PB
Available
Blog
AI workflow September 19, 2026

How I Do SEO With AI Agents (and Where They Get It Wrong)

SEO is full of work that's mechanical, repetitive, and spread across thousands of pages — which is exactly the shape of task an AI coding agent is built for. It's also full of judgment calls an agent has no business making. Here's how I split the two on a site with several thousand pages.

What agents are genuinely good at

The mechanical layer. When an audit flags a hundred pages with the same problem — a missing meta description, a schema field that should be an absolute URL, a broken internal link, a page that got too heavy to crawl — an agent can apply the fix consistently across all of them in one pass. On this site I've had agents paginate a five-megabyte index page down under the crawl limit, make every product's structured data valid, keep the sitemap in sync as pages come and go, and fix crawl errors in batches. That's slow, error-prone work by hand and fast, consistent work for an agent.

Internal linking is the standout. Instead of hand-linking posts, you write one rule and let the build apply it everywhere — routing generic informational traffic toward the pages that actually convert:

internal linking, from a rule applied at build
// Route generic search traffic to the pages that actually earn money.
// A small rule, applied at build across every post, beats hand-linking.

const SERVICE_LINKS = [
  { match: /shopify|storefront|checkout/i, href: "/shopify-experts/",  text: "Shopify help" },
  { match: /wordpress|web design|website/i, href: "/website-design/",  text: "website design" },
  { match: /\bai\b|automation|chatbot/i,   href: "/ai-consulting/",  text: "AI consulting" },
];

for (const post of posts) {
  const rule = SERVICE_LINKS.find(r => r.match.test(post.body));
  if (rule) addContextualLink(post, rule);   // one relevant internal link, not ten
}

The same mindset covers programmatic pages, letting real analytics data drive on-site decisions (I recently reordered a 2,400-tool index so the most-visited tools lead, straight from the traffic numbers), and pinging search engines the moment content changes. All repeatable, all a good fit for an agent.

Where they get it wrong

Strategy and intent — every time. Ask an agent to "improve SEO" and left unchecked it will generate thin, near-duplicate landing pages that look like a penalty waiting to happen, invent schema properties that don't exist, and optimize copy until it reads for a crawler instead of a person. It doesn't know which pages are worth ranking, what a searcher actually wants, or where the line is between thorough and spammy. It has no taste. That part stays human, and it's the part that decides whether the mechanical work pays off or gets you filtered.

The split that works

Treat the agent as the hands and keep the head. You decide the strategy — which pages matter, what intent each one serves, where the quality bar is. The agent executes the mechanical work under that plan, at a scale you couldn't match by hand. Then you verify: check the rendered schema in a validator, confirm the new links actually resolve, and make sure every generated page earns its place instead of padding the index. Done that way, agents make SEO faster without making it worse — which, if you've watched AI "SEO content" tank a site, is not a given.

Frequently asked questions

What SEO work are AI agents actually good at?

Mechanical, repeatable work across many pages: fixing crawl errors, adding and validating schema, keeping sitemaps correct, and wiring internal links from a rule. An agent applies a fix to thousands of pages consistently, which is where humans get slow and inconsistent.

Where do AI agents get SEO wrong?

Strategy and intent. They will happily generate thin programmatic pages, invent schema fields, and over-optimize copy until it reads for robots. They do not know which pages are worth ranking or what a searcher wants — that judgment stays human.

Can I let an AI agent run my whole SEO?

No. Let it execute the mechanical work under a plan you set, and verify the output — check the rendered schema, confirm links resolve, and make sure new pages earn their place. The agent is the hands; the strategy and the quality bar are yours.

Want this done on your site?

Technical SEO at scale, internal linking that funnels to revenue, clean schema and sitemaps — done with AI agents and a human keeping the strategy honest. That's a lot of what I do for clients. Let's talk.

AI consulting with Patrick Bushe

Keep reading