Updated in October 2026.
Picture this: someone asks ChatGPT which clinic in Valencia does dental implants, and the answer names yours, with a link to your website. Or they ask Perplexity to compare three invoicing tools, and yours shows up, quoting your own pricing page.
It isn't science fiction. Google has shown AI-generated summaries (AI Overviews) in Spain since March 2025, and its AI Mode is now available in Spain, in Spanish. And in May 2026, Google Analytics launched a channel of its own for visits that come from assistants like ChatGPT, Gemini or Claude: if it gets its own channel, that traffic already exists.
In this guide you'll find what decides whether an AI cites you, what to change on your website (robots.txt, the much-talked-about llms.txt and structured data, explained without the hype) and how to measure the visits that reach you from assistants.
Why should I care about LLMs (Large Language Models)?
OK, "LLM" sounds like a spaceship, but it's simpler than that. Large language models (LLMs) are the engine behind ChatGPT, Claude, Gemini or Perplexity: they learned to write by reading huge amounts of text and, instead of giving you a list of links the way Google used to, they give you the answer directly, as if you were talking to an expert.
What matters for your website is where that answer comes from. When the question is about businesses, prices or anything recent, the assistant doesn't rely on memory alone: it searches the web there and then, reads several pages and links the ones it uses. Google explains that AI Overviews and AI Mode retrieve those pages with its usual ranking systems and run several related searches at once (what it calls "query fan-out"); ChatGPT searches with its own crawler and with third-party search providers.
The good news is that SEO done right is already the foundation for showing up in AI answers. Google says so itself: for its search engine, optimizing for AI search is still SEO. What GEO (ranking in AI answers) adds is looking after which pages can be quoted, letting each assistant's crawlers in and measuring what arrives from them.
The 5 golden rules for AI to like you (and quote you)
How do you win over these artificial "little brains"? With the same things that convince a good client:
- "Quotable" content with substance:
An assistant links the pages that back up its answer, so the clearer and more verifiable yours is, the better. Microsoft explains it on its webmaster blog, writing about citations in Copilot: examples, data and cited sources build trust when an AI answer reuses your content.- Use specific data: Instead of saying "we're the best", publish the figure and where it can be checked. "1,048 appointment requests from Google in 2025, 82% of them all" says far more, and it's what the Clínica Albareda case study publishes.
- Cite your sources: If you use statistics, link to the original source (the ministry's report, not the blog that summarized it).
- Make it genuinely useful: Google's guidance asks for "non-commodity" content: your experience, your cases and your point of view, not a summary of what ten other articles already say (or of what the AI itself could write).
- Clear structure, like a good table of contents:
Well-organized content is easier to understand, whether a person or a machine reads it.- Headings and subheadings (H1, H2, H3): Use them to break up your text. Try to phrase subheadings as questions people would ask. For example: "How do I choose the best insurance for my van?"
- Lists and tables: Numbered or bulleted (like the one you're reading), and a table whenever you compare options or prices.
- Frequently asked questions (FAQs): A FAQ section is pure gold. Answer your customers' most common questions.
- Trust and authority (E-E-A-T):
AI, like Google, wants to recommend reliable, expert sources. Google uses E-E-A-T for what its systems try to recognize in a page: experience, expertise, authoritativeness and trustworthiness. It isn't a ranking factor in itself, but of the four, Google says, trust matters most.- Experience: Show that you know what you're talking about because you've lived it. If you run a clinic or a firm, share real cases (with your clients' permission).
- Expertise: Show your credentials: degrees, certifications, years in the field. And sign your articles with the name of whoever wrote them, linking to their profile: Google explicitly recommends it.
- Authoritativeness: Do others see you as a reference? Being mentioned by trade media, directories or lists of the best companies in your field counts, and so do links from other quality websites. Earned mentions, though: Google lists chasing inauthentic mentions among the myths that don't work.
- Trustworthiness: Your website must be secure (HTTPS), with clear contact details and a privacy policy. Keep your Google Business Profile and your Bing Places listing up to date: Google says its profile helps your business show up in AI responses, and Bing recommends it for local answers.
Imagine you're looking for a lawyer. Would you trust one who just appeared out of nowhere, or one with a physical office, years of proven experience, good reviews and colleagues who recommend them? AI thinks along the same lines. Having a real brand that people know and respect is pure gold.
- Structured data (schema): the "labels" for machines:
It's like putting labels on your content so a machine understands it the first time: "this is the company's name", "this is its address", "this is the article's author", "these are the FAQs". Google makes clear that no special schema is needed for AI Overviews or AI Mode and that the data must match what's visible on the page; it's still worth it for Google's rich results. WordPress plugins help with this, and it's easy to implement in Webflow too. - Speed and a good experience (UX), on mobile too!
Nobody likes waiting, least of all someone arriving from an answer with their question already asked. Google lists page experience among its recommendations for these features.- Fast and in text: It should load fast, and what matters should be written on the page, not inside an image: that's another of Google's recommendations for AI search.
