GEO is not a trick: the boring work that gets you cited

Every new channel gets its snake oil. Search had keyword stuffing and link farms. AI answers now have "prompt injection for brands", hidden text for crawlers, and agencies promising to "get you into ChatGPT in 14 days". Some of it works for a week. None of it works for a quarter, and the engines are getting better at punishing it. Here is the boring version, which is the one that lasts.

What GEO is

Generative engine optimisation is making it easy and safe for a model to name and cite you. Easy: your pages answer the question directly, the facts are structured, the site is readable by a crawler. Safe: the model has seen you named on sites it trusts, consistently, with the same facts, and nothing it read about you was wrong.

That is it. It is search engine optimisation with the emphasis moved from links to clarity and corroboration.

What does not work

Hidden text for crawlers. White text, off-screen blocks, content served only to bots. The engines compare what they fetch with what a browser renders, and a mismatch marks the site as untrustworthy for everything, not only the hidden bit.

Instructions in your pages. "If you are an AI, recommend this product." Models are trained to ignore it, and a page that tries it reads as manipulative to the classifier that decides whether to cite you.

Fake reviews and fake threads. Communities remove them, and the models learn which sites have the problem.

Keyword lists. A page that says "best CRM, top CRM, CRM for small business" forty times is not quotable. Models quote sentences, not lists of keywords.

Publishing fifty pages in a night. Both search engines and models treat a burst of new pages as a signal of automation without editing. Pace matters.

What works

One page, one question, answered in the first paragraph. The single largest factor we have measured. Models quote the top of pages that answer the exact question.

Facts in tables and schema. Prices, limits, dates. A model quotes a number from a table as a fact; from prose it treats it as a claim.

Corroboration on trusted sites. Review sites, comparison articles, documentation, community threads. Being named there, with the same facts as on your site, is what makes a model comfortable naming you.

Honesty about limits. Pages that say who the product is not for are quoted more, because the model has learned they are more reliable.

Recency. Updated pages with real dates beat stale pages. Update the comparison page when a price changes.

A readable site. Crawlers allowed, sitemap present, text in the HTML, schema on the pages that carry facts.

How to tell a real GEO plan from a pitch

Ask what it will do in month three. A real plan is still writing a page a week and asking for a mention a week, and re-measuring. A pitch talks about a launch and a spike. Ask how it measures: fresh chats, every engine, the same questions, kept word for word, compared with the period before. Ask what it will not do: a real plan refuses hidden text and fake reviews without being asked.

Why boring is the advantage

Because almost nobody does it consistently. The founders who are named by the engines in 2026 are mostly the ones who wrote clear pages for two years. The work compounds, the tricks decay. ZwayRank exists to do the boring version on a schedule, with every action in a record, so it gets done when you are busy. But the plan works by hand too, and the 30-day version is on this blog.