How I actually use AI in my SEO process (and where I don't)

Here is the short version: I use AI for the absolute minimum amount of thinking possible, and I make every decision that matters myself. That probably sounds backwards for someone who uses AI every day, so let me show you what it looks like in practice - the parts I hand over, the parts I never do, and the step in the middle that stops the whole thing going off a cliff.

I am writing this partly because "we use AI" has become a marketing line that tells you nothing. Some people mean "AI writes everything and hits publish." I mean something much narrower and, I think, much more useful.

What I hand to AI

The tasks I delegate are the ones that are mechanical - lots of volume, clear rules, easy to check. Think of AI here as a tireless assistant doing the photocopying while I do the deciding.

• Grouping a raw keyword export into topic clusters so the shape of a market appears in minutes instead of an afternoon.
• Drafting title tags and meta descriptions across a whole site, which I then rewrite where they miss.
• Pulling a wall of Search Console or analytics data into a plain-English summary I can react to.
• Turning my outline - my structure, my angle, my points - into a first draft.
• Flagging obvious technical issues across hundreds of URLs at once: thin pages, duplicate titles, missing headings.

In every one of those, the AI produces raw material. It never produces the decision. That distinction is the whole game.

What I never hand over

Some things I keep entirely in human hands, because getting them wrong is expensive and AI has no way to know it is wrong.

Which keywords are actually worth chasing

AI ranks keywords by volume because that is the number it can see. It cannot tell that a low-volume term is where the money is. It is genuinely useful for the mechanical side of keyword work and useless at the commercial side - I have written about that split in using AI for keyword research. The value call stays with me.

Strategy and trade-offs

Should this be one page or three? Do we go after the competitive head term now or build authority first? Is this link worth pursuing? These are judgement calls that depend on the specific business in front of me. AI will give you a confident answer to all of them, and it is guessing.

Anything a reader will actually trust

Claims, statistics, the way a Google feature works, advice a business owner will act on - I check all of it, because AI hallucinates. It states false things with a straight face. I would rather be slow and right.

The manual override step

Between "AI produces something" and "it goes live" there is always a step where I read the output looking for the dumb thing. And there is usually a dumb thing - a hallucinated fact, a robotic internal link, a title that technically fits but reads like a machine wrote it, a suggestion that would cannibalise an existing page.

I treat every AI output as a confident first draft from someone who has never met the client. Useful, fast, and never the final word.

- Tom McSherry

This override step is not me being precious. It is the part that turns AI from a liability into a genuine speed-up. Remove it and you are just publishing a machine's guesses at scale. Keep it and you get the best of both - the AI's speed on the grind, a human's judgement on everything that counts. That balance is what I mean by human-in-the-loop SEO.

Why I won't use set-and-forget tools

There is a whole category of set-and-forget AI SEO tools built to remove the human entirely - point it at your site, let it publish and optimise on its own. I understand the appeal. I also know exactly what step they are removing: the override. When the AI decides to reshuffle your internal links or spin up forty thin pages, there is no one in the loop to say no.

So no, I have not automated myself out of a job, and I am not trying to. I have automated the photocopying. Whether a client is in Auckland or Adelaide, the work that moves their rankings is still a human weighing the specifics of their business - AI just clears the busywork out of the way so I can spend more time on it. I hold that line strongly, and I would be happy to be proven wrong the day a tool can make those calls better than a person. It has not happened yet.