What Is GEO (Generative Engine Optimization)? How to Rank in ChatGPT and Google AI Mode in 2027

47 min read

If you're searching “what is GEO,” odds are you've already run the test yourself: you asked ChatGPT or Google's AI Mode about your industry in your city, something like “best dental clinic in Valencia” or “accounting firm in Andorra la Vella,” and your business didn't show up. Two competitors did, along with a directory and a website you'd never even heard of.

That's not a fluke. More and more people ask an AI before they compare websites, and the answer arrives already written: with names, pros and cons and, sometimes, a recommendation. If your business isn't in that answer, then for that customer it doesn't exist.

A new acronym has grown up around all this, along with a lot of noise and quite a few tricks that don't work. In this guide I separate what's official from what's my own judgment: what GEO is, how AI tools decide who to cite, what I'd do step by step to get them talking about your business, and how to measure it without fooling yourself.

Definition

What is GEO (generative engine optimization)?

GEO (generative engine optimization) is the work of getting the AI tools that answer questions, such as ChatGPT, Gemini, Claude, Perplexity and Google's AI Mode, to understand your business correctly, mention it and cite your website as a source when someone asks about what you do.

Put more simply: SEO works to get your website into the list of results; GEO works to get your business inside the answer. And that answer doesn't have ten slots. It has three or four names.

The term wasn't coined by an agency. It comes from a paper by researchers at Princeton and IIT Delhi, published in November 2023 and presented at the KDD conference in 2024. In their experiment, citing sources, adding quotations or including statistics boosted a page's visibility in generated answers by up to 40%. I'll come back to that study, because there's a takeaway in it that almost nobody talks about.

What does GEO mean in marketing and AI?

In digital marketing and AI, GEO stands for generative engine optimization: optimizing for search engines and assistants that write the answer instead of showing you a list of links. That's why you'll also see it searched as “GEO SEO,” “AI SEO” or “SEO for AI.” It has nothing to do with geotargeting or location-based “geo marketing,” even though the acronym invites the mix-up. In fact, when I searched the Spanish equivalent of “GEO positioning” on Google, its own AI summary started talking about satellite positioning.

What is AEO (answer engine optimization), and what is LLMO?

They're different names for almost the same work. Every agency and every tool has picked its own acronym, and it's worth knowing them all so you don't end up paying twice for the same thing:

Acronym	Stands for	Where the focus is
GEO	Generative Engine Optimization	Showing up and getting cited in AI-generated answers: ChatGPT, Gemini, Claude, Perplexity, AI Mode
AEO	Answer Engine Optimization	Being “the answer”: it started with featured snippets and voice assistants, and today it includes AI answers
LLMO	Large Language Model Optimization	Getting language models to know your brand and describe it accurately
AIO or AI SEO	AI Optimization	An umbrella term for all of the above (careful: AIO is also used as shorthand for AI Overviews)

In practice, the work is the same. Asked about AEO and GEO in its official guide, Google answers that, from its point of view, optimizing for AI search is optimizing for search.

Comparison

GEO vs SEO: the differences and what doesn't change

GEO doesn't replace SEO; it's built on top of it. AI tools that search the web need pages that are crawlable, indexed and trustworthy, which is exactly what SEO takes care of. What changes is the result you're after, the weight of what happens off your website and the way you measure it.

	SEO	GEO
Goal	Get your website into the results and earn clicks	Get your brand into the answer and your website among the cited sources
What the user sees	A list of links	A written answer with a handful of names and a few links
Unit of work	The keyword and the page	The full question, the sub-questions the AI generates, and the entity: your business
Weight of off-site signals	Links and reviews	What's said about you in media, directories, forums and reviews, with or without a link
Where it plays out	Google and Bing	AI Mode, AI Overviews, ChatGPT, Gemini, Claude, Perplexity and Copilot
How it's measured	Clicks, impressions, rankings and CTR	Appearances, citations, how you're described and visits from assistants
Stability	Fairly stable rankings	Answers that change from one query to the next

Is SEO dead, or is GEO just SEO with a new name?

Neither, quite. SEO isn't dead: Google says its AI features rely on the same ranking and quality systems as Search, so SEO best practices still apply. And GEO is almost SEO with a new name, but not entirely. If your SEO foundation is broken, GEO won't fix it. What changes is that there are more platforms, what's said about you off your website carries more weight, and measuring is harder.

