GEO in practice: how to get AI models to cite your brand

More of your buyers now start with a question typed into ChatGPT, Gemini or Perplexity rather than a search box. Instead of a list of ten blue links, they get a written answer with a handful of named sources. If your brand is one of those sources, you are in the conversation. If it is not, you may never know the conversation happened.

Generative Engine Optimisation, or GEO, is the practice of making your organisation the kind of source these models reach for and cite. It overlaps with SEO, but the goal is different, and so are several of the moves. This is a hands-on guide to what actually works, and how to start without a large budget.

The landscape has widened in 2026. Google now answers a large share of queries not only with AI Overviews but with a conversational AI Mode that has become the default experience for many searches. Alongside it sit ChatGPT with live web search, Perplexity, Gemini and Microsoft Copilot. They differ in how they work, but they share one habit: they synthesise an answer and name a small set of sources. GEO is about being one of those named sources across the surfaces your buyers actually use.

Being cited is not the same as ranking

Classic SEO aims to place your page high in a list so a person clicks through. GEO aims to make the model use your content inside its answer and, ideally, name you as the source. There is often no list and no click. The reader gets a synthesised answer, and your brand either appears in it or does not.

That changes what good content looks like. A model does not skim a page and rank it against competitors on a results page. It reads for a clear, self-contained statement it can lift, attribute and trust. Your job is to be the passage that is easiest to quote and safest to attribute.

The winning position is no longer a rank on a page. It is being named as the answer, wherever the reader happens to ask.

The levers that actually move citations

A handful of factors do most of the work. None of them are tricks; they reward being genuinely clear and genuinely trusted.

• Answer the question early and plainly. Phrase a heading as the question your buyer would ask, then resolve it in the first sentence or two before you explain. A self-contained answer near the top of a section is far easier for a model to lift than one buried under introductions.
• Make content structured and quotable. Short, factual sentences, clear headings, definitions, and lists give a model clean units it can extract. Vague, meandering prose is hard to cite.
• Be clear and consistent about who you are. Use the same brand name, address, founding details and description everywhere. Publish a strong About or Organisation page and mark it up with schema so the entity is unambiguous.
• Earn references on sources AI already trusts. Reputable press, industry directories, review platforms and Wikipedia-grade references all feed the models’ sense of who is credible. A mention on a trusted third party often matters more than another page on your own site.
• Keep it fresh. Visible publication and update dates, and content that is genuinely current, help a model prefer you over a stale alternative.

Notice the pattern: clarity for the reader, credibility in the wider web, and structure a machine can parse. You are not writing for an algorithm so much as writing so plainly that both a person and a model can act on it.

Different engines trust different sources

One useful nuance in 2026: the answer engines do not all draw on the same places. Each leans on the sources it considers authoritative, so the same page can be cited by one and ignored by another. As a rough guide, ChatGPT tends to rely on broad reference sites and community discussion, Google’s AI Overviews and Gemini lean on Google’s own index and video, and Perplexity favours established news and reference publishers. The practical takeaway is not to chase each engine’s quirks, but to make sure you show up credibly in more than one type of source: your own clear pages, a reputable third-party mention, and a well-maintained reference entry together cover most of what these systems draw on.

A starting sequence for this quarter

You do not need to do everything at once. Work through this order and you will cover the highest-value moves first.

1. List the ten to twenty questions your buyers genuinely ask before they choose someone like you. Use sales calls and support tickets, not guesswork.
2. Map each question to a page. Where no page answers it well, plan one. Where a page half-answers it, rewrite the opening to answer it directly.
3. Tighten those pages: lead with the answer, add clear headings, break dense paragraphs into facts and lists, and add a short FAQ where it fits.
4. Fix your entity basics. Make name, description and contact details identical across your site, Google Business Profile, LinkedIn and any directories. Add Organisation and, where relevant, FAQ schema.
5. Pursue two or three credible external mentions: an industry directory, a guest article, a review platform, or a listing on a trusted local or sector site.
6. Set a monthly reminder to re-check and refresh. GEO is maintenance, not a one-off project.

How to tell whether it is working

You cannot manage what you do not observe, and AI answers are less transparent than search rankings. The most reliable check is also the simplest: ask the models the questions your buyers ask, and watch what happens.

• Run your buyer questions through ChatGPT, Perplexity, Gemini and Google’s AI Overviews and AI Mode. Record whether your brand appears, whether it is named as a source, and whether the summary is accurate. Because engines draw on different sources, note which ones cite you and which do not.
• Keep a simple log over time so you can see movement after you make changes. Answers vary between runs, so look at the trend, not a single result.
• Watch your analytics for referral traffic from AI tools and for a rise in branded searches, which often follows growing visibility in AI answers.
• Note when a model states something wrong about you. That is a signal your entity information is unclear or inconsistent somewhere, and worth fixing.

Common mistakes to avoid

Most failures are self-inflicted. These are the ones we see most often.

• Thin content that says little. If a page does not answer the question clearly, there is nothing worth quoting.
• Key facts locked inside images or graphics. If your value proposition, pricing logic or specifications live only in a picture, a model cannot read them. Put them in text as well.
• Inconsistent naming and details across the web. Every variation makes it harder for a model to be confident it is talking about you.
• Chasing every model and tactic at once. Pick the platforms your buyers actually use, get the fundamentals right, and expand from there.

Where to start

Pick your three most important buyer questions and make the pages that answer them genuinely excellent this month: answer first, structure clearly, and confirm your name and details are consistent everywhere. Then ask the models those same questions and write down what you see. That single loop, repeated, is the core of GEO in practice, and human judgement stays firmly in the loop deciding what to publish and how to describe your business. The rest is patience and consistency.

Frequently asked questions