Ranking first matters less than it used to.

GEO is optimising to be cited by systems that answer questions directly — AI Overviews, ChatGPT, Perplexity, Copilot — rather than to rank in a list of links.

It overlaps heavily with good SEO and then diverges in a few ways that matter. AI crawlers do not run JavaScript, so anything that appears only after the page loads is invisible to them. They quote passages, not pages, which rewards writing that answers a question in its first sentence instead of building to a conclusion. And they can only read you if you let them: a robots file that blocks or forgets to mention them is a decision most sites make by accident.

The good news is that none of this conflicts with ranking well. It is mostly the same work, done with a second reader in mind.

We are developers, so the fixes land in the codebase rather than in a report.

Most technical SEO findings are engineering problems — rendering, structured data, canonical handling, how routes are generated, what the crawler receives versus what a browser shows. An audit that identifies them still leaves you needing someone to make the changes, and that handoff is where most recommendations quietly die.

We also will not sell you a monthly retainer for content nobody asked for. What earns links for a technical business is teardowns, real postmortems, and tools that are genuinely useful. It is slower than publishing ten listicles and it is the only version that keeps working.

It starts with what the crawler sees, which is often not what you see.

Then: structured data that states facts about the business rather than decorating the page, an information architecture where each page answers one question properly instead of every page answering all of them badly, and passages written so a machine can lift one and have it still make sense. Core Web Vitals get fixed as code, not as advice.

For local and service businesses this compounds fast, because the questions are specific and the competition is usually a page of stock photography. Our own automotive client is the worked example — twenty-three pages, each answering one question, all of it readable by the systems now doing the answering.