That's why, in our work as an SEO agency in Andorra, GEO doesn't start from scratch: it's the next layer. Here I'll focus on helping you understand how it works so you can get started on your own.

Data

Why GEO matters: what the data says

GEO matters because more and more searches are resolved inside the answer, with no click at all. And that comes from sources that don't sell GEO services.

Pew Research Center analyzed the real browsing activity of 900 U.S. adults in March 2025: 68,879 Google searches, 18% of which showed an AI summary. When that summary appeared, people clicked on a traditional result about half as often:

And adoption is already massive, according to the latest official figures:

And it doesn't only affect winning new customers. Picture a client who's been buying from you for years and one day asks an AI what the best option in your category is. If three other brands come up and yours doesn't, they had no doubts until that moment. Now they do.

If AI doesn't name you, you don't lose a click: you lose your spot on the shortlist.

How it works

AI search visibility: how ChatGPT, Gemini, Claude and Perplexity decide who to cite

AI tools cite pages they can retrieve at the moment they answer and that look useful and trustworthy for each part of the question. To understand this, you need to separate two things that almost everyone mixes up: what the model knows from memory and what it looks up when it answers.

When the AI searches, the process looks something like this:

• User's question
• Related queries
• Retrieves pages
• Picks reliable passages
• Answer with citations

Google calls this query fan-out: the system breaks the question down into subtopics and runs several searches at once. OpenAI describes something similar: ChatGPT rewrites your question into one or more targeted queries that it sends to its search providers. In business terms: you're not competing for one keyword; you're competing for every one of those sub-questions.

How do you rank in ChatGPT, and how does it choose the sources it cites?

Ranking in ChatGPT, what some people call “ChatGPT SEO,” starts with understanding that it cites the pages its web search finds, and that search combines third-party search providers with content from partners. OpenAI says any public website can appear as long as it doesn't block its search crawler, OAI-SearchBot, and that it ranks results using several relevance and trust factors. It doesn't name its providers: “ChatGPT uses Bing” is a widely repeated inference, not something OpenAI publishes.

Each platform has its own way in:

Platform	Where it gets what it cites	What has to be able to get in	How to check
AI Overviews and AI Mode	Google's index, with query fan-out	Googlebot, plus the Search generative AI control in Search Console set to “Include”	Search Console, Generative AI performance report
Gemini (app)	Google Search, to ground its answers	Googlebot, and not blocking Google-Extended	Manual tests
ChatGPT	Its web search: third-party providers and partners	OAI-SearchBot	Analytics (utm_source=chatgpt.com) and manual tests
Claude	Its web search	Claude-SearchBot (and Claude-User for visits a user requests)	Manual tests and visits from claude.ai
Perplexity	Its own index	PerplexityBot (and Perplexity-User at a user's request)	Visits from perplexity.ai and manual tests
Copilot	Bing's index	Bingbot	Bing Webmaster Tools, AI Performance report

A summary of documentation from Google, OpenAI, Anthropic, Perplexity and Microsoft. None of them publishes its exact selection criteria.

Why does AI recommend my competitors and not me?

When I look into a case like this, I almost always find one of these four causes:

• They can't read you: a bot blocked in robots.txt, at the CDN or by a security plugin, or content that only loads with JavaScript.
• You don't answer the question: your website talks about “end-to-end solutions,” and your competitor has a page that answers exactly what's being asked.
• Nobody talks about you off your website: your competitor shows up in directories, comparison pages, forums and media; you only show up on your own site.
• The AI isn't sure who you are: a different name, address or set of services depending on the source.

The data backs up the third one in particular:

The first figure says the top 10 helps, but it's no longer enough: fan-out goes looking for answers on pages that don't rank high for the original query. The third is the takeaway from the GEO paper that almost nobody mentions: the websites that weren't ranking first were the ones that gained the most. GEO gives a real shot to whoever isn't the biggest, as long as they're the clearest.

Google

Google AI Mode and AI Overviews: how they work

AI Overviews are the AI-generated summary that appears above the results for some searches; AI Mode is a full conversational search experience, built for complex questions and follow-ups. Both run on Gemini models and draw on Google's index, so on Google, GEO starts with SEO.

