If I type: Give me keywords about SEO.

I will probably get exactly what I asked for: a broad list of predictable SEO terms.

SEO tools.
SEO strategy.
SEO tips.
Keyword research.
Technical SEO.

Nothing wrong with them. Nothing particularly useful either.

The quality of AI keyword research depends heavily on the information you give the AI before asking for keywords.

Here is how I do it.

Before Getting Started

Before asking ChatGPT, Claude, Gemini, or another AI tool for keywords, I explain what the business does.

I normally give it:

• The product or service

• The target customer

• The main problem the product solves

• The country or market

• The type of content I want to create

For example:
I run an SEO consultancy in Berlin. I work mainly with B2B SaaS companies that want to increase organic traffic and improve their visibility in Google and AI search engines. I want to find topics for educational blog content targeting people responsible for SEO or marketing.

Now the AI has some context.

That immediately gives me better results than asking for "SEO keywords."

Tell AI Who Is Searching

This part changes the results more than people expect.

Compare these audiences:

• Someone learning SEO

• A SaaS marketing manager

• An ecommerce owner

• An SEO specialist

• A founder looking for an agency

They might all search for SEO information, but they will not search in the same way.

I usually include the audience directly in my prompt: The target audience is B2B SaaS marketing managers and founders who understand basic SEO but are not SEO specialists.

That stops the AI from filling the list with beginner queries like "what is SEO?"

Unless those are the keywords I actually want.

Give It a Seed Topic

AI works better when I give it somewhere to start.

Let's say I want to write about AI visibility.

Instead of: Find keywords about AI.

I would start with: My seed topic is AI visibility and how brands appear in ChatGPT, Gemini, Perplexity, and other AI search experiences.

Then I ask it to expand around that topic.

For example: Generate related topics and keyword ideas around this seed topic. Include problems, questions, comparisons, use cases, tools, processes, and commercial searches.

Now I am exploring a topic instead of asking AI to guess what I mean.

Ask for Search Intent

A keyword list becomes much more useful once you understand why someone would search each query.

I normally ask AI to classify keywords into:

• Informational

• Commercial

• Transactional

• Navigational

I also like adding a more practical description of the intent.

For example:

This makes content planning much easier.

Ask for Different Keyword Angles

One prompt normally gives me obvious keywords.

The useful ideas often come from the second and third prompts.

I ask AI to explore the same topic from different angles.

For example:

Find questions beginners would ask about this topic.

Then:

Find problems an experienced SEO manager might search for.

Then:

Find comparison keywords related to this topic.

Then:

Find keywords someone might search shortly before buying a tool or hiring a consultant.

Then:

Find long-tail queries that describe a very specific problem.

I prefer doing this in several rounds instead of asking for 500 keywords at once.

The results are easier to review and usually less repetitive.

Make AI Group the Keywords

A list of 200 keywords is not a content strategy.

I ask AI to group related keywords into clusters.

A prompt can be as simple as:

Group these keywords by topic and search intent. Keywords that could realistically be covered by the same page should stay in the same cluster. Do not create separate pages for small wording variations of the same search.

That last sentence helps.

Without it, AI has a habit of treating tiny keyword variations like completely different topics.

For example:

• AI visibility tracking

• track AI visibility

• how to track AI visibility

• AI search visibility tracking

Those could easily belong to the same page.

I want clusters based on the actual subject and intent, not every variation AI can generate.

Ask AI to Find the Gaps

Another prompt I use is:

Looking at these keyword clusters, what important questions, problems, comparisons, or subtopics are missing?

This is useful because I am no longer asking AI to generate everything from scratch.

I am asking it to review an existing structure and find holes.

I can also give it my current site content:

These are the articles already published on my website. Based on my seed topic, identify relevant areas I have not covered yet. Do not suggest topics that substantially overlap with existing articles.

That can turn AI into a decent content-gap brainstorming tool.

Ask for a Table You Can Actually Use

Formatting matters.

If I do not specify the output, I often get a long response that is annoying to move into Sheets or Excel.

I normally ask for something like:

Return the results as a table with these columns:

Keyword
Topic cluster
Search intent
Funnel stage
Suggested page type
Suggested article angle

You can add whatever matters to your workflow.

For local SEO, I might add location.

For international SEO, I might add country and language.

For ecommerce, I might add product category.

Keep the columns useful. More columns do not automatically mean better research.

Do Not Ask AI for Fake Search Volume

This is where I would be careful.

You can ask an AI model to estimate whether a keyword sounds broad or niche.

I would not treat numbers it invents for monthly search volume or keyword difficulty as SEO data.

If I need actual metrics, I check the keywords with tools such as Ahrefs, Semrush, Google Keyword Planner, or whatever keyword database I am using.

My AI workflow is usually:

1. Generate ideas.

2. Expand the interesting ones.

3. Group them.

4. Remove duplicates.

5. Export the list.

6. Check the keywords against real search data.

7. Prioritize what is worth targeting.

AI speeds up the thinking.

The SEO tool gives me the data.

A Keyword Research Prompt I Would Actually Use

Instead of asking:

Do keyword research for me.

I would use something closer to this:

Act as my keyword research assistant.

I run a B2B SEO consultancy targeting SaaS companies in Germany and other European markets.

My target audience is founders, marketing managers, and SEO managers who already understand basic SEO.

I want to create content around AI visibility.

Generate keyword ideas around:

• Problems people experience

• Questions they ask

• How-to searches

• Comparison searches

• Tools

• Commercial searches

• Long-tail queries

Group closely related keywords into topic clusters.

For every keyword, include:

• Keyword

• Cluster

• Search intent

• Likely audience

• Suggested content type

Do not invent search-volume or keyword-difficulty numbers.

Avoid near-duplicate keywords that represent the same search intent.

Start with 30 keywords.

Then I review the output.

I might follow with:

Cluster 3 looks interesting. Give me 30 more specific keywords around that topic.

Or:

Which questions inside these clusters have not been answered yet?

Or:

Separate keywords suitable for blog posts from keywords that would make better service, comparison, or landing pages.

That conversation is usually much more useful than one giant prompt.

One More Thing: Give AI Real Data

The workflow gets better when I stop making AI guess.

I can paste in:

• Search Console queries

• Ahrefs exports

• Semrush exports

• Competitor keyword lists

• Existing article titles

• Customer questions

• Sales call notes

• Reddit discussions

• Support tickets

Then I can ask:

Group these Search Console queries into topics and identify clusters where I already have impressions but weak coverage.

Or:

Here are 300 competitor keywords. Remove branded queries, group the remaining keywords by intent, and find topics relevant to my SaaS SEO service.

That is much stronger than starting with an empty chat window.

My Rule for AI Keyword Research

I use AI to expand my thinking, organize messy keyword lists, find patterns, and explore search intent.

I do not use it as the source of truth for keyword metrics.

Give it context.

Tell it who the customer is.

Give it a clear seed topic.

Ask for different search intents.

Make it cluster the results.

Then validate the keywords with real SEO data.

That gives me a much more useful list than asking ChatGPT to "find some keywords."