Search stopped being ten blue links
For twenty years, search meant typing words into Google and picking from a list. That is no longer how a large share of buyers behave. They ask a chatbot a question in full sentences, get one answer, and act on it. If your client is not in that answer, they may as well not exist for that buyer.
This changes the job completely. Ranking third for a keyword used to be a win. Now the question is whether an AI engine quotes you when someone asks about your subject, and the things that earn a citation are not quite the same as the things that earned a ranking. That gap is what you will spend your time on. It is a genuinely new field, which means nobody has ten years of experience in it and a sharp beginner can be genuinely useful within months rather than years.
What actually earns a citation
The honest answer is that it is still being worked out, publicly and by us, which is exactly why this is an interesting place to start. Anyone claiming certainty here is selling something. What we can see from our own sites and our clients’ is that certain things repeat often enough to act on.
Self contained sections that fully answer one question do better than long meandering pages. Specific figures with a named source do better than vague claims. Questions as headings, answered immediately underneath, do better than clever headings that make the reader work out what the section is about. Structured data that correctly describes what the page is helps a machine understand it without guessing. None of this is a trick, it is writing that respects the reader, which is why it also converts better for humans.
The part where the data argues with you
You will form a theory about what people search for. The data will frequently disagree, and learning to accept that quickly is most of the job. The alternative is defending an idea while the traffic quietly goes elsewhere.
We recently found that one of our own sites was ranking well for a phrase that expired every month, so the traffic vanished on schedule and the page looked broken when it was only stale. Another time we found a whole set of pages with meta descriptions cut off mid sentence, every one of them inside the recommended length range, because a length check cannot see a severed sentence. Both were found by looking rather than assuming, and both had passed automated checks. You will be taught to look, and to distrust any check that could not have failed.
Writing that sounds like a person
We have firm rules about voice and you will learn them fast. British English. No em dashes. No filler openers. Never invent a number, ever, not even one that seems harmless, because a client can be asked to defend it in a meeting you are not in.
That last rule is the one people find hardest. If you cannot trace a figure to a source, you leave it out or you go and find the source. We would rather publish a page with fewer numbers than a page with one we cannot stand behind. This is not pedantry, it is the difference between content that builds trust and content that quietly destroys it the first time somebody checks. It also makes the writing better, because you stop reaching for a statistic to prop up a sentence that should have been clearer.
What a month looks like
Early on you will be given a page to research and write, with the questions already identified, so you can concentrate on the writing and the structure without also having to work out the topic. Fairly quickly you will do the research yourself, using real search data rather than guessing what people ask.
Later you will take a live site, read its search data, work out which pages are slipping and why, and propose what to do about it. Then you will do it and watch what happens. Seeing your own change move a number, or fail to, is the fastest way to actually learn this field rather than reciting it. It also teaches you patience, because search moves slowly and the temptation to change five things at once makes the result impossible to read.
Why our own sites are the training ground
We publish for ourselves as well as for clients, which means you can experiment on something real without a client bearing the cost of your learning. If a change goes badly on our page, we learn something. If it goes badly on theirs, they lose money.
We also keep a written record of every SEO mistake we have made, including the ones that cost us weeks, and you will read it early. We run our own AI SEO agent that works on our sites every week, and you will work alongside it. That is a strange and useful experience, because you quickly learn what a machine is genuinely good at, where it is confidently wrong, and which parts of this job still need a person who can read a page as a reader rather than as a crawler.
How to apply
Send one email with a short piece of writing you are proud of, a link or a paste is fine, plus one paragraph on something you looked up recently and what surprised you about the answer. End it with the sentence “I want the AI SEO and content role”.
We read the writing sample first and we read it properly. It does not need to be about marketing or technology, and we would honestly rather it were not. It needs to be clear, and it needs to sound like you rather than like something written to impress. The paragraph about what surprised you is the second test: it shows whether you go and check things, and whether you are willing to be interested in an answer you did not expect. Both are the daily work here.
