Strategy
What an AI SEO Strategist Actually Does All Day
Every SEO on LinkedIn added AI to their title in about eleven months. Almost none of them changed what they do on a Tuesday. Here is what the work looks like when the change is real.
Two years ago the SEO strategist job was reasonably stable. You owned a keyword map, a content calendar, a technical backlog, and a rankings dashboard. The arguments were about priority and resourcing, and the scoreboard was a number everyone had agreed to look at.
Then the scoreboard split in two, and one half of it became something you cannot query.
The title spread fast. Adding AI to a job title takes eleven seconds. What follows is what the work actually looks like when the change is real, written partly so you can hire for it and partly so you can tell whether you are doing it.
The job in one sentence
An SEO strategist decides what to work on and why. Everything else in the role is in service of that.
That distinction matters more than the org chart suggests. Specialists and executives do the work: publishing, fixing, optimizing, reporting. The strategist owns the diagnosis and the priority order, and has to be able to defend both to a room that would rather spend the sprint on something else.
Which is why the role got harder. The diagnosis used to be available from tools. Now half of it is not.
The four things the job is now
Diagnosis you run by hand. There is no rank tracker for answer engines that anyone should fully trust yet. So you build a prompt set, run it across ChatGPT, Perplexity, Claude, and Google AI Overviews, and record what comes back. Which brands are named. Which sources get cited. What those pages have that yours does not.
It is manual, it takes half a day a quarter, and it is the single most valuable half day in the role. Everything downstream gets its priority order from it.
Evidence design. This is the genuinely new skill and the hardest one to hire for.
Ranking rewarded coverage: say enough about a topic, in enough depth, with enough internal support. Citation rewards something narrower. An engine quotes a source when the source states something specific, attributable, and checkable. A number with a method behind it. A named person taking a position. A comparison somebody actually ran.
So the strategist’s content job changed from “what should we cover” to “what can we say that is worth quoting.” Those produce completely different briefs. The first produces a 3,000-word guide that reads well and gets cited by nothing. The second produces 900 words with one defensible claim, and it shows up in answers for two years.
Specs, not requests. Most of the technical work that moves AI visibility is schema, structure, crawler access, and rendering. All of it goes through engineering, and engineering is busy.
The strategists who get things shipped write specs an engineer can pick up without a meeting: what to change, in which template, what the output should look like, and how it will be verified. The ones who send a Slack message saying we should add schema stay blocked forever.
Measurement that admits what it cannot see. AI referrer data is partial and will stay partial. Some engines pass no referrer. Attribution is genuinely broken here, and everyone senior enough to matter has noticed.
The credible move is to say so, then report the things you can defend: citation share by cluster, branded search lift, conversion rate on the AI traffic you can see, and assisted revenue. Strategists who present a clean-looking AI attribution model lose the room the first time someone asks how it works.
What shrank
Two things, and they were the comfortable ones.
Rank reporting. Still useful, no longer the headline. A dashboard of position changes describes an increasingly partial view of a query landscape where a third of the clicks were absorbed above the results.
Keyword mapping as a primary artifact. Keywords still matter for classic search. But answer engines fan a single question into several internal queries and synthesize across them, which means the unit of planning shifted from the keyword to the question and the entity. Strategists still doing spreadsheet-first keyword mapping are optimizing for the half of the environment that is shrinking.
How to hire one
Four filters, in order of how much signal they carry.
Make them read a live result. Open an answer engine, run a query in your category, and ask why those sources were chosen. A real strategist will look at the cited pages and reason out loud about structure, specificity, and authority. Someone who rebranded will talk about quality content and user intent without opening anything.
Ask what they stopped doing. Anyone who has genuinely adapted has killed something: a report nobody read, a content format that stopped working, a tool they cancelled. If the answer is “we added AI optimization on top of everything else,” they have added a slide, not a practice.
Ask for a spec they wrote. Not a strategy deck. An actual ticket or requirements doc that went to engineering. You will learn in thirty seconds whether they can make anything happen in a company.
Ask what they would not do. Strategy is mostly refusal. A strategist with no examples of talking a stakeholder out of something is a coordinator with a better title.
The uncomfortable part
A fair amount of what made someone senior in this field five years ago has depreciated. Deep knowledge of a rank tracker’s quirks, mastery of a crawler’s configuration, the ability to build an elaborate keyword model. Those were real skills that produced real results, and they are worth noticeably less now.
What appreciated: the ability to decide what is true, say it clearly enough that a machine can quote it, and get engineering to ship the thing that makes it visible.
That is not a technology skill. It is closer to editorial judgment plus product sense, which is an awkward thing to discover mid-career if you got here through spreadsheets.
The upside is that it is learnable, and the people learning it right now are competing against a field that mostly updated its LinkedIn headline and went back to work.
The week, concretely
Job descriptions are useless for this. Here is roughly how the time distributes for a strategist owning one mid-sized site, based on what the work actually demands rather than what a calendar invite claims.
Monday, mostly reading. Not dashboards. The things your buyers read: the subreddit, the industry newsletter, the two competitors who publish well, the support tickets from last week. Strategy comes from knowing what people are confused about, and that information does not live in a keyword tool.
One block for engineering. Standing time with whoever ships your changes. Long enough to write a spec together, short enough that nobody resents it. Strategists who skip this and communicate only in tickets ship about a third as much.
One block for editorial. Reviewing drafts against a single question: what in this piece is quotable. If nothing is, the fix is not a rewrite, it is a missing claim, a missing number, or a missing opinion. That conversation is uncomfortable and it is the job.
A recurring hour on measurement. Not building the report. Checking whether the report still describes reality, which it stops doing every few months as engines change what they surface.
Half a day a quarter on citation testing. Blocked in the calendar, done properly, written up. The rest of the quarter’s priorities fall out of it.
What is not on the list: daily rank checks, and reformatting a monthly deck that three people skim.
Tools that still earn their place
Plenty of the stack survived. What changed is which part of the job it serves.
Search Console is more useful than before, not less. Query-level impression data is one of the few honest signals about how a page behaves when an AI Overview sits above it. Watch impressions holding while clicks fall, which is the signature of being summarized rather than being replaced.
Crawlers still find template defects better than anything else. Their weakness was never the crawl, it was the report format.
Log files got more valuable, because they are the only reliable way to see what the AI crawlers actually fetch. Configuration files tell you what you intended. Logs tell you what happened.
Rank trackers dropped from primary scoreboard to supporting evidence. Keep one, stop opening it daily.
The engines themselves became a research tool. Running your own category prompts weekly is now a normal part of the job, and it costs nothing but attention.
What to learn if you are moving into this
Three things, in priority order, based on what actually blocks people.
Learn to write a spec. Not a strategy document. A ticket an engineer can execute without asking a question: current state, desired state, which template, how to verify. This is the fastest way to double your effectiveness and it takes about a month to get decent at.
Learn enough structured data to argue about it. You do not need to write JSON-LD from memory. You need to know what an entity is, why identity fields matter, and what happens when markup and page content disagree. That is a week of study and it separates the strategists who can direct technical work from the ones who defer to whoever speaks last.
Learn to state a claim. The hardest one, and the least technical. Most SEO writing is hedged into meaninglessness because hedging is safe. Answer engines cannot quote a hedge. Practise writing sentences that could be wrong.
If you would rather bring that judgment in from outside than build it internally, our AI SEO consulting page covers how that engagement works.