GEO, explained: getting mentioned when people ask an AI

Generative Engine Optimization (GEO) is the practice of shaping your website and your presence around the web so that AI assistants find you, understand what you do, trust you, and mention or cite you when someone asks them a question. When a customer types "who does good window cleaning in Tallinn" into ChatGPT instead of Google, they don't get ten blue links. They get a short answer with two or three names. GEO is about being one of those names.

It is not a replacement for SEO. Most AI answers are built on a search index, so SEO basics still decide whether you're in the running. GEO adds a layer: pages that answer questions directly, facts that survive being lifted into an answer, an identity that looks the same everywhere, and other people talking about you.

How do AI assistants find and choose their sources?

AI assistants answering questions about current things mostly work in two steps: they search, then they write. The model turns the question into searches, reads a handful of the pages that come back, and writes an answer that quotes, paraphrases or links some of them. Nobody outside these companies knows the exact rules, and I won't pretend to. The broad shape is public:

• ChatGPT search uses OpenAI's own search crawler, OAI-SearchBot, alongside other search sources, and shows cited links in its answers.
• Perplexity runs its own index, crawled by PerplexityBot, and puts numbered citations next to every claim.
• Google AI Overviews and AI Mode draw on Google's normal search index. Google says it may use "query fan-out": running several related searches on subtopics to build one answer. Its documentation for site owners is explicit that a page needs to be indexed and eligible to show a snippet, and that no special files or markup are required.
• Gemini and Copilot do the same with Google's and Microsoft's own search infrastructure respectively.
• Claude can search the web when that feature is on, and Anthropic has its own crawlers for search and for user-requested fetches.

Two things follow. If a crawler can't reach your page, no clever writing gets you into the answer. And the model picks what it can use: a clear answering sentence beats a paragraph of marketing warm-up.

There is a second route into answers: training data. What the web said about you when a model was trained shapes what it "knows" without searching. You can't edit that directly; you can only keep the public record about your business accurate and consistent.

GEO vs SEO: what's the same and what's different

GEO and SEO share a foundation and differ in what counts as winning. SEO wins a position on a results page; GEO wins a sentence inside an answer.

If your SEO is weak, start there. A page Google can't index won't appear in an AI Overview.

What the research says

The term GEO comes from a research paper: GEO: Generative Engine Optimization by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, from IIT Delhi, Princeton University and independent researchers. It was first posted in November 2023 and published at KDD 2024 (arXiv:2311.09735).

The authors built a benchmark of search queries across many topics, rewrote source pages in different ways, and measured how much of each source ended up in the generated answers. Their main findings, in plain terms:

• Adding citations, quotations from relevant sources and statistics made content noticeably more visible in generative answers. These were among the best-performing methods; the paper reports gains of up to about 40% on its visibility measures.
• Keyword stuffing, the old SEO trick, did little or nothing, and in some tests made things worse.
• Lower-ranked pages gained the most. Sources that sat further down the traditional search results benefited more from these changes than the top result did.
• What works varies by topic. No single method won everywhere.

Keep this in proportion. It is one study, on a benchmark the authors built, with AI systems as they worked in 2023–2024. Assistants have changed since. But the direction matches common sense: an answer engine wants material it can quote and back up.

The GEO checklist: nine things to do

Here is the list in the order I'd do it for a small business. The first three are technical and take an afternoon; the rest is ongoing.

1. Let the AI crawlers in

Check your robots.txt file (at yoursite.com/robots.txt). Many sites block AI crawlers without anyone deciding to: a hosting default, a security plugin, a copied template. The ones that matter:

Search and training are separate switches. OpenAI's crawler documentation says that blocking OAI-SearchBot keeps you out of ChatGPT search answers, while GPTBot is about training. You can allow search bots and block training bots if that's your preference. For visibility, the search bots are the ones you can't afford to block.

Google-Extended needs a special note because it is often misunderstood. According to Google's crawler documentation, it controls whether content Google crawls may be used to train future Gemini models and for grounding in Gemini Apps and Vertex AI. It does not affect whether your site appears in Google Search, and it is not a ranking signal. So blocking Google-Extended does not remove you from AI Overviews; those run on the normal search index that Googlebot feeds.

2. Consider an llms.txt file, without expecting much

llms.txt is a proposed standard, published by Jeremy Howard in 2024 at llmstxt.org: a Markdown file at your site's root that gives language models a short summary of who you are and links to your most important pages. It is cheap to make and does no harm. Be honest with yourself about what it does, though: none of the big AI companies has confirmed that its assistants use it, and Google has said plainly that you don't need AI text files to appear in its AI features. Treat it as a tidy business card, not a lever.

3. Add structured data

Structured data is a block of code (JSON-LD, using the Schema.org vocabulary) that states facts about your page in a form machines read without guessing. The useful types for a small business:

• Organization or LocalBusiness: name, address, phone, opening hours, area served, and sameAs links to your official profiles.
• Person: for the people behind the business and the authors of your articles.
• Article: headline, author, date published and updated.
• FAQPage: for real question-and-answer sections.

Google says no special schema is required for AI Overviews, and structured data is not a ticket in. What it does is remove ambiguity about who you are, and ambiguity is what makes an assistant leave you out or mix you up with someone else.

4. Make your identity consistent everywhere

An assistant pieces together who you are from your site, Google Business Profile, business registers, directories, social profiles and press. If your name is spelled three ways, or half the listings still have your old address, you become a blurry entity. Blurry entities don't get recommended.

Pick one name, one short description, one address format and one phone number. Use them on every profile you control, and link those profiles to each other and from your site.

5. Write answer-first

Open every page and section with a sentence that directly answers its question. "A deep clean of a two-room flat in Tallinn takes four to five hours" is quotable. "We believe every home deserves to shine" is not. Question in the heading, answer in the first line, detail after. Each section should make sense lifted out on its own, because that's what an assistant does.

6. Back claims with numbers and name the source

This is the research finding put to work. Where a number helps, use it: prices, durations, how many jobs you've done. When the number isn't yours, name the source and link to it. Never invent one.

7. Get mentioned on sites that aren't yours

Assistants lean on what others say about you, not just what you say about yourself. The legitimate places to earn that:

• Local and trade press: a useful comment or a real story, not a paid puff piece.
• Directories and registers that people in your industry actually use.
• Reviews on Google and on the platforms your customers check.
• Forums and communities, including Reddit, where people ask for recommendations. Take part as yourself and be useful; planted posts get spotted and removed.
• Wikipedia and Wikidata, only if your business genuinely meets their notability rules. Writing your own Wikipedia article usually backfires.

8. Keep pages fresh and dated

Show when a page was published and last updated, and actually update the ones that matter. An assistant answering "what does X cost in 2026" has reason to prefer a page that visibly covers 2026.

9. Make it readable without JavaScript

Some crawlers don't run JavaScript. If your prices or contact details only appear after scripts load, a crawler may see an empty page. Use server-rendered or static HTML for anything you want quoted.

You can check the technical basics of your own site in a minute with our free GEO readiness checker. It looks at crawler access, structured data, llms.txt and a few other signals.

How do you measure GEO?

You measure GEO by asking the assistants your customers' questions on a schedule and writing down what comes back. There is no Search Console for ChatGPT, so the honest method is manual:

1. Write 10–20 real questions. Take them from your sales calls, emails and reviews, in your customers' words. Example: "who can fix a leaking roof in Tartu quickly", not "roofing services Estonia".
2. Ask them every month in ChatGPT, Perplexity, Google (look at the AI Overview and AI Mode), Gemini, Copilot and Claude. Use a logged-out or fresh session where you can, so your own history doesn't colour the result.
3. Log the result in a spreadsheet: date, assistant, question, whether you were named, whether you were linked, which competitors were named, and which sources were cited.
4. Look at the cited sources. If Perplexity keeps citing one directory for your category and you're not in it, that's your next task.
5. Check referral traffic. In your analytics, look for visits referred by chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. ChatGPT usually adds utm_source=chatgpt.com to the links it shows. Visits from assistant apps often arrive without a referrer and land in "direct", so treat these numbers as a floor.

Answers vary between runs, so judge the three-month trend, not a single result. Google counts clicks from AI Overviews and AI Mode inside Search Console's normal web search figures, with no separate filter.

Where Pilgud fits

We do this work as part of Pilgud Services – classic Google rankings have their own SEO service: crawler access, structured data, content that answers your customers' questions, and earning the mentions that make assistants confident about you. We did it on pilgud.com first. For local visibility, read our guide to local search, Google Maps and AI answers. If you'd rather look at it together, get in touch.