Search stopped being a list of links and became an answer with footnotes. The work of getting into those footnotes is not the work of ranking.
Marku Leone · Founder, donebyverde · South Florida
For twenty years the goal was a position. You wanted to be the first blue link, because the first blue link got the click and the click got the customer. Answer engines broke that chain. Someone asks ChatGPT which fence company to use in their town, or asks Perplexity how much a website should cost, and the machine writes a paragraph. Sometimes it names three companies. Sometimes it names one. There is no page two, and for most questions there is no click at all.
That is the whole change, and it is bigger than it sounds. Ranking was a competition for attention. Citation is a competition for trust. The model is not choosing who paid the most or who has the most backlinks. It is assembling an answer from sources it has reason to believe, and deciding which of those sources is worth naming.
Generative engine optimisation, GEO, is the practice of being one of those sources. It overlaps with SEO, because a page that cannot be crawled cannot be cited. But the two diverge quickly, and the places they diverge are where most businesses are still doing the wrong work.
Simplify the pipeline and three things are happening. The model retrieves candidate material, either from what it learned in training or from a live search it runs in the moment. It weighs that material against everything else it retrieved, looking for agreement. Then it writes an answer and decides which sources to attribute.
Each stage rewards something different. Retrieval rewards being findable and parseable. Weighing rewards consistency, because when four sources say one thing and yours says another, yours is the outlier and outliers get dropped. Attribution rewards specificity, because a model names a source when the source is the reason the claim can be made at all.
That last point is the one worth sitting with. Generic content is absorbed and never credited. If your page says websites cost between three and ten thousand dollars, the model already knew that from a hundred other pages and will not name you. If your page says a one time build ships in three weeks and a subscription includes hosting, SEO, and unlimited small changes on a twelve month minimum, that is a specific, attributable claim about a real business. The model has nowhere else to get it. That is what a citation looks like from the inside.
The most reliable way to get cited is to publish something that only you can publish. For a service business that is easier than it sounds, because you are sitting on primary data nobody else has.
You know what a project in your category actually costs, because you quote them every week. You know how long the work really takes, including the parts clients forget. You know which questions come up on every single discovery call. You know what happened to a client's numbers after you did the work. None of that exists anywhere on the internet until you write it down, and all of it is the kind of concrete claim an answer engine has to attribute.
Compare that to another post explaining what SEO stands for. There is no version of that page that earns a citation, because there is no sentence in it that the model could not have written on its own.
Answer the question in the first hundred words. Models weight the top of a document heavily, and a page that opens with three paragraphs of throat clearing before it gets to the point is a page whose point may never be extracted. Put the answer first, then earn it back with the reasoning.
Use headings as questions and sub-answers. A page structured as a genuine outline, with real heading elements rather than styled divs, is a page a retrieval system can slice into passages. This is not a metaphor. Most retrieval works on chunks, and clean headings determine where the chunk boundaries fall.
Publish your facts as structured data. Organization markup with a consistent name, URL, logo and contact. Service markup for what you sell. FAQ markup for the questions you already answer on a call. Structured data does not make a model like you, but it removes ambiguity about who you are and what you do, and ambiguity is the thing that gets a source dropped in the weighing stage.
Then keep the facts identical everywhere. Your business name, your service area, your positioning, your prices. When your website, your Google profile, your LinkedIn and a directory listing disagree about basic facts, every one of those sources gets less trustworthy. Consistency is not a nice to have here. It is the mechanism.
Citation is not only about your own site. Models lean heavily on sources they already consider reliable, which for local service businesses means industry publications, well moderated communities, local press, supplier and partner pages, and the handful of directories that are genuinely curated rather than spam.
One substantive mention on a page the model already trusts moves more than ten thin pages on your own domain. This is the part that looks like public relations rather than marketing, and it is the part most agencies skip because it cannot be automated.
Keyword density does not work, because the model is not counting keywords. Publishing volume for its own sake does not work, and actively backfires: a domain full of generic AI written filler is a domain with a low signal to noise ratio, and every thin page dilutes the pages that are real.
Trying to be everywhere does not work either. A service business that tries to be the authority on all of marketing will be the authority on none of it. The businesses getting cited are narrow. They own one question completely rather than owning a hundred questions partially.
You cannot check a rank position, so check the answer. Write down the fifteen or twenty questions a real prospect would ask before hiring you, phrased the way a person actually types. Ask them to ChatGPT, Perplexity and Gemini once a month, in a fresh session each time, and log whether you are mentioned, how you are described, and who is mentioned instead.
That log is the closest thing to a rank tracker this era has. It tells you which questions you already own, which competitor keeps appearing where you should be, and whether the description the machine gives of your business is the one you would give yourself. When the description is wrong, that is not a ranking problem. It is a consistency problem, and it is fixable.
GEO rewards businesses that have something specific and true to say, and punishes businesses that publish to fill a calendar. That is inconvenient if your content strategy was volume, and it is very good news if you actually do the work and have never bothered writing it down.
Founder of donebyverde, a studio in South Florida. We build the site, shoot the content, and wire the follow up that answers first, for service businesses across South Florida. This article is how we actually work.
Talk it throughReal partnerships start with one strategy call.
Get It Done
