I asked AI to do keyword research, here’s what it got right and wrong

Keyword research has always been one of the most methodical parts of SEO. Traditionally, it meant opening multiple tools, exporting keyword lists, analyzing intent, grouping topics, and slowly building a content strategy.

Then AI entered the picture.

Today, many marketers ask a simple question: Can AI do keyword research for me?

I decided to test this properly. Instead of relying on traditional keyword tools alone, I asked AI to generate keyword ideas, analyze intent, and structure a potential content strategy.

The results were surprisingly useful in some areas—and clearly flawed in others.

If you plan to use AI in your SEO workflow, understanding where it helps and where it misleads is essential.

Let’s walk through what actually happened.

Why I tested AI for keyword research

The promise of AI in SEO is speed. Tasks that once required hours of research can now produce outputs in seconds. For keyword discovery especially, AI seems appealing because it can quickly generate long lists of related search queries.

But keyword research isn’t just about ideas. It involves understanding search intent, identifying realistic ranking opportunities, estimating search demand, and structuring topics around actual business goals.

So the real question wasn’t whether AI could produce keywords.

The question was whether those keywords were strategically useful. To test this, I gave AI a more structured prompt.

“Act as an SEO strategist. Generate 50 keyword ideas related to SEO for SaaS companies.
Group them by search intent (informational, commercial, and transactional).
Focus on realistic search queries rather than marketing language, and include long-tail keywords that SaaS companies might realistically target.”

Within seconds, I had a full list organized by intent. At first glance, it looked thoughtful and comprehensive.

But once I started evaluating the suggestions carefully, checking how closely they resembled real search behavior, a more nuanced picture emerged.

The first thing AI did well

The moment the results appeared, the list looked structured and surprisingly thoughtful.

AI had organized the keywords into three clear groups: informational, commercial, and transactional. At a glance, this mirrored how many SEO teams structure their content funnels.

For example, under informational intent, it suggested queries such as:

• how SEO works for SaaS companies

• SEO strategy for SaaS startups

• how SaaS companies generate organic traffic

Under commercial intent, the suggestions shifted toward evaluation:

• best SEO agencies for SaaS companies

• SaaS SEO tools

• SEO services for SaaS startups

And under transactional intent, the keywords became more action-oriented:

• hire SaaS SEO agency

• SaaS SEO consulting services

• SaaS SEO pricing

At first glance, this looked like a solid starting point. The queries were clearly related to the topic, and the intent grouping resembled a usable content structure.

In fact, this is where AI immediately proves useful. It can turn a broad topic into a structured list of potential search queries in seconds. For marketers who are starting with a blank page, that alone can save time.

But once the initial impression faded and the list was examined more carefully, some limitations started to become obvious.

Where the list started to break down

The problems became visible the moment I started reviewing the keywords the way an SEO strategist normally would.

At first glance, the list looked credible. But when you read the keywords closely, some of them didn’t resemble real search behavior.

For example, AI suggested phrases like:

• SaaS organic growth optimization framework

• scalable SEO architecture for SaaS companies

• SaaS traffic acceleration strategy

These phrases sound polished, but they read more like marketing language than actual search queries.

People searching on Google usually type much simpler phrases. They search for things like:

• SaaS SEO strategy

• how to do SEO for SaaS

• SaaS SEO agency

Search behavior tends to be direct and problem-focused. AI, however, often generates language that resembles blog headlines or internal strategy documents, not raw search queries.

This happens because large language models are trained on massive amounts of written content—articles, guides, marketing materials, and documentation. As a result, they sometimes reproduce the language patterns of marketers rather than the language patterns of searchers.

The difference may seem subtle, but it matters.

Keyword research is not about identifying phrases that sound intelligent. It is about identifying phrases that real people actually type into search engines.

And that is where the next limitation of AI becomes even more significant.

The biggest limitation: AI doesn’t know search demand

The next issue appeared when I tried to evaluate the keywords the way any SEO team eventually must: by checking demand.

Good keyword research answers a simple question: are people actually searching for this?

AI, by itself, cannot answer that.

