The technical SEO checklist that still matters in an AI-search world

There's a narrative going around that technical SEO is legacy work — that the real game now is getting cited inside ChatGPT or showing up in an AI Overview, and that crawl budgets and canonical tags are yesterday's problem. That's backwards. Every retrieval-augmented system still has to fetch, parse, and trust your pages before it can cite them. Technical SEO is the plumbing underneath AI visibility, not a separate track from it.

Start with what a crawler can actually see

Before anything else, confirm the basics haven't quietly broken:

1. Robots.txt isn't blocking pages you want found. A single misplaced Disallow can hide your entire blog from every crawler, including the ones feeding AI retrieval indexes.
2. XML sitemaps are current and submitted. A sitemap that still lists pages you deleted two years ago, or omits ones you published last month, is actively misleading crawlers about what's worth fetching.
3. Canonical tags point where you mean them to. Duplicate or conflicting canonicals split authority across near-identical URLs and confuse which version should be treated as the source of truth.
4. Status codes are honest. A "deleted" product page that quietly 200s instead of 404ing or 410ing keeps getting recrawled and keeps diluting your site's signal.

Core Web Vitals: still a ranking factor, now also a trust signal

Page speed and stability were never just about pleasing Google's algorithm — they're proxies for whether a site is well-maintained. A page that takes nine seconds to become interactive, or that visibly shifts content while loading, reads as neglected to both users and crawlers.

• Largest Contentful Paint (LCP) under 2.5s
• Interaction to Next Paint (INP) under 200ms
• Cumulative Layout Shift (CLS) under 0.1

These thresholds haven't moved. What's changed is that a slow, unstable site now has two audiences to disappoint: the human bouncing off it, and the crawler deprioritizing it on the next pass.

Schema markup is doing more work than it used to

Structured data used to be mostly about rich snippets — star ratings, breadcrumbs, that kind of thing. It's now one of the clearest ways to hand a retrieval system pre-parsed facts instead of making it infer them from prose. FAQPage, Article, Product, Review, and LocalBusiness schema all give AI systems structured shortcuts to the same information your copy is trying to convey.

If your content answers a question, mark it up as a question. Don't make a model guess where the answer starts and ends.

The audit that actually matters

A technical SEO audit is only useful if it turns into fixes, not a PDF nobody reads. The pattern worth building toward:

• Crawl errors get triaged and resolved on a schedule, not discovered once a quarter.
• Broken internal links get caught before they compound.
• Schema gets validated against Google's Rich Results Test and Schema.org's own validator, not just pasted in and forgotten.
• Every fix ships — as a deployed change or a reviewable pull request — instead of sitting in a backlog labeled "someday."

None of this is glamorous. It's also the reason some sites get cited confidently by AI assistants and others, with objectively better content, don't get retrieved at all.