What is Google AI Mode and how does it work?

It's a search mode where AI writes the answer. According to Google, the system breaks your question down into subtopics, fires off a burst of simultaneous queries and returns an answer with links to the websites it used. You can ask by text, voice or photo, and keep asking follow-up questions in the same conversation. There's nothing to set up to use it: it appears as a button on the results page and in the Google app, and you can also go straight to google.com/ai.

	AI Overviews	AI Mode
Where it appears	Above the results, only when Google thinks it adds something	In its own search mode, separate from the classic results
For which questions	Ones that can be answered with a summary	Complex ones, comparisons and ones that call for follow-up
How it searches	May use query fan-out	Uses query fan-out heavily
Links	Supporting links inside the summary	Supporting links and, if there's no reliable answer, web results
Requirements to appear	An indexed page that's eligible to be shown with a snippet	The same

Google warns that the two may use different models and techniques, so their answers and links won't always match.

What is an AI Overview and how do you show up in one?

An AI Overview is the AI-generated summary Google places above the results, with links to the websites it used. To appear as one of those links, here or in AI Mode, your page needs to be indexed and eligible to be shown in Google with a snippet. Google says there are no additional requirements: no special markup and no new files. Meeting them doesn't guarantee you'll appear; from there, quality and how well you answer each sub-question decide. Watch out for nosnippet and max-snippet: they limit what Google can show from your page, including in its AI features, and sometimes a plugin sets them without anyone noticing.

Step 1

Crawling: let in the bots that can cite you

An AI can only cite what it can read. This is the most technical part of GEO and the easiest to check: if a platform's search crawler can't get into your site, that platform won't cite you, no matter how good your content is. The tricky part is that there are lots of bots, and each one does something different:

Bot	Owner	What it's for	If you block it
Googlebot	Google	Indexing for Search, AI Overviews and AI Mode	You disappear from Google, including its AI features
Google-Extended	Google	It isn't a separate bot: it's a permission that controls whether your content is used to train Gemini and to ground answers in the Gemini app	You stay in Search (it affects neither inclusion nor ranking), but the Gemini app can't use your content to ground its answers
Bingbot	Microsoft	Bing's index, which Copilot draws on	You disappear from Bing and from Copilot's citations
OAI-SearchBot	OpenAI	Showing websites in ChatGPT's search answers	You don't appear in those answers (at most, as a navigational link)
GPTBot	OpenAI	Content that may be used to train its models	Your content isn't used for training; ChatGPT search isn't affected
ChatGPT-User	OpenAI	Visits a user requests inside a conversation	OpenAI warns that robots.txt may not apply, because a person initiates the action
Claude-SearchBot	Anthropic	Improving Claude's search results	You may lose visibility in Claude's answers that use search
ClaudeBot	Anthropic	Training its models	Your future content stays out of training
PerplexityBot	Perplexity	Showing and linking websites in its results; according to Perplexity, it doesn't train models	You don't appear in its results

Sources: official documentation from Google, OpenAI, Anthropic and Perplexity. OpenAI and Perplexity note that robots.txt changes can take about 24 hours to take effect.

Should I block GPTBot or Google-Extended?

Keep the two decisions separate. Showing up in answers depends on search bots; handing over content to train models depends on training bots. They're independent of each other: OpenAI confirms you can allow OAI-SearchBot and block GPTBot.

My take for a business that lives on winning customers, not for a publisher that sells its content: I wouldn't block any of them. What a model learns in training also shapes how it describes your brand when it isn't searching. And Google-Extended has a nuance almost nobody mentions: besides training, it controls whether the Gemini app uses your content to ground its answers. If you still don't want to hand over content to train models, here's how to do it:

robots.txt template, ready to copy

# Show up in AI answers without handing over content to train models.
# Search bots (Googlebot, Bingbot, OAI-SearchBot,
# Claude-SearchBot and PerplexityBot) stay untouched: they can still get in.

