Generative engine optimisation: how to get named in the AI answer.

The term comes from a 2023 research paper by Aggarwal and colleagues, which tested how changes to a page altered how often generative engines quoted it. The right changes raised a page’s visibility in the generated answers by up to 40%, and the strongest were adding citations, quotations from relevant sources and statistics. That matches what we see. Assistants quote what is specific, sourced and easy to lift.

You will also see it called answer engine optimisation (AEO), AI SEO or LLM SEO. The labels differ. The job is the same.

GEO and SEO: what actually changes

GEO isn’t a replacement for SEO. Most assistants still lean on a search index to find what to read, so a page that can’t be crawled or ranked rarely gets quoted. What changes is the unit of success.

	SEO	GEO
What you win	A position in a list of links	A mention, a description or a citation inside the answer
Who reads you first	A person scanning results	A model deciding what to repeat
What it rewards	Relevance, links, page experience	Clear facts, consistent naming, third-party agreement
How you measure it	Rankings, clicks	How often you are named, how accurately, and next to whom

So the old work still matters. It just isn’t enough on its own any more.

How assistants decide who to name

When someone asks an assistant “which agency should I use for X”, the answer is built from 2 places: what the model learned in training, and what it retrieves live from the web. You can influence both, but on different timescales. Retrieval responds in weeks. Training data moves when the next model ships.

The SetpointHQ Index scores 100 UK employer brands every month on how ChatGPT, Claude, Gemini and Perplexity describe them, across 5 dimensions. They make a decent map of what moves an AI answer, whatever your sector:

• Citation likelihood. Whether the assistants name you at all when asked, and how near the top of the answer.
• Schema completeness. Whether your site states, in structured data, who you are and what you do.
• Knowledge-graph presence. Whether you exist as a clear entity in places like Wikipedia and Wikidata, so the model can tell you apart from a business with a similar name.
• LLM crawlability. Whether AI crawlers are allowed in, and can read your pages without running JavaScript.
• Social-signal density. Whether you show up on the review and profile sites the models absorb.

The one most businesses miss sits outside their own site. A claim that several independent sources agree on is far easier for an assistant to repeat than one that only appears on yours.

What to do, in order

1. Let the crawlers in. Check robots.txt for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. Blocking them is a legitimate choice, but it should be a choice, not an accident left by a plugin.
2. Make the pages readable without JavaScript. If your content only appears after scripts run, many AI crawlers see an empty page. Server-rendered HTML is the safe default.
3. Say who you are, in a way a machine can’t misread. Organization schema with your legal name, address and sameAs links to your LinkedIn page, Companies House record and Wikidata item. Use the same name everywhere.
4. Write pages that answer the question people type. Lead with a 40 to 60 word answer, then the detail. Use real numbers and say where they came from. A vague page gives the model nothing worth quoting.
5. Get other sites saying the same thing. Directories that assistants read, trade press, partner pages, review sites, a Wikipedia mention if you have earned one. This is slow, unglamorous work, and it moves answers more than anything you do on your own site.
6. Add an llms.txt file. It is a plain-text map of your site for language models. There is little public evidence that the big assistants rely on it yet. It takes an hour, so we add one anyway.
7. Measure it every month. Pick the 20 to 50 questions your buyers actually ask, run them through ChatGPT, Claude, Gemini and Perplexity, and record who gets named. Then change something and run them again.

How to measure GEO

Rankings don’t transfer. The measures that do:

• Share of answers: out of your question set, how often each assistant names you.
• Accuracy: whether what they say about you is true. A wrong description is worse than none, because buyers act on it.
• Company you keep: which competitors are named alongside you, and in what order.
• Citations: which pages, yours or other people’s, the assistants link to when they mention you.
• Referral traffic: visits from chatgpt.com, perplexity.ai and the rest in your analytics. Small numbers, usually high intent.

What GEO won’t do

It won’t guarantee a mention. Nobody controls what a model says, and anyone promising a fixed spot in ChatGPT’s answer is guessing. It won’t rescue a business that buyers don’t talk about anywhere else, either. And it won’t work fast on training data, which only changes when a new model is released.

What it will do is stop you being invisible by default, and make sure that when you are named, the description is yours.

If you want to pay for a placement instead of earning one, that is a different channel. See ChatGPT ads.

Common questions

Sources: Aggarwal and others, “GEO: Generative Engine Optimization”, arXiv 2311.09735 (2023). Google Search Central guidance on generative AI features. Dimensions from The SetpointHQ Index methodology.