signal of what people are actually searching for. Data from Google Search Console, Ahrefs, or Semrush clearly shows which information or products users need, how often they search for them, and how interest evolves over time.

💡 We can think of keyword analysis as a compass. It helps us gather the data and serves as a litmus test for the real needs of our potential customers.

Prompts as a Path Closer to Customers

The difference between traditional SEO and AI search lies in what happens to the keyword. In the past, we optimized pages for a specific phrase. Today, the keyword is only the starting point of the process.

This aligns with the principle of query fan-out, on which AI search is built: a single phrase expands into a network of related queries. This creates a range of long-tail topics that, in practice, better reflect people’s real needs.

The more comprehensively our content covers a given topic, the more likely AI will consider it a valuable source. And since AI aims to address as many user intents as possible, it pushes us to create content that answers real questions and meets real expectations.

Ultimately, it’s not about the numbers—it’s about satisfied users who find the answers they need in our content, build trust in our brand, and can turn into loyal customers.

💡 Within the next two to four years, AI search may surpass traditional search. Those who don’t start optimizing for LLMs in time risk falling behind their competitors—who will secure top spots in AI-generated results and, with them, the majority of user traffic and attention.

Practical Example: Keyword vs. Prompts

Before: The goal was to include the keyword in the headline and content.

“best yoga mat”

Today: AI search breaks down the user query into related phrases, such as:

“What thickness of mat is best for beginners?”
“Which material is most durable and non-slip?”
“Which brands are recommended by professional yogis?”
“How to choose a mat based on yoga style (hatha, vinyasa, hot yoga)?”

If your content only answers the original phrase but doesn’t address these specific scenarios, AI will most likely select a different source.

The user isn’t just looking for a single phrase—they are trying to solve a broader need. That’s why prompt analysis helps us create content that is more useful, more comprehensive, and closer to the real expectations of customers.

💡 “Good SEO is good GEO,” says Danny Sullivan from Google.

How to Work with Prompt Analysis

1) Collect Data for Keywords

From tools like Google Search Console, Ahrefs, Marketing Miner, or Semrush, you can extract countless data points:

specific queries, search volume, trend, seasonality
intent variations (what, how, why, vs., best, for [segment], within [budget])
SERP features (People Also Ask, Top Stories, videos) and a note on whether an AI Overview appears for the keyword
pages that already bring in high-quality traffic (strong engagement/conversions)

Output: a raw list of seed topics and queries grouped into thematic clusters.

Practical Tip

When working in a small market like the Czech Republic, we can also draw on data from abroad. For example, in an analysis for a client in the medical supplies segment, we expanded our research of long-tail keywords with results from the German market—keywords that didn’t appear in Czech data due to low search volumes.

The assumption was simple: patients there face similar needs as those in the Czech environment. We then recalculated the collected data using an appropriate coefficient and created a set of relevant queries tailored to the local context.

2) From Keywords to Prompts

Once we’ve collected the seed topics from a classic keyword analysis, the next step is to expand them into AI prompts. In practice, this means mapping the so-called query fan-out—the breakdown of a single query into a tree of related sub-questions.

Typical directions AI tends to develop when generating answers include:

Basics / definitions: what it is, how it works, when it makes sense
Parameters: thickness, material, dimensions, compatibility, performance
Variants & alternatives: for whom, for which style/purpose, budget vs. premium
Comparisons & recommendations: A vs. B, “best for…”
Price & availability: budgets, where to buy, warranty, service
Maintenance & risks: cleaning, lifespan, common mistakes, safety

You don’t have to do all of this manually—AI tools themselves (ChatGPT, Gemini, and others) can help. They can quickly expand a single seed topic into dozens of query variations and, based on the data provided, estimate which ones are most relevant for users and worth developing further.

Most common prompt categories:

Definitional: “What is…”, “Explain the difference between…”
How-to: “How to choose…”, “How to fix…”, “Step-by-step guide to…”
Comparative: “Compare A vs. B”, “Best X under Y dollars”
Constrained choice: “Recommend X for [beginner / allergy sufferer / small car] under [budget]”
Diagnostic / problem-solving: “Why isn’t my… working?”, “What to do if…”
Local: “Where in [city]…”, “Nearest service for…”

Output: a simple list of relevant prompts for each seed topic, which then becomes the foundation for structuring your content.

3) Mapping Seed Topics to Specific Landing Pages

Thanks to data from the tools, we can also see which keywords our website already ranks for and which it doesn’t. This allows us to create an overview that reveals two key things:

Where landing pages are missing: for seed topics or prompts that don’t yet have a corresponding landing page, a new one needs to be created.
Where we already rank: if the website already holds a position for a given keyword or topic, we enrich the existing landing page with prompts and fan-out sub-questions to expand it and increase its chances of appearing in AI-generated results.

At this stage, we have already connected the keyword and prompt analysis and built a clear plan of what to create from scratch and what to improve.

Output: an overview of seed topics including specific prompts and landing pages (either existing or new).

4) Creating a Prompt Canvas

For larger websites or highly searched topics, it’s worth keeping prompt analysis organized. A simple table can help, where for each seed topic you record:

Prompt and its type: is it a definition, a how-to, or a comparison?
Query intent: is the user just looking for information, comparing options, or already close to making a purchase?
Recommended answer format: table, FAQ, checklist, short guide…
Evidence (EEAT): what can we support with our own experience (tests, data, photos)?
Priority: how important is the prompt or seed topic from a business perspective, and how much effort will it take to cover it?
Demand & seasonality: volume, trend, regions.

5) Working with the Content Itself

Once the prompt analysis is complete, we can move from theory to practice—content creation:

Content brief. Turn the output of the prompt analysis into a clear content plan. This plan should be delivered to content creators in a way that ensures the final output follows best practices for AI search while reflecting the fan-out map we identified—covering all key sub-questions and user scenarios.
Content creation and publishing. Based on the analysis, create new pages or update existing ones.
Measurement. In the era of AI search, it’s increasingly important to focus less on raw traffic and more on visit quality. It’s not just about how many people land on your page—it’s about what they do there. How much time do they spend, how do they interact with the content, and do they take the desired actions? These indicators reveal whether the content truly meets user needs.
Maintenance and updates. Regularly check for changes in trends or offerings. Update content to keep it relevant and sustainably visible in AI results. To quote Google: “The only constant in search is change—just like the needs of the people who use it.”

Conclusion

The goal of optimization is no longer to cover a handful of narrowly defined keywords as effectively as possible. Yet organic data remains irreplaceable. Especially in the early stages of AI search, there is no better source of insight into what our potential customers are looking for.

Moreover, organic data enables us to expand prompts through query fan-out, connect topics within the knowledge graph, and create content that brings our brand closer to users.