User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

# Google-Extended: think twice. Blocking it also stops
# the Gemini app from using your content in its answers.

And the usual technical SEO basics: important content in the HTML rather than hidden behind JavaScript, fast loading and a complete mobile version. If you're going to rebuild your website, a web design in Andorra project that comes with all of this solved from day one will save you months. Here's what that looks like in these 13 real websites designed with AI.

Step 2

Citable content: answer first and add what nobody else has

An AI cites the passage that best answers each sub-question, and it prefers one that adds something verifiable. In practice, that takes two things: answering up top, without beating around the bush, and offering what isn't already on twenty other websites.

Google calls this non-commodity content: expert or experience-based contributions that go beyond what everybody already knows. A list of generic tips that anyone, or any AI, could write gives nobody a reason to cite you. Here's how I'd see it for a renovation company in Madrid:

The second one answers in the first sentence and gives a figure only that company has: exactly what an AI can cite without worrying about getting it wrong.

What type of content does AI cite most?

There's no official list, but the GEO paper offers some measured clues. On its benchmark of 10,000 queries, these three techniques boosted visibility the most:

And the one that performed worst: repeating keywords. In the Perplexity test, keyword stuffing came out 10% below doing nothing at all. What was already a bad idea in SEO is downright counterproductive in GEO.

On top of that, as my own judgment, I'd add: pages that answer specific questions (ballpark price, timelines, requirements), honest comparisons, tables when you're comparing options, and a visible author with demonstrable experience, which is what Google calls E-E-A-T.

Which pages do AI tools want to cite?

The ones that give them reasons to recommend you. When someone asks for “a divorce lawyer in Madrid who isn't expensive,” the AI needs something to compare, and most websites don't give it that. This is my own judgment, but I see it clearly: a service page an AI feels confident recommending looks more like a good comparison sheet than a brochure.

In an online store, the page that gets cited is the product page. The test I'd run on each one: if the reader had an expert sales associate standing in front of them, would they still have a question? If the answer is yes, that product page is missing content.

Does structured data help with GEO?

It helps, but it isn't magic. Google says it isn't required to appear in its AI features and that there's no special schema for AI, although it's still useful for rich results. I'd get the basics right: Organization or LocalBusiness with exact details and links to your official profiles (sameAs), Product if you sell online and Article on the blog. Always faithful to what's actually visible on the page.

Step 3

Entity and brand: make sure AI knows exactly who you are

To an AI, your business is an entity: a name with a set of attributes, like what it does, where, for whom and what people think of it. If every source says something different, the AI hesitates. And when it hesitates, it cites someone else.

Signal	Where it has to match	Common mistake
Business name	Website, Google Business Profile, social media and directories	Three versions of the name depending on the site
What you do	Homepage, About page and profiles	“End-to-end solutions” instead of specific services
Where you work	Website, Google Business Profile and directories	A different address, phone number or hours depending on the source
Who it's for	Service and industry pages	“For all kinds of clients”
Who's behind it	Team page and LinkedIn	Not a single name or face
Proof	Case studies, reviews and mentions	Cases with no data, no context or no date
Official profiles	Linked from the website and in the structured data	Abandoned profiles telling a different story

Your About page deserves more love than it usually gets. It's the page an AI uses to answer “who is this company?”, so in the first few lines, with no frills, it should say what you do, where, since when and for whom.

What does it mean to move from keywords to entities?

It means you stop thinking about which words to repeat and start thinking about what the AI needs to know about you. A dental clinic in Bilbao gains nothing by writing “dentist in Bilbao” ten times. It gains if the AI can confirm, across several sources, that it's a clinic in Bilbao, with these licensed specialists, these treatments, these ballpark prices and these reviews on Google and on Doctoralia.

And that all those sources tell the same story. If your website says “advanced implant dentistry,” the local news site calls you a “family clinic” and the directory lists you under a different category, the AI doesn't know what to believe. That consistency across sources is often called entity consensus, and it's what turns three scattered mentions into a reliable description.

Do you need to be on Wikipedia?

No. It helps if your business meets Wikipedia's notability guidelines, which require significant coverage in independent sources, but writing your own company's article runs straight into its conflict-of-interest rules. Wikidata is more accessible for basic facts (name, website, headquarters), always with references. For most small businesses, the priority comes earlier: business profiles, industry directories and mentions in online media.

