Type a question into Google right now and there's a decent chance you never leave the search page. The answer sits right there, generated on the spot, with maybe two or three source links tucked into the bottom of it. Ten blue links used to compete for a click. Now one paragraph competes for a citation. That shift matters more than most content advice has caught up with. Ranking on page one used to be the finish line. Increasingly, the finish line is getting pulled into an answer that someone reads and never clicks through from at all. And getting pulled into that answer takes a different kind of writing than getting ranked ever did.
For twenty years, ranking well meant reverse-engineering an algorithm that was trying to guess what a human typed and wanted. That produced a specific kind of writing, one built around keyword placement and phrasing that matched whatever a person typed into the box. Length mattered too, since word count signaled thoroughness to an algorithm even when the extra words were just padding. None of that was really about the words themselves. It was about satisfying a system that stood between the writer and the reader, on the assumption that satisfying the system was the only way to reach the reader at all.
An AI answer engine isn't ranking pages. It's extracting claims. It reads through a pile of sources, pulls out the sentences that most directly answer the question, and stitches them into a response. Nobody scrolls past that response to see where it came from unless they specifically want to check.
That changes what counts as good writing in a fairly specific way. A sentence that gets pulled out of a paragraph and dropped into someone else's answer either holds up on its own or it doesn't. If a claim only makes sense next to the three sentences before it, it never gets picked. If it depends on a "however" two paragraphs earlier to be accurate, it gets misquoted or skipped entirely.
Writing for extraction means writing sentences that survive being ripped out of context. That's a real skill, and it isn't the one a decade of on-page SEO trained people to build.
A few things shift once extraction, not ranking, becomes the target.
Answer the question first. Content built for search engines often built up to the point slowly, partly because length signaled depth to the algorithm and partly because writers were taught to hook readers before delivering value. An AI system isn't hooked by anything. It wants the clearest possible statement of the answer, and it wants it early.
Make claims that are true in isolation. A lot of content hedges by design: "it depends," "in some cases," "generally speaking." That hedging protects the writer, but it also makes a sentence useless to something trying to extract a clean answer. Compare "email marketing can be effective depending on your list and industry" with "a small, engaged list will outperform a large, cold one." The second version might be wrong in some situation somewhere. That's fine. It's specific enough to be useful, and specific enough to get pulled into an answer. Vague ones get passed over simply because there's nothing to grab onto.
Let one section stand on its own. A page written for a human reader can build an argument across ten paragraphs, each one leaning on the last. A page written to be extracted needs sections that work as standalone units, so any one of them could be the piece an AI system pulls without losing its meaning. That's a structural change as much as a sentence-level one, and it's probably the hardest habit to build, since most writers were trained to make paragraphs depend on each other. SEO and AI-answer writing overlap more than they conflict. Clarity was always the goal, even when the tactics for getting there were clumsy. Specificity has always outperformed vagueness in both worlds. Expertise still shows, whether a machine is extracting a sentence or a person is scanning a page for the part that answers their question.
The overlap is bigger than the split, and what changed was emphasis more than values. Writing that was already clear and direct probably works fine in both systems. Writing that leaned on length and keyword density to get by is the writing that's now exposed.
The practical move is to write the clearest possible answer first and think about structure after that, rather than chasing a new checklist for AI citation or throwing out SEO altogether. Say the thing plainly, make the claim specific enough that it could be wrong, and don't make a reader, or a machine, dig through three paragraphs of throat-clearing to find out what you actually think.
That was probably good advice before any of this started. It just wasn't the advice that got rewarded. Now it is.
Have you noticed which of your own pages show up in AI answers, and whether they're the ones you expected?