- Easy to use on mobile: Lots of people search from their phone. Your website has to look and work great on small screens.
Let AI crawlers in (robots.txt)
No AI can quote what it can't read. Each company uses its own crawlers, and your robots.txt file decides which ones get in. What almost nobody explains is that they don't all do the same job: some read your site to answer whoever is asking, others collect text to train models, and you can let the first ones in without opening the door to the second.
| Company | Crawler | What it does | If you block it |
|---|---|---|---|
| OpenAI | OAI-SearchBot | Finds and links websites in ChatGPT search | You don't appear in its answers (at most, as a bare link) |
| OpenAI | ChatGPT-User | Visits a page when a user asks ChatGPT a question | OpenAI warns that robots.txt rules may not apply |
| OpenAI | GPTBot | Collects text to train its models | Your content isn't used for training; you can still appear in ChatGPT |
| Anthropic | Claude-SearchBot | Indexes websites to improve Claude's search results | You may lose visibility in its answers |
| Anthropic | Claude-User | Visits a page when a user asks Claude a question | Claude can't read your site to answer |
| Anthropic | ClaudeBot | Collects text to train its models | Your future content isn't used for training |
| Perplexity | PerplexityBot | Finds and links websites in Perplexity's answers (doesn't train models) | Perplexity recommends allowing it to appear in its results |
| Perplexity | Perplexity-User | Visits a page when a user asks a question | Perplexity says it generally ignores robots.txt |
| Googlebot | The usual crawler: it decides whether you appear in Google, including AI Overviews and AI Mode | Google stops reading your site, for AI features too | |
| Google-Extended | Not a separate crawler: a signal so Google doesn't use your content to train Gemini or to ground the Gemini apps' answers | No effect on your presence or ranking in Google Search |
Careful with Google: Google-Extended doesn't take you out of AI Overviews or AI Mode. If you ever wanted out of them without leaving Google, Search Console has its own setting for that (Search generative AI, under Settings), which keeps your site included by default.
A robots.txt that lets AI search crawlers in and keeps OpenAI's and Anthropic's training out looks like this:
# AI search crawlers: let them read and cite you
User-agent: OAI-SearchBot
User-agent: Claude-SearchBot
User-agent: PerplexityBot
Allow: /
# Model training: closed, in this example
User-agent: GPTBot
User-agent: ClaudeBot
Disallow: /If you sell a service and want to be known, shutting the door on training makes little sense; for a publisher that lives off its content, maybe it does. Think twice about Google-Extended: besides training, it stops the Gemini apps from using your site to ground their answers. It's your call, but make it on purpose, not with a robots.txt copied from another site.
And check your server too: OpenAI and Perplexity ask you to allow the IP ranges they publish, because a CDN or firewall that blocks "bots" can keep them out even when your robots.txt lets them in.
llms.txt: what it is and who actually reads it
llms.txt is a proposal by Jeremy Howard, published in September 2024 and updated in August 2026: a plain-text Markdown file at the root of your website (yoursite.com/llms.txt) with your business's name, a one- or two-sentence summary and links to your important pages. The idea is that an AI agent can understand your site without reading all of it. An example:
# Your business name
> What you do, for whom and where, in one or two sentences.
## Services
- [Service one](https://yoursite.com/service-one): what it solves and for whom.
- [Service two](https://yoursite.com/service-two): what it solves and for whom.
## Case studies
- [Client](https://yoursite.com/case-studies/client): the result, with its number.What's confirmed, and what isn't:
- It's a proposal, not a standard. Its author maintains it at llmstxt.org, which talks of thousands of sites that already publish one.
- Google Search doesn't use it. Its guide to AI search is clear: you don't need llms.txt or other "special" files to appear in Google, AI Overviews and AI Mode included, because its search engine ignores them; having one neither helps nor hurts.
- The AI labs publish one, but for their developers. OpenAI, Anthropic and Google have an llms.txt for their technical documentation. Yet neither OpenAI nor Anthropic mentions it in their guidance on getting a website into their answers: there they talk about robots.txt.
- Where it is used: by agents that read documentation (a coding assistant looking up a library, for example), by documentation platforms that generate one automatically, and by Lighthouse, Chrome's audit tool, which checks for it among its agentic browsing audits.
Our take: publish one, because it's little work and has no downside, but don't expect it to get you into ChatGPT on its own. What gets you there is everything above.
Bing, Copilot and IndexNow
Bing matters more than it seems. Copilot and Bing's AI summaries cite pages from the web, and OpenAI counts Microsoft among the search providers ChatGPT works with.
IndexNow is a ping your website sends when you publish, change or delete a page, so Bing recrawls it without waiting for its next visit. The ping is shared with the protocol's other search engines (Yandex, Seznam, Naver, Yep and Amazon); Google isn't among them. Microsoft recommends it precisely so AI answers use the current version of your page. You turn it on with a key file on your site and a call to its API on every change.
How to measure the visits AI sends you
Today you have four places to measure it:
- Google Analytics 4: since 13 May 2026 it has a channel of its own, AI Assistant, in the default channel