The SEO that still works when half of search is a robot

I rewrote this post because the version I had up was the kind of generic SEO checklist you’ve read a hundred times — and because search itself changed underneath it. There’s an AI-generated answer sitting on top of the results now, and it’s quietly eating the clicks that used to belong to position one. So this is less “tips” and more “what I’d actually do,” in the order I’d do it.

The throughline is the same one I bring to operations work: the metric everyone watches is usually the one that’s lying. More on that at the end.

1. Be the source the AI answer cites

The game is no longer just ranking — it’s being the page a language model pulls from when it writes the answer box. That rewards the same things good content always rewarded, just more strictly: a clear, direct answer near the top of the page; real structure (headings that match how people ask); and specifics a model can lift cleanly. Vague, hedgy, “it depends” content doesn’t get cited. Confident, concrete, well-organized content does.

2. Structured data, for real this time

Schema markup went from “nice to have” to “this is how machines understand your page.” Organization, Person, Article, FAQPage, BreadcrumbList — the ones that fit your content. It’s cheap, it’s mechanical, and it’s one of the few things fully in your control. I add JSON-LD to every site I build now as a default, not a flourish.

3. Core Web Vitals, because slow still loses

Google’s field metrics are a real ranking input and, more importantly, a real user input. The big shift: INP (Interaction to Next Paint) replaced FID — it measures responsiveness across the whole visit, not just the first tap. Largest Contentful Paint and Cumulative Layout Shift still matter. If your site janks when someone scrolls or taps, you’re losing people before the content even gets a vote. (I took one site from a 47 to a 98 — that’s its own post.)

4. Topical depth beats keyword sprinkling

One thorough page that genuinely covers a topic outperforms ten thin pages chasing variations of the same phrase. Search engines — and the models reading them — are good enough now to reward actual depth and punish padding. Write the page you’d want to find. Delete the three you wrote to game the algorithm.

5. The boring technical foundation

Crawlable architecture, a real sitemap, clean canonical tags, no orphan pages, HTTPS, mobile parity. None of it is glamorous and all of it is load-bearing. This is the plumbing — when it’s wrong, nothing above it works, and you’ll waste months blaming your content.

The metric that’s lying to you

Here’s the operations lesson hiding in an SEO post. The number most people obsess over — sessions, or “traffic” — is the easiest one to inflate and the easiest one to misread. I’ve audited funnels where the headline traffic looked healthy while the actual qualified, converting visits had quietly collapsed. Bots, brand searches, and one viral-but-useless page can paper over a real decline for months.

So don’t optimize for the vanity number. Trace it back: which pages bring people who do the thing you care about? Optimize those. Delete or fix the rest. Same method I’d use on any process — question the metric, delete the noise, then go make what’s left faster.