Step 4

Mentions: get people talking about you beyond your website

Mentions are the factor most closely tied to showing up in AI answers, and the one most often neglected. In Ahrefs' study of 75,000 brands, branded web mentions had a 0.664 correlation with appearances in AI Overviews, compared with 0.218 for backlinks. It's correlation, not causation, but the message is clear: AI trusts what others say about you more than what you say about yourself.

These are the places where I'd work on mentions, always online and always where your customers already are:

All of this is online work for a simple reason: whatever leaves no trace on the internet, AI can't read, link to or verify. That's why, in my view, a mention in an outlet the AI knows counts even without a link, or with a nofollow one: the model reads the text, not the link attribute.

And how do you know whether an outlet carries weight with AI before you invest time in getting featured there? A simple trick: turn off web search in ChatGPT or Claude and ask what it knows about that outlet, using its exact name. If it answers in detail, the outlet is in its memory; if it barely knows anything, that outlet carries little weight. Compare it with a major outlet and with a made-up name to calibrate. It's not a guarantee, but it helps you choose. You'll find the prompt further down.

AI doesn't believe what you say about yourself. It believes what others repeat, on sites it considers trustworthy.

Step 5

Reviews and business profiles

Business profiles and reviews are the most direct source for local questions, like “which accounting firm in Andorra la Vella would you recommend?” or “best physical therapist in Seville.” Google confirms it in its guide: Google Business Profiles and Merchant Center feeds can make your products and services visible in its AI answers.

• A complete Google Business Profile: categories, services, hours, real photos and answers to users' questions.
• Bing Places, because Copilot relies on Bing, and Apple Business Connect for people searching from an iPhone.
• Reviews on Google and on your industry's platforms, all of them answered, the good ones and the bad ones.
• A Merchant Center product feed if you run an e-commerce store in Andorra, with prices and availability kept up to date.

A note that isn't an official rule: when an AI summarizes reviews, it uses the text of the reviews, not just the star rating. Ten reviews that explain what problem you solved are worth more for GEO than fifty stars with no comment.

Step 6

First-party data, video and images: what AI can't make up

First-party data is the hardest thing to copy and the easiest to cite. An AI can write a thousand generic articles about your industry, but it can't make up how long your projects take, what your service costs on average or what your customers ask before they hire you.

Step 7

GEO in Andorra: the same work, in four languages

In Andorra, GEO is done in Catalan, Spanish, French and English, because your customers ask in all four and AI tools don't answer the same way in each language. You'll see it as soon as you run the test: a question in French tends to lean on French-language sources. If your business only exists in Spanish, then for that customer it doesn't exist.

The same thing happens in Spain, on a different scale. A hotel in Mallorca gets questions in Spanish, Catalan, English and German; a business in Barcelona or Valencia has customers who ask in Catalan or Valencian; one in Bilbao or Vigo, in Basque or Galician. And in the U.S., a business in Miami or Los Angeles will often get the same question in English and in Spanish. Every language you sell in is a different AI answer that you need to work on.

• One version of the website per language, with its own URL, hreflang and adapted content, not translated word for word.
• The same business name in every language, even if you translate the description.
• Mentions in each market's online media: Andorran outlets in Catalan, French ones if you sell across the border.
• The audit prompt set, repeated in every language.

It's also a competitive advantage: in a small market, or in a language with less content, there are few sources on each industry, and many businesses only work in one or two languages. Every page and every mention you add carries more weight.

Defense

Reputation: GEO is also defense

What an AI says about your brand when someone asks for opinions is largely decided by forums, Reddit and long-form reviews, not by your website. If there's nothing of yours there, someone else decides it for you. That's why GEO isn't only about showing up: it's also about making sure that, when you do, what's said about you is true.

What happened to Codesmith, and what does it teach us?

Codesmith is a New York coding bootcamp that reached $23.5 million in revenue. According to the investigation published by consultant Lars Lofgren in October 2025, the most active moderator of the industry's go-to forum on Reddit was the co-founder of a competing bootcamp, and over 487 days, 425 negative posts about Codesmith were published there. When Lofgren asked ChatGPT whether Codesmith was a good bootcamp, the answer repeated the same doubts raised in those threads.

