In July my job title changed to AI Marketing Engineer. I was a few weeks short of 20 years working in SEO, and I'll admit I was quite glad to see "SEO" leave my job title (not because I've fallen out with it, it just doesn't describe what I do any more).
It's tempting to tell that as a story about starting over. It doesn't feel like that to me though. The more I've looked back at what I was doing in 2007, 2012, 2016, the more it feels like I was being trained for this the whole time, I just didn't know it. Which is a slightly funny thought, given what I'm tinkering with now (I've been training a small add-on for an open image model on centuries-old woodcuts, so a hobby project leans less on other people's copyrighted art - not training a model from scratch, but giving it a different basis to work from).
I think technical SEOs in particular have a really good grounding for the kind of building work that AI has opened up. I want to be careful with that - I don't think we're uniquely suited to it, and I'm sure there are plenty of other routes in that end up in much the same place. But it's the route I took, so it's the one I can talk about.
So I went back through my old blog posts (the ones I could recover, which was an experience) to see if the threads were really there or if I was just telling myself a nice story.
2007 to now
Making things machines can read (and understand)
My first ever live website was built entirely out of images. Every word on it was a picture of a word. I couldn't work out why nobody could find it, so I stuffed the meta tags full of keywords and installed a hit counter that told me precisely nothing. That bafflement is more or less how I got into search in the first place.
By 2007 I'd learnt enough to write, a little smugly, that "if the text is in an image, then a search engine cannot read it." Obvious now. It wasn't obvious to a lot of people then. (Although AI can demonstrably read text in images these days, so I wouldn't be surprised if search engines can too - it's probably just a costly overhead to do that on every page they crawl.)
A lot of what came after was a more sophisticated version of the same problem. Not just helping a machine read a page, but helping it understand what the page was about, who it was for and how it related to everything else. I wrote a post in 2012 called "Adventures with Rich Snippets" and ended up talking about rich snippets at Think Visibility the following year. I was testing review markup, recipe markup, authorship, all sorts, and measuring what it did to click-through. There was a kind of contract with Google back then - you did the work of explaining your content properly, and you'd often (not always) get something visible back for it.
hreflang was the same idea for international sites: telling search engines which version of a page was meant for which country and language. More on the tool we built for that in a bit.
When I joined Sage in 2016, one of the first conversations I had was about where a new China website should live. The obvious answer was /ch, except that was already the Swiss site. I suggested we just use the ISO country and language codes that hreflang already uses as our folder structure. That became the principle for consolidating hundreds of sites, country sites, product sites and help sites into one structure, and it's still how things are organised today. Alongside that came years of domain work and redirects, which I've often described as being handed a plate of tangled spaghetti - each strand a domain, a hostname or a redirect chain, and the job is finding where each one starts, where it ends up and whether it should still be on the plate at all. Not glamorous, but it's all about making a big messy web presence make sense to a machine (and to people, to be fair).
And now the machines doing the reading are AI crawlers and models. This site is built for them as much as anyone - plain prerendered HTML, structured data on everything, and none of the default AI bot blocking that a lot of sites are running without realising it. It's the same instinct really, just a different thing doing the reading. I'm less sure what the contract is this time round, to be honest. That's probably a post of its own.
2008 to now
Not trusting the number on its own
If there's one habit that has followed me through my whole career, it's not trusting a number in isolation. Where did it come from, how was it measured, what else moved at the same time?
In 2008 I wrote that as personalisation and geo-targeting grew "people will get very different results for rankings". A rank report was already starting to be a bit of a fiction. A year later I was writing up Matt Cutts' advice to look at your server logs rather than getting hung up on ranking reports, and I've been looking at log files ever since (these days mostly to see how the AI bots are crawling).
Also in 2009, I wrote about Wolfram|Alpha, a new "answer engine" that kept people on its own site rather than sending them anywhere. My conclusion was that it had no impact on SEO for now, but "whether or not this changes and evolves over time remains to be seen". Seventeen years later... well.
Whether or not this changes and evolves over time remains to be seen.
