Technical SEO is the part small business owners hand to an agency because it looks like programming. A lot of it is. That used to mean paying a retainer for work that amounted to a few hours of careful edits — structured data, image handling, a sitemap that reflects reality. Coding agents changed the economics of that specific job more than almost anything else in marketing. This guide covers what you can safely do yourself with Codex, and just as importantly, where you should stop.

Why a Coding Agent Rather Than a Chat Window

Chat assistants are good at telling you what schema markup should look like. You still have to paste it into the right template, in the right place, without breaking the page. A coding agent works directly in your project: it reads the actual files, makes the change across every template that needs it, and shows you a diff before anything is committed. For repetitive edits spread across a codebase — which is most technical SEO — that difference is the whole ballgame.

Two ground rules before you start, and I say this as someone who has broken a live site by ignoring them. Work in version control. If your site is not in Git, that is the first task, because the ability to undo is what makes any of this safe. Never point an agent at production. Work locally or on a staging copy, review every diff, and deploy deliberately.

Start With an Audit, Not With Changes

Resist asking for fixes immediately. Ask for a survey first:

Audit this site for technical SEO issues. Report only — make no changes. Check: pages missing title tags or meta descriptions, duplicate or missing H1s, images without alt text or explicit width and height, missing or invalid structured data, internal links that 404, and anything blocking indexing in robots.txt or the sitemap. Give me a list ordered by likely impact.

You now have a prioritized backlog grounded in your actual code rather than a generic checklist. Work through it one item at a time, reviewing each change before moving on. Batching six unrelated fixes into one change is how you end up unable to tell which one broke something.

The Fix That Is Actually Worth Your Time: Structured Data

Of everything on a typical list, structured data is where a small business gets the most return — because it is the layer that tells search engines what your pages are rather than leaving them to infer it, and because most competitors have not bothered.

The types that matter for most small businesses are few. LocalBusiness (or the specific subtype that fits you) on your homepage and contact page, with name, address, phone, hours, and geo coordinates matching your Google Business Profile exactly. Article or BlogPosting on posts, with author, publish date, and modified date. BreadcrumbList on anything nested. Product or Service if you sell something specific. That is most of it.

A useful prompt:

Add JSON-LD structured data to the blog post template. Use BlogPosting with headline, description, datePublished, dateModified, author, image, and mainEntityOfPage. Pull every value from the existing post data — do not hardcode anything and do not invent fields the data does not contain. Show me the diff before changing files.

That last clause is not optional. The common failure mode is an agent generating plausible markup containing values that do not correspond to anything real on the page, which is worse than no markup at all. Validate the result with Google’s Rich Results Test and Schema.org’s validator afterward.

One caveat that saves wasted effort: not all schema produces a visible search feature. Google removed FAQ rich results for most sites, so adding FAQ markup expecting expandable questions in the SERP is chasing something that no longer exists. Structured data is still worth adding — it helps machines understand the page, which increasingly matters for how AI systems summarize and cite you — but add it for comprehension, not for a rich snippet that was retired.

Entity Consistency: The Underrated One

If you take one thing from this post beyond schema basics, make it this. Search engines are trying to resolve your business to an entity — a specific thing in the world — not just match strings. Anything ambiguous makes that harder.

Practically: your business name, address, and phone should be byte-identical across your site, your structured data, your Google Business Profile, and your directory listings. If you use an abbreviation or an alternate name, declare it explicitly with alternateName in your schema rather than hoping it is inferred. Use a stable @id so the entity on your homepage and the one on your contact page are understood as the same thing rather than two similar ones. If your name collides with another business, say plainly on the page what you are and are not.

This is unglamorous and it compounds. I have covered the underlying mechanics in more depth in Google’s LLM patent and what it implies for entity SEO.

Performance, Handled Proportionately

Core Web Vitals — loading (LCP), responsiveness (INP), and visual stability (CLS) — are a genuine but modest ranking consideration. They matter far more as a conversion issue: slow sites lose customers regardless of where they rank.

The wins are usually boring and mechanical, which makes them ideal agent work: convert oversized images to modern formats and serve them at sensible dimensions, add explicit width and height so layout does not jump while loading, lazy-load anything below the fold, defer non-critical scripts. Ask for these one at a time, measure with PageSpeed Insights before and after, and stop when you hit diminishing returns. Chasing a perfect score is a poor use of a business owner’s afternoon.

Make Bing Part of the Routine

Bing is a smaller share of search but a meaningfully larger share of AI answer surfaces, and it is far less competitive. It also offers something Google does not: IndexNow, a protocol that lets you notify participating engines the instant a page changes rather than waiting to be crawled. It is a small amount of code and an agent can wire it into your publishing flow in one sitting. Google does not use IndexNow, so this is a Bing-side win, not a universal one — but it is close to free. I have written up why IndexNow is a gateway to AI search answers.

Where to Stop

Some things are not worth the risk of a confident agent and a live site. Be careful with redirects — a bad rule can loop or quietly orphan pages, and mistakes here are costly. Be careful with robots.txt and canonical tags, where a single wrong line can deindex your site. Do not let an agent restructure your URLs without a considered migration plan. And skip anything sold as a clever trick to influence rankings: if it works by manipulating a search engine rather than by making the site genuinely better, it is a liability with a delay on it.

The honest summary is that technical SEO for a small business is perhaps ten hours of real work, most of it one-time, now within reach of a non-developer with a coding agent and some discipline. Once it is done, the recurring work is content and distribution — covered in writing posts that rank and in the wider JSB Media Plan for WordPress SEO pillar.