When a language model generates keyword ideas, it is predicting phrases that sound like searches, based on patterns it has seen in text. It does not have direct access to live search data, search volume estimates, or trends in user behavior.

That means some of the suggested keywords might exist in real search queries—but others might not be searched at all.

For example, a phrase like “SaaS SEO strategy” is likely to have measurable search demand. But something like “SaaS organic growth optimization framework” might never appear in real search logs.

Without external validation, there is no way to know.

This is where traditional SEO tools still play a critical role. Platforms such as keyword research databases, Google Search Console, and SERP analysis tools exist for a reason—they reveal whether a query actually appears in search behavior.

When I checked several AI-generated keywords against these tools, the pattern became clear. Some keywords aligned closely with real search demand. Others showed extremely low volume or none at all.

In other words, AI can produce plausible keyword ideas, but it cannot confirm whether those ideas correspond to meaningful search activity.

For keyword research, that distinction matters more than it might initially seem.

Another gap: AI doesn’t understand ranking difficulty

The next limitation became clear when I looked at the list from a competitive standpoint.

Keyword research isn’t only about identifying what people search for. It also involves evaluating whether you have a realistic chance of ranking.

AI does not perform that type of analysis.

For example, some of the keywords generated in the list included phrases like:

• SaaS SEO

• SaaS marketing strategy

• SEO strategy

These topics certainly exist in search demand. But they are also extremely competitive.

If you look at the search results for keywords like these, you’ll often see large, established websites ranking at the top—major marketing blogs, well-known SaaS platforms, and high-authority publications.

For a newer website or a smaller SaaS company, competing for these keywords would be extremely difficult.

A human SEO strategist immediately considers factors such as:

• domain authority of competing sites

• backlink strength

• content depth in the existing search results

• whether niche opportunities exist within the topic

AI does not evaluate any of these elements. It generates keyword ideas without assessing the competitive landscape.

This means the output may include keywords that are theoretically relevant but practically unrealistic to target.

In real-world SEO work, that distinction matters a great deal. A keyword might be popular and relevant, but if the competition is overwhelmingly strong, it may not be the best place to focus effort.

What AI actually did well

Despite these limitations, the exercise revealed something useful.

AI may not be reliable for validating keywords, but it is surprisingly effective at expanding the topic space around a subject. When I reviewed the full list again, many of the suggestions pointed toward related subtopics that could easily become separate articles.

For example, the generated keywords naturally branched into areas such as:

• SaaS link building

• SaaS technical SEO

• SEO for SaaS startups

• SaaS content marketing and SEO

• SaaS SEO tools

Instead of thinking only about one keyword, the list began to resemble a cluster of related topics.

This is where AI becomes genuinely helpful.

In traditional keyword research, marketers often start with a seed keyword and then explore related queries through tools, search suggestions, or competitor analysis. AI can accelerate that exploratory phase by generating many adjacent ideas quickly.

Even if some individual keywords are imperfect, the broader topic map can still be valuable.

In practice, this means AI works well as an ideation engine. It helps marketers move from one core topic to multiple supporting angles that could form part of a larger content strategy.

The important distinction is that these ideas still need to be validated with real data before they become part of an SEO plan.

The role AI should play in keyword research

After going through this exercise, the conclusion became fairly clear.

AI should not be treated as a replacement for keyword research tools. But it can be a useful assistant in the early stages of the process.

In practice, the most effective approach is to combine both.

AI works well when you want to quickly explore a topic. It can generate variations, identify adjacent themes, and suggest questions people might ask about a subject. That makes it helpful when you are trying to move from a single idea to a broader set of content possibilities.

However, once those ideas exist, they still need to be tested.

Search demand, ranking difficulty, and actual user intent must be validated using real search data. Tools that analyze search volume, competition, and existing search results remain essential for that part of the workflow.

A practical process might look like this:

First, use AI to expand a topic and generate potential keyword ideas.

Next, verify those ideas using keyword research tools and search data.

Then analyze the search results to understand what type of content Google is rewarding for those queries.

Finally, structure your content strategy around the opportunities that are both relevant and realistically achievable.

In other words, AI can help you move faster at the beginning, but strategic judgment and real data are still required to decide where to focus.