What is GEO? The basics of being visible in AI search engines

There is a complaint we hear from business owners more and more often: they asked ChatGPT about their own industry, the answer listed their competitors one by one, and their company was nowhere to be found. It isn’t surprising. Some people no longer open ten blue links one after another; they type their question into ChatGPT, Gemini, Perplexity or Copilot, or read Google’s AI summary and move on. What they get is one consolidated answer, usually compiled from pages on the web, sometimes with links back to them. GEO (generative engine optimization) is the name for the work that gets your brand into that answer, described accurately and shown as a source.

Is GEO replacing SEO?

No. It sits on top of it. Many AI-assisted search experiences draw on search engine indexes and on content gathered by crawlers when they generate an answer. A page that can’t be crawled, loads slowly or is vague about what it covers has a hard time getting in. A site with weak SEO rarely does well at GEO.

The real difference is what you are aiming for:

• With SEO, the goal is to rank high on the results page and win the click.
• With GEO, the goal is for your brand, product or point of view to appear correctly inside the answer a language model writes, and to be cited as a source.

So “where do we rank for this keyword?” is still a fair question. GEO simply puts another one next to it: “what does AI say about us, and what is it basing that on?”

How do AI answers pick their sources?

Each system does it differently, and most of the details aren’t public. The broad logic is similar, though: the system interprets the question, finds relevant documents (often with a search backend doing that part), picks the passages that fit and writes the answer. In the literature this is called retrieval-augmented generation (RAG).

For your page to be the one that gets picked, it has to do three things well:

1. Be findable: crawlable, indexable and fast.
2. Be quotable: each section answers one question clearly, so a model can lift a meaningful passage out of a long page.
3. Be trustworthy: who wrote it, when was it last updated, what backs up the claims? Readers look for these signals, and so do the systems.

The checklist we work through

1. Technical foundation: crawlability and speed

The first file we open is robots.txt. Is anything important blocked? Are pages delivered from the server as HTML? Do they load quickly on a phone? Pages whose content is built only in the browser with JavaScript may be read incompletely by some crawlers, and we regularly come across sites that look polished to a visitor but are close to empty from a crawler’s point of view. If you are starting from scratch, our SEO-ready website templates come with these basics already in place.

Some AI companies run their own crawlers; OpenAI’s GPTBot, Anthropic’s ClaudeBot and PerplexityBot are a few examples. Whether to let them in is a business decision. If you want to be visible, at the very least make sure you aren’t blocking them by accident.

2. Write in questions and answers

Make a list of the questions your customers put to AI assistants. Your sales team, or whoever answers the phone, is usually the quickest source, because the same questions come up again and again. Give each one a clear heading on your site and make the first sentence underneath it the direct answer. Detail, examples and limitations can follow. This layout makes your content easier to quote in featured snippets and in AI summaries.

3. Structured data

Schema.org markup such as Organization, Product, Article/BlogPosting, FAQPage and BreadcrumbList tells machines plainly what the entities on the page are: the company, the product, the author. On its own it won’t get you into an AI answer. It helps your content be read correctly, and that’s about it, which is why we treat the page itself as the priority and the markup as support.

4. Same name, same details, everywhere

Your company name, product names, address and contact details should be written the same way on your website, your Google Business Profile, LinkedIn and industry directories. Language models recognise a brand by stitching together what different sources say about it. If an old address is still sitting in a directory after you moved, or your company name circulates in three different spellings, don’t be surprised by the occasional wrong answer.

5. A short introduction with llms.txt

llms.txt is a proposed format for gathering your most important pages into a plain Markdown file that language models can read easily. It isn’t an official standard yet, and not every system uses it. Still, it doesn’t take long to write, and it makes the team agree on how you describe yourselves, which is a useful exercise in its own right.

6. Information nobody else has

If a page repeats the same generic information found everywhere else, a system has no reason to choose it as a source. Write about your own experience, the way you work, your checklists and your case studies. If you quote figures, name the source, and keep anything time-sensitive up to date.

Mistakes we see most often

• Looking at keywords only: questions typed into AI assistants tend to be long and full of context. Think about the sentences people actually write and the intent behind them, not single-word targets.
• Burying key information in images or PDFs: scope of service, service area and delivery times belong on the page as plain text. A service list uploaded as a brochure image may not be readable to a crawler at all.
• Letting things go stale: old product names, discontinued services and changed contact details can keep circulating in answers for a long time. Keep publication and update dates visible on the page.
• Blocking crawlers by accident: security plugins and CDN rules sometimes turn AI crawlers away too. Check these settings every so often.

How do you know it’s working?

Measuring GEO is far less settled than measuring SEO, so don’t expect a neat dashboard. What we do at the start is fairly simple:

• Ask different AI assistants your target questions at regular intervals, and note in a spreadsheet whether your brand appears and which sources are shown.
• In your analytics tool, track traffic referred by sources such as chatgpt.com, perplexity.ai and gemini.google.com in a separate segment.
• In Search Console, follow changes in impressions and clicks for informational queries.

The first version of that spreadsheet is often discouraging: the assistants either don’t mention you at all or describe a service you dropped long ago. It is still the right place to start, because it tells you which page to fix first. This week, write down the five questions your customers ask most and put each one to two different assistants. Where to begin will usually be obvious within the first half hour.