The company's revenue fell by about 80% from its peak, and its leadership attributes roughly half of that drop to Reddit. The other side denies that account and blames the collapse on the industry-wide downturn, so it's worth reading for what it is: a disputed case. But the mechanism is real, and it can happen to anyone.

If you don't write your company's version on the sites AI reads, someone else will.

What is a defensive corpus?

It's the detailed, honest and verifiable content about your business that you build on the sites AI tools consult, before you need it. Codesmith didn't have one: when the criticism started, there was nothing to balance it out. Reversing a narrative the models have already absorbed is slow and expensive; preventing it isn't nearly as hard.

• Fully developed case studies and testimonials, with names and context, not just stars. My sense is that a text describing an experience tells a model more than a high average rating.
• Useful, signed answers in the forums and communities where your industry is discussed, before there's a problem.
• Your own content that tackles the uncomfortable questions: prices, timelines, what happens if something goes wrong, why you're not for everyone.
• Reviews on Google and on industry platforms, always answered, including the bad ones.

How do people manipulate AI, and how do you defend yourself?

It's worth knowing the techniques so you can recognize them, not so you can use them. These are the ones that worry me most, with what I'd do in each case:

Warning sign	What may be going on	What I'd do
Criticism that keeps coming back, at regular intervals, in the same forum	A slow campaign: a steady drip doesn't set off alarms, but it builds up a tone the AI takes at face value	Log the author and date of every negative mention, look for patterns and check who moderates that forum
Forum threads ranking on Google for “[your brand] reviews” or “[your brand] scam”	Those pages feed both Google and AI search engines at the same time	Monitor those searches every month and respond with data and transparency wherever you can
Fake reviews, positive or negative	They violate the policies of Google and of review platforms	Report them through each platform's channels and document everything
“Summarize with AI” buttons that hide instructions	In two months, Microsoft detected more than 50 instructions from 31 real companies trying to get the assistant to remember them as a trusted source	Never use them on your website: Microsoft treats it as an attack. As a user, review your assistant's memory

Measurement

How to measure your AI visibility

AI visibility is measured in layers, because no single tool sees everything: Search Console for Google, Bing Webmaster Tools for Copilot, Analytics for the visits that come from assistants, and a fixed set of prompts for what nobody else measures.

What to measure	With what	What it tells you	What it doesn't tell you
Impressions in AI Overviews and AI Mode	Search Console, Generative AI performance report	How many times a link to your site was shown, by page, country, device and date	It only reports impressions. Clicks are in the regular performance report, under the “Web” search type, mixed in with everything else
Citations in Copilot and Bing	Bing Webmaster Tools, AI Performance report	How many times, and which pages, are cited as a source	The clicks those citations generate
Visits from assistants	Analytics: utm_source=chatgpt.com and visits from perplexity.ai, gemini.google.com or claude.ai	Visits, behavior and conversions	The times you're mentioned without anyone clicking
Whether you're recommended, and how	A fixed set of prompts, every month	Whether you appear, in what order, what's said about you, who shows up alongside you and which sources are cited	A statistical figure: it's a sample
Brand	Search Console (searches for your name) and reviews	Whether more and more people are searching for you by name	Where that interest comes from

Google points out in its guide that no third-party tool has access to its internal ranking or AI systems. AI visibility tools work with samples of prompts: they're good for spotting trends, not for exact figures.

How many customers is AI bringing you, or taking away?

No report will tell you, but you can estimate it by cross-referencing four free sources. This is the method I'd use:

What I wouldn't do is judge GEO only by the share of visits coming from AI today, which in almost every industry is small. The real risk is a sudden shift. Ask yourself this: if Google made AI Mode the default tomorrow, would you show up?

How do I know if ChatGPT recommends my business?

Ask it the way a customer would, not by your brand name. Without paying for tools, the most reliable approach I know is a fixed set of 10 to 20 real questions, run every month on the same platforms, without signing in or in a private window to reduce personalization, and logged in a spreadsheet. And in Analytics, filter the visits with utm_source=chatgpt.com: OpenAI adds that parameter to the links that come out of ChatGPT, so you can see how many people arrive from there and what they do next.

