§1The front page moved and most brands did not notice
For a technical brand in 2026, the highest-leverage place to be found is no longer a ranking on a search results page. It is inside an LLM's answer. When a builder asks ChatGPT or Perplexity "how should I structure model routing" or "what breaks agents in production", the model does not hand back ten blue links. It writes an answer and cites a handful of sources. Being one of those sources beats ranking third on a page almost nobody scrolls.
This has a name, generative engine optimisation, and it is not a rebrand of SEO. The unit of victory changed. The old game was rank for a query. The new game is get quoted in an answer. Those reward different things, and a lot of what we were told mattered for the first barely moves the second.
§2What we actually changed
We did this to our own site this week, so this is a report from the work, not a checklist copied off a blog. Four changes did the heavy lifting.
First, structured data on every page. A small block of JSON-LD that tells a machine plainly what each page is, who wrote it, when, and that it belongs to us. We added it to all 74 of our notes and news pages and wired it into the template so every future page carries it without a thought. This is the difference between a model guessing what a page is and being told.
Second, a clean Markdown twin of every article. We already published one. The point is that an LLM reading the Markdown spends its limited context on the substance, not our navigation, hero images and animations. If you make a model wade through chrome to find your argument, it often will not.
Third, an llms.txt that maps our best work for any model crawling the site, and now regenerates on every build instead of quietly freezing at whatever it said months ago. A map that goes stale is worse than no map.
Fourth, pillar pages. We gathered scattered notes into three clear themes so the site reads as "deep on this subject" rather than a reverse-chronological pile. Both humans and models navigate depth better when it is signposted.
§3The bug we found doing it
Here is the part the tidy version of this story would leave out. While auditing the pages, we found that 18 of our own news pages had their canonical tag pointing at the wrong path, a URL that returns a 404. For months we had been telling every crawler and every model "the real version of this page lives over here", and over here did not exist. We were quietly instructing machines to discount our own work.
That is the most useful thing GEO work surfaces. Before you chase being cited, check you are not actively telling the machines to ignore you. The audit is worth more than the optimisation, because the audit finds the own-goals.
§4What is theatre
Not all of the playbook earns its keep. Chasing schema markup for rich-result badges you will never qualify for, bolting FAQ blocks onto posts that do not need them, fussing over exact keyword phrasing for a system that reads meaning rather than strings. These feel like progress and mostly are not.
The things that matter are dull. Be machine-readable. Be unambiguous about who you are. Be genuinely the best answer to a real question someone asks. Structure helps a good answer get quoted. It cannot rescue a thin one, and a confident block of JSON-LD wrapped around a weak post just makes the weakness easier to parse.
The honest limit is worth stating plainly. Structured data does not make you authoritative. It makes you legible. Authority still comes from being right and being referenced by others. GEO is necessary now, not sufficient. We have made ourselves easy to read. Whether we are worth citing is a separate, harder question, and the only one that ultimately counts.
§5The takeaway
If you run a technical brand, the question is no longer only "do we rank". It is "when a model answers a question we should own, are we in the citations". Most brands are not, and most have never checked. The fix is half plumbing and half the hard part. The plumbing, being legible and unambiguous, we did in a week and you can too. The other half, being the best answer in the room, is the work it always was. Do the plumbing so the second half is not wasted, then go and earn the citation.