GEO (generative engine optimisation)

A generative engine is a search product that retrieves several sources and writes one answer from them, with citations. ChatGPT search, Perplexity and Google’s AI Overviews all work this way. GEO is the work of making your content more likely to be used in that answer, and more visible within it.

The name comes from a paper by Pranjal Aggarwal and colleagues, first posted in November 2023 and published at KDD 2024. They built a benchmark of queries, rewrote source pages in different ways and measured how much of each page showed up in the generated answers. Adding citations to sources, quotations and statistics improved visibility by roughly 30 to 40% on their main metric, while keyword stuffing did not help. Results also varied by topic, so there is no single formula. GEO and AEO describe nearly the same practice: GEO is the research term, AEO the marketing one.

Why it matters for a SaaS website

When a buyer asks an assistant to compare tools, the answer is written from a few pages. The page that supplies the clearest facts, figures and named sources tends to shape the wording, and a page full of claims with nothing to back them gets skipped. For SaaS, that decides whether your pricing, integrations and positioning reach the buyer in your words or a competitor’s.

How DesignMe handles it

We apply the findings as part of normal page writing, not as a separate project. Pages open with a direct answer, back claims with named sources and dated numbers, and include specifics such as prices, timelines and client names that an engine can lift as they stand. Main content stays in plain HTML so crawlers that do not run JavaScript can read it. We then check the result by running real buyer questions through ChatGPT, Claude, Perplexity and Google each month, as described in our AEO for SaaS post.

Example

Say an analytics startup has a comparison page that says it is “faster and easier” than a rival. Rewritten for GEO, it states the query speed it measured, links the benchmark method, quotes a named customer and gives the date of the test. An assistant asked to compare the two now has a sentence it can cite, with the source attached.

Common mistakes

• Treating GEO as a trick. The paper’s gains came from adding evidence, not from gaming the model.
• Unsourced statistics. Numbers without a source or date are weaker material and a risk if they are wrong.
• Ignoring SEO. Engines still retrieve from a search index first. A page nobody can find is never quoted.
• One-off optimisation. Models and answers change. Retest on a schedule.