At Sage a lot of my later SEO years went into joining data up - logs, Search Console, analytics, CRM - to try and tie what we did to what the business actually got out of it. Some of that was inspired by Chris Green's SEO metric chain (you wouldn't believe how many conversations I've had in real life about that blog post). We were trying to prove cause and effect, and I don't think we ever fully nailed it. Causation has always been hard in search. Throw in cookie consent taking a chunk of the data away, and now AI search, where a lot of the influence happens somewhere that never reaches your website at all - even though the log files show plenty of evidence that something is calling that content. That's more correlation, not causation, and it's not definitely being shown to a human either, with all the prompt trackers out there asking questions that no human ever sees the answers to.
None of that has made me stop measuring. It's just made me a lot more careful about what I think a number is telling me, and that turns out to be quite a useful instinct when you're working with tools that will confidently tell you all sorts of things.
2012 to 2025
The ideas I couldn't build
For most of my career I've had more ideas than I had the skills to build.
In 2012 theMediaFlow (the agency I was a director of) had a hreflang sitemap generator built. I specced it, and a small local dev firm built it. It was one of the first tools of its kind, you just uploaded a CSV, and it got shortlisted for Innovation at the UK Search Awards in 2013. I'm still proud of it. But I couldn't have built it myself. A year or two before that, I'd been talking to Screaming Frog about their SEO Spider while it was still in development, because I was looking for something that could check how the canonical tag had been implemented across a whole website. Getting early access really helped me get to the bottom of a sitewide canonical issue I was wrestling with at the time. Again, I had the problem and the ideas, but somebody else did the building.
I even wrote in January 2012 that whilst a lot of SEOs were talking about learning to code that year, "for me maximizing and improving my Excel usage is a considerably higher priority." I don't think I was wrong at the time. It's quite funny to read now though.
What changed was AI. In July 2025 I gave up a weekend (and quite a few evenings) to build a keyword research tool so I could apply for an AI squad at work. A week before, I'd never built a web application of any kind. I got in. The squad started out with everyone building their own thing, and I ended up pushing for us to build one joined-up workflow instead, which eventually became a tool that took some of our asset production from weeks down to hours.
That's the engineer part of my job title. I build in partnership with AI, and it writes a lot of the code, but I can debug things, test things, break things and fix things now, and I just keep building. That hreflang tool that cost a few thousand pounds and took a dev team? I think we could probably knock it up in Python in a few minutes now.
2008 to now
How I've always worked
Some of what carried over isn't technical at all.
Back in 2009 I was writing that there was "no point being mysterious" about what you were doing. SEO had a bit of a reputation for being dark arts, and I always preferred explaining what I was doing and why. I still do. Teaching people to fish rather than doing it for them is most of how I try to help people at work now.
I also found myself running the SEO team at Vertical Leap within a year or two, somewhat by default, long before I had a title that said so. In 2010 I wrote a post called "Becoming a Leader" which said "a title isn't very important - what is important is how you fulfil that role." I'd still stand by that. Do the job you want to be doing, and the title tends to follow eventually.
At Sage I've never been a line manager (other than as temporary cover). I've helped build the team, mentored people, spotted people in other parts of the business who had a real interest in SEO, set standards and built the foundations the team built on. I've always felt I add more as a senior person who does the work than as a manager, and luckily my bosses have agreed.
2008, and again now
Who is Peter Handley?
In January 2008 I was playing with Wikia Search, a short-lived search engine that let anyone add a short description to the results for a search. I added a line to the one for my own name: "Peter Handley is a search engine marketer working for Vertical Leap".
Nearly 20 years later, I've rebuilt this website partly so that when somebody asks an AI "who is Peter Handley?", it has something accurate to go on. It's kind of the same thing I was doing in 2008, just with a different machine (and, thankfully, a different answer).
I don't know exactly what the next 20 years look like. But I'm pretty sure the curiosity, and the habit of questioning things and trying to make them make sense to machines, will keep being useful. It's served me alright so far.