'Set and forget' AI SEO tools: why I don't trust them (yet)

I don't trust 'set and forget' AI SEO tools, and I'll be straight with you about why before I've earned the right to sell you anything: because every version I've tested or watched a client get burned by has done one of two things. It has either produced garbage that a human would be embarrassed to publish, or it has produced something that looked fine on day one and became an expensive mess to untangle three months later. Neither of those is 'set and forget'. That's 'set and pay for it later'.

This is a post about autonomous AI SEO tools - the ones that promise to research, write, publish, and internally link your whole site while you get on with running your business. I want to talk a few of you out of buying one. And because I'd be a hypocrite if I didn't, I'll also answer the obvious question at the end: aren't you using AI to write this very blog? Yes. And the difference is the entire point.

What 'set and forget' actually promises

The pitch is seductive because the pain is real. If you run a plumbing business in Hamilton or a physio clinic in Geelong, you don't have time to write content, build internal links, and keep an eye on which of your pages are quietly competing with each other. So a tool that says 'connect your site, we'll handle the rest' sounds like exactly what you want. Turn it on, walk away, come back to more traffic.

The trouble is that SEO isn't a mechanical task with one right answer. It's a series of judgement calls, and judgement is the one thing these tools are worst at. Deciding what to write, deciding which page should own which keyword, deciding how pages should link to each other and with what words - those aren't data-entry jobs. They're the actual work. When a tool automates the judgement instead of the labour, it isn't saving you the hard part. It's doing the hard part badly and hiding the evidence.

The failure you don't see on day one

Here's what makes these tools genuinely dangerous rather than just useless: the damage is invisible at first. On day one, you've got twenty new pages, they're all indexed, everything looks productive. The problem shows up later.

The most common version is cannibalisation - where two or more of your own pages compete for the same keyword, so Google can't tell which one to rank and ends up half-ranking both. A human doing this deliberately would notice they'd already written about 'emergency plumbing' and consolidate. A tool publishing at volume has no reliable way to judge that at mass scale without oversight. So it keeps producing near-duplicates, and your own pages start eating each other. I've written more about why publishing more pages can actually make your rankings worse, because it's counterintuitive and it catches people out constantly.

It isn't 'set and forget'. It's 'set, and pay someone to untangle it later'.

The second version is the thin-content pile-up. AI can produce a thousand words on any topic in seconds, but volume isn't value. You end up with a hundred pages that each say a little about a lot and nothing much about anything - and a search engine that increasingly reads that as a low-quality site. The hidden cost of AI articles at scale is exactly this: the bill doesn't arrive when you publish, it arrives when someone has to work out which pages to delete, merge, or rewrite to stop dragging the rest of the domain down.

The anchor text and linking problem

There's a more specific red flag I look for the moment someone shows me a tool. Does it give you control over anchor text - the actual words inside a link - and internal linking? If it decides those for you and won't let you override them, that's a pretty big red flag for me. Anchor text and internal links are two of the strongest levers you have over what ranks. Handing them to a black box means handing away the steering wheel. A tool that won't let you drive is one I won't trust with the car.

How I actually use AI (and where I refuse to)

None of this makes me anti-AI. I use it every day. The distinction I care about is human-in-the-loop: AI does the mechanical lifting, a human makes every judgement call. I basically use AI for the absolute minimum amount of 'thinking' in any process possible, with a manual override step to fix it when it does something dumb or hallucinates.

• AI is good at: reformatting, drafting from a brief I've written, summarising research, pulling data into a table, handling repetitive labour at speed.
• A human decides: what topic is worth writing about, which page owns which keyword, what the anchor text says, how pages link together, and whether anything is competing with something we already rank for.
• There is always an override step - a point where I can catch the tool doing something silly before it goes live, not after.

If you want the full breakdown of where I draw that line, I've written it up in how I actually use AI for SEO and the broader case for keeping a human in the loop. The short version: the AI is a very fast, very confident junior who needs checking. It is not a consultant.

'But aren't you using AI for this blog?'

Yes. I'd be dishonest to pretend otherwise, so let me be specific about how, because the how is the whole argument.

A human - me - decides what each post is about, what opinion it argues, which examples make the point, and every internal link and anchor in it. AI does the mechanical work: turning my notes and structure into clean prose, tidying formatting, keeping things consistent. It never decides what I think or where a link points. That's the difference between a tool that assists a human's judgement and a tool that replaces it. This blog is produced human-in-the-loop, on purpose, as a working example of the thing I'm recommending.

And to be fair to the other side: I hold this view with humility. The tools are improving fast, and I'd be happy to be proven wrong. The day an autonomous tool can reliably judge cannibalisation, write with genuine control over its own linking, and flag its own mistakes before publishing, I'll change my tune in public. I don't know what I don't know. But 'yet' is doing a lot of work in this post's title, and today it's still doing it honestly.

One last thing worth saying: before you spend money on any tool at all, it's worth checking whether the activity is even right for your business. For a lot of local operators, more blog content isn't the lever they think it is - I've argued that plainly in whether blogging is a waste of time for local businesses. Automating the wrong task faster is still doing the wrong task.

And there is a version of this that is not about SEO at all. A plugin you install has code execution on your live site, and a lot of the new AI plugins were built by people who are not developers - which is a different and less discussed exposure: AI WordPress plugins and the security risk nobody is talking about.

If you'd still like to use AI for your website content, just with you rather than a tool holding the wheel, the structure I'd build around it first is set out in using AI for your website content without wrecking it.