Templates

Prompts to audit your AI visibility

These are the prompts I'd use in a first audit. Replace whatever is in brackets and run them in ChatGPT, Gemini, Google's AI Mode, Claude and Perplexity. Save the answers exactly as they come: you'll need them for prompt 6.

1 · Discovery, without mentioning your brand

I need [service] in [area]. Which companies would you recommend, and why?
Give me three options with their strengths and weaknesses, and list the specific
sources you're basing your answer on.

2 · Comparison with your competitors

Compare [your company] with [competitor 1] and [competitor 2] for someone who
needs [specific need]. What are the pros and cons of each one?
Cite the sources for every claim.

3 · What the AI knows about your brand

What do you know about [your company], based in [city]? Tell me what it does,
what kind of client it works for, what its customers think of it and how long
it has been around. If you're not sure about any detail, say so. List your sources.

4 · The sources it checks for your industry

To answer the question “[a real customer question],” which websites,
directories, media outlets, comparison sites or forums would you check? List the
specific sources, not types of sources.

5 · The same question in Spanish, Catalan and French

Necesito [servicio] en Andorra. ¿Qué empresas me recomiendas y por qué?
Indica las fuentes en las que te basas.

Necessito [servei] a Andorra. Quines empreses em recomanes i per què?
Indica les fonts en què et bases.

J'ai besoin de [service] en Andorre. Quelles entreprises me recommandes-tu
et pourquoi ? Indique les sources sur lesquelles tu t'appuies.

6 · Putting the results into a table

Below are the answers several AI assistants gave me to the same questions.
Build a table with these columns: platform, question, does [your company]
appear? (yes/no), position, what it says about it, competitors that appear,
cited sources and possible errors about [your company].
Underneath, summarize which sources come up most often and in which ones
[your company] doesn't appear. Don't make up anything that isn't in the text I give you.

7 · Reputation: what people think of your brand

What do customers think of [your company], based in [city]? Summarize the
positives and negatives you find, and tell me which specific sources each
opinion comes from (forums, reviews, media). If a criticism keeps coming up, tell me where.

8 · Memory test for a media outlet (with web search turned off)

Without searching the internet, what do you know about [exact name of the outlet]?
What topics does it cover, for which audience and from where? If you don't know it
or aren't sure, say so clearly instead of guessing.

Prompt 4 and the last column of prompt 6 are the most valuable ones: the sources AI already cites tell you where you need to be. That's your to-do list for step 4.

Myths

GEO myths: llms.txt, magic schema and other tricks

Most of the “GEO tricks” going around have no official backing, and the platforms themselves debunk some of them. The funny part is that, while I was putting this guide together, Google's own AI summary for the Spanish equivalent of “how to show up in ChatGPT” recommended three of them.

Myth	Reality	Who says so
“You need an llms.txt file”	Google Search doesn't use it, and creating one neither helps nor hurts your visibility on Google	Google
“Verify your domain in ChatGPT to show up”	That verification is for publishing GPTs in the GPT Store and for enterprise single sign-on. It has nothing to do with appearing in answers	OpenAI
“There's a special schema for AI”	There isn't. Structured data isn't required for AI Overviews or AI Mode	Google
“Chunk your text into 100–200-word blocks”	You don't need to chunk content for AI to understand it, and there's no ideal length	Google
“Repeat the keyword to get cited”	It was the worst technique in the GEO experiment: on Perplexity, 10% worse than doing nothing	GEO paper
“Blocking Google-Extended takes you out of AI Overviews”	It doesn't affect inclusion in Search. To opt out of AI Overviews and AI Mode, there's the Search Console control	Google
“ChatGPT uses Bing, so Bing is all you need”	OpenAI talks about third-party search providers without naming them. Bing matters, especially for Copilot, but it isn't everything	OpenAI
“You can pay to get ChatGPT to recommend you”	Ads in ChatGPT are separate and labeled, and OpenAI says they don't influence the answers	OpenAI

What is an llms.txt file, and does it actually do anything?

llms.txt is a Markdown file with a summary of a website and links to its key pages, meant for language models to read when they answer. Jeremy Howard proposed it in September 2024, and that's still what it is: a proposal. Google explicitly says its search doesn't use it, and it doesn't appear in the crawler documentation from OpenAI, Anthropic or Perplexity. It can make sense for developer documentation; for a small business, it's not a priority, and I wouldn't pay anyone to make one.

