Why You Shouldn't Trust AI Tools to Give You Your Keywords

If you've ever asked ChatGPT, Claude or another AI tool to suggest keywords for your wellness website, the list it give you was pure guesswork.

And if your website copy has been optimised around that guesswork, you've spent time and effort optimising for keywords that nobody is actually searching for.

What AI Tools Are Actually Doing When They "Recommend Keywords"

To understand why this happens, it helps to understand what an AI tool is actually doing when you ask it for keyword suggestions.

AI language models generate responses based on patterns in the text they were trained on. When you ask for keywords related to, say, "nutritional therapy for gut health," the model produces phrases that sound like the kind of thing someone might search for in that context based on the patterns of language it has seen associated with that topic.

What it is not doing is checking search volume. It's not using how many people, actually typed any of those phrases into Google recently. It has no access to that information and no mechanism for finding it.

It's producing phrases that sound like searches but not phrases that are definitely searches.

"Sounds Like a Search" vs. "Is a Search"

Ask for keyword ideas around "skin therapy for hormonal acne" and you'll get back a list that includes things like "best treatment for hormonal acne," "how to clear hormonal acne naturally," "skin therapist for adult acne near me" and a dozen variations along similar lines.

Some of these will be genuinely searched but some will have actually have zero search volume.

Phrases that sound exactly like something a real person MIGHT type that nobody actually searches for.

The AI has no way of distinguishing between these categories.

It doesn't know if "how to clear hormonal acne naturally" gets searched thousands of times a month or if "natural remedies for hormone related chin acne in your thirties" never gets searched.

Why This Matters More Than It Sounds

It might seem like a minor issue if a few of the suggestions aren't great?

The problem isn't just that some or all suggestions are weak. It's that you have no way of knowing which ones work, without doing the research the AI didn't do.

When you actually run proper keyword research on a list of AI-generated suggestions and check real search volume data for each phrase that AI suggested, maybe one or two turn out to have meaningful search volume. The rest range from negligible to literally zero.

If you take that list at face value and build your website copy around it, you've optimised your copy for terms that nobody is typing into Google.

Your website copy would read fine. It might even read like it's been thoughtfully optimised because the phrases sound like real searches. But the actual effect on your search visibility would be non existent, because you've built around guesses rather than data.

The Confidence Problem

One of the things that makes this particularly easy to miss is how confidently AI tools present this information.

When you ask an AI tool for keyword suggestions, it doesn't hedge. It doesn't say "here are some phrases that sound plausible but I haven't verified whether anyone actually searches for them." It presents a professional-looking list, often with additional commentary about why each keyword might be valuable, what search intent it reflects and how it relates to your business.

All of that additional context is still guesswork.

AI tools can be a starting point for generating ideas to then validate.

But validating them is non negotiable.

As is finding the real search intent keyword phrases that AI will never recommend.

What Real Keyword Research Actually Involves

Real keyword research starts from a fundamentally different place: actual search data.

Tools designed for this purpose query databases that track real search behaviour: how many times specific phrases are searched, how that volume changes over time and how competitive a given term is, what related searches exist and how search intent varies between similar-sounding phrases.

This means that instead of generating a list of phrases that sound like searches, real keyword research identifies phrases that are searches with actual numbers attached. This phrase gets searched roughly this many times a month. This one gets searched more. This one, despite sounding good, gets searched essentially never.

From there, the research goes further than volume alone. It looks at search intent, not just whether a phrase is searched but what the person searching it is actually looking for and whether that matches what your business offers.

A high-volume keyword that doesn't match genuine buying intent for your services isn't necessarily worth optimising for, even if the volume looks impressive.

It looks at competition, how difficult it would realistically be to rank for a given term and whether there are lower-competition alternatives that represent a more realistic opportunity, particularly for smaller or newer websites.

And it looks at the specific language real people are using in their search queries that AI-generated guesses often miss entirely.

Why This Matters Specifically for AI Search Too

There's an additional layer to this given how search itself is changing.

AI search tools are themselves drawing on real search behaviour to determine what's relevant and credible.

Website copy that's been optimised around unvalidated keywords doesn't just fail to rank well in traditional search. It also fails to align with the actual language patterns that AI search tools are matching against when they're deciding which websites are relevant to which queries.

The Step That Gets Skipped And Why

Given all of this, why does the AI-generated keyword list get used so often?

Mostly because the alternative aka proper keyword research requires access to research tools, the knowledge of how to use them effectively and the time to validate and interpret the results.

Asking an AI tool for a quick list feels like a reasonable shortcut.

The problem is that it shortcuts the part of the process that actually determines whether the copy built around those keywords has any chance of working.

It's the equivalent of building a house on a foundation you haven't checked is solid. The house might look fine and the structure above ground might be excellent. But if the foundation isn't right, the quality of everything built on top of it doesn't matter as much as it should.

What This Means for Your Website Copy

If your website copy has been written or optimised based on a keyword list that came from an AI tool without that list being validated against real search data, your SEO game is probably a lot weaker than you think.

The gap between your website copy and what your ideal clients are typing into Google or asking AI tools might be much more than it appears because the phrases your copy was built around were never validated as real searches.

It means going back to the keyword stage and doing the research that should have happened before any copy was written at all.

Keyword Research As Standard In MY SEO Website Copy Packages

This is why keyword research isn't an optional extra in my premium website copy packages and blog packages. It's the foundation everything else is built on.

You can choose to work with me on positioning-only website copy with no optimisation for SEO or AI search.

But if you choose a website copy package with positioning and search optimisation, keyword research is a key stage of the prep phase.

Check out my website copy packages here.

Sally Aquire is a website and blog copywriter for wellness, beauty and skin professionals. She does proper keyword research as standard in every website copy and blog package so your website and/or blog copy is built on what people are actually searching for, not what AI thinks they might be.

Next
Next

You Worked With a Copywriter On Your Website & Weren't Happy With The Results But Did They Actually Do This?