SEO, GEO, AEO: A Plain Guide to the New Age of Search

Marketing has a habit of inventing a new acronym every time the ground moves, and the ground has genuinely moved. SEO you know: search engine optimisation, the work of earning positions in classic search results. AEO stands for answer engine optimisation, aimed at the direct answers search engines assemble at the top of the page. GEO stands for generative engine optimisation, aimed at the answers AI assistants write from scratch when someone asks ChatGPT, Gemini, or Google's AI Overviews for a recommendation.

Strip the labels away and all three answer the same question: when a customer looks for what you sell, does your business appear, and is it described accurately and favourably? The difference is who does the reading. In SEO a person scans a results page. In AEO a ranking system extracts a snippet. In GEO a language model reads a handful of retrieved pages and composes an answer, usually with citations.

We run all three for clients, and the most useful thing we can tell you is that they are not three separate projects. They are three doors into the same house, and the house is a website that answers real questions clearly enough for any reader, human or machine, to quote it.

Classic SEO rewards relevance and authority built over time: pages matched to what people search, a site that crawls cleanly, and links and mentions that suggest you are a real, referenced business. None of that has been cancelled. AI assistants lean on conventional search indexes to shortlist candidate pages before they write anything, so if your SEO is broken, your GEO is broken too, just less visibly.

AEO rewards extractability. Answer boxes and featured snippets are assembled by systems looking for a question stated plainly and an answer sitting right next to it. Pages structured as clear questions with concise, complete answers underneath win these placements. FAQ schema and sensible headings do a lot of quiet work here.

GEO rewards being quotable and being trusted. A language model on a deadline lifts plain factual sentences where the claim and the subject sit together. It cross-checks who you are against structured data and consistent business details across the web. And it reaches for first-hand specifics, because your own numbers, your own cases, and your own sector knowledge are the one thing it cannot synthesise from anywhere else. We wrote about the mechanics of this in our earlier article on how AI search picks businesses, and the pattern has only firmed up since.

The practical reason to care is not the technology, it is the queue. A growing share of buying research now ends inside an AI answer without a single click on a traditional result. When a patient asks an assistant which London clinic to trust, or a founder asks how to enter the UK market, the businesses named in that answer have effectively skipped the queue. The rest are invisible, however well they rank underneath.

Our own client work shows the shift in the lead source data. "I asked ChatGPT" started appearing in our clients' enquiry forms in 2025, first as a novelty, now as a steady line item. For businesses in considered-purchase sectors, health, legal, property, B2B services, the assistant is often the first stop, and classic search the second, used to verify what the assistant said.

That order matters. If the AI answer names you and the search results then confirm you look credible, the two channels compound. If the AI answer names your competitor and search confirms them instead, you lost the enquiry before you knew it existed.

Budgets are finite, so here is the sequence we use with clients, in plain terms.

Notice that the first three steps are shared. This is why we are sceptical of pitches that sell GEO or AEO as a standalone product with its own retainer. Most of the work is the same work. The honest question to ask any supplier is which parts of their proposal would still exist if the acronym disappeared tomorrow.

Some of what travels under the GEO banner is noise. Hidden prompts addressed to language models, keyword-stuffed pages in white text, and mass-produced AI articles all attack systems that are retrained continuously, and the trick that works in March is a liability by June. We do not sell any of it.

We would also keep perspective on the protocol layer: llms.txt files and agent-readable endpoints are cheap to add and we include them on every build, but they are insurance for where the web is heading, not the lever that moves this quarter's enquiries. The lever is the boring one: pages that answer the questions your customers actually ask, on a site machines can read, from a business whose details check out everywhere. That was true in the age of ten blue links, and the new age of search has made it more true, not less.