Using AI for keyword research: useful, with a big caveat

AI is genuinely useful for keyword research - as long as you understand which half of the job it is doing. It flies through the mechanical parts: generating variations, clustering, sorting. It is close to useless at the part that actually decides whether a keyword is worth your time - the commercial judgement. Lean on it for the first and you save hours. Trust it with the second and you will chase traffic that never turns into customers.

Keyword research is picking the search terms you want to be found for. The trap is thinking it is about finding the biggest numbers. It is not. It is about finding the terms that bring people who are ready to buy - and that is exactly the bit AI cannot see.

Where AI genuinely speeds things up

The mechanical parts of keyword research are repetitive and rule-based, which is precisely where AI shines. It is like having someone sort a thousand-card deck into neat piles while you decide which piles matter.

• Expanding a seed term into hundreds of variations in seconds.
• Clustering a messy keyword export into topic groups so you can see the market's shape.
• Grouping terms by likely intent as a rough first pass - informational, commercial, navigational.
• Spotting phrasing and long-tail variants you would not have thought of.
• Tidying a raw data dump into something you can actually read and sort.

This is real, hours-saving help. If keyword research used to take you a day of spreadsheet wrangling, AI can get you to the interesting part by morning tea. No complaints from me on that front.

The big caveat: it can't judge value

Here is where it falls over. AI ranks keywords by the numbers it can see, and the number it loves most is search volume. But search volume is a weak signal for how much a keyword is actually worth - I have made that case in full in why search volume is a weak signal.

Picture two keywords. One gets 8,000 searches a month: "how does SEO work." The other gets 70: "SEO consultant Christchurch." AI will point you straight at the 8,000 because bigger looks better. But the first is mostly students and curious browsers, and the second is a business owner in Christchurch with a budget, looking to hire this week. The small number is worth vastly more. AI has no way to feel that difference.

AI weights keywords by volume. You have to weight them by business value. Those two numbers are often pulling in opposite directions.

- Tom McSherry

Two more things AI gets wrong

Intent, past the obvious

AI can sort keywords into broad intent buckets, but it misses the subtle ones that decide whether a page converts. "Best CRM" and "CRM pricing" look similar to a model. To a human they are two different buyers at two different moments, needing two different pages. Miss that and you write the right page for the wrong person.

Cannibalisation

Ask AI for a content plan and it will happily suggest a separate page for near-identical terms. Now you have several pages fighting each other for the same search, and Google is not sure which to rank. This is how adding more pages can make rankings worse. AI does not feel that tension - it just sees more keywords and suggests more pages.

How to actually use it

The workflow that works is simple: let AI do the mechanical grind, then you do the weighting. Have it generate and cluster the list. Then go through and score each cluster by business value - how likely is this searcher to become a customer, and how much is that customer worth? That scoring step is human, every time. It is the same override step I keep in every part of my process, which I have described in how I actually use AI for SEO.

If you want the fundamentals underneath all this - especially for a smaller business where every keyword has to earn its keep - start with keyword research for small business.

So: useful, with a big caveat. AI is a brilliant research assistant and a poor strategist. Let it clear the busywork, keep the judgement for yourself, and you get the best of both. Hand it the judgement too, and you will end up ranking for a lot of words that never pay the bills. I would be happy to be proven wrong, but I have not yet seen AI pick the 70-search keyword over the 8,000 - and the 70 is usually the one worth having.