Mistakes

Common mistakes when getting started with GEO

The most expensive GEO mistakes aren't technical: they're about approach. These are the ones I see most often:

Summary

How to do GEO (AI SEO) step by step: summary

GEO comes down to seven steps, plus defending your reputation and measuring, and the order matters: first, make sure AI can read you; then, that it understands you; finally, that it has reasons to recommend you.

Does GEO work for a small or local business?

Yes, and sometimes even more than for a big brand. Local questions (“which physical therapist in Escaldes-Engordany…”, “which dentist in Zaragoza…”) have fewer sources to choose from, so a complete business profile, reviews that tell real stories and a few mentions in the right places make all the difference. The GEO paper points the same way: the websites that weren't at the top were the ones that gained the most.

Example

GEO example: an accounting firm in Andorra

Here's how I'd approach GEO for an accounting and business-services firm (a gestoria, as they're called locally) in Andorra la Vella that works with residents and with companies setting up in the country. It's a hypothetical case, not a real client, and it works as a template for any service business: change the industry and the questions, and the framework stays the same.

Step	Decision made	What gets delivered
Audit	20 real questions in four languages, such as “which firm in Andorra can help me set up a company?”, “gestoria per obtenir la residència a Andorra” or “cabinet comptable Andorre création de société”	A spreadsheet with appearances, competitors and cited sources
1 · Crawling	Review robots.txt, the CDN, the security plugin and the AI control in Search Console	Search bots allowed and confirmed in the logs
2 · Content	One page per service that answers up top (requirements, timelines, ballpark price) and one guide per procedure: active residency, passive residency, company formation	Pages with an author and links to the official sources
3 · Entity	The same name, address and description on the website, the business profile, the directories and LinkedIn	A consistency sheet and organization structured data
4 · Mentions	Start with the sources AI already cites: online professional directories, comparison sites and forums where newcomers to the country ask questions	A list of target sources, ranked by how often they come up
5 · Reviews	Ask for a review when each procedure is completed, in the client's language, describing the case	Reviews with useful text on Google and in the directories
6 · First-party data	Publish the real timelines for each procedure, based on anonymized files	A timeline table no competitor has
7 · Languages	Catalan, Spanish, French and English versions with hreflang	Four adapted versions, not literal translations
8 · Defense	Monitor what comes up for “[name] reviews” on Google and in AI tools, and in the forums where newcomers ask questions	Monthly alerts and client cases told in detail
9 · Measurement	The same 20 questions every month, plus Search Console, Trends and GA4	A monthly trend report, with no isolated screenshots

What matters in this example isn't the list, but where each decision comes from. The audit doesn't just tell you whether you appear: it shows you which sources AI uses for your industry. If, for “residency in Andorra,” it always cites the same two directories and one forum, that's where the work starts, not in publishing twenty more articles.

What I wouldn't do

• Pay for an llms.txt, a “secret GEO schema” or any other magic file for AI.
• Create a page for every variation of a question to “cover the fan-out”: Google considers that scaled content abuse.
• Buy mentions on websites nobody reads, or buy fake reviews.
• Pay for campaigns against competitors in forums, or hide instructions for AI on my website. It's deception, and when it comes to light, the damage falls on whoever did it.
• Block “all AI bots” with a single click without knowing which ones are for search and which ones are for training.
• Exclude my website from AI Overviews and AI Mode in Search Console without a clear business reason.
• Write my company's Wikipedia article myself.
• Believe anyone who promises to get me “ranked first in ChatGPT.” Nobody controls what an AI answers.
• Abandon SEO to “do GEO.” Without that foundation, the rest doesn't hold up.

The bottom line

GEO isn't a discipline that changes everything. It's the same old SEO, with more platforms, more weight on what's said about you off your website and less precise measurement. If I had to get a business ready for 2027 today, this would be my order:

• Make sure AI can read you
• Make sure it understands you
• Get people talking about you
• Get recommended

In that order. Whoever starts at the end, buying mentions or magic files, is building on a website that AI can't read or doesn't understand.

AI doesn't reward whoever publishes the most. It rewards whoever is easiest to cite without fear of getting it wrong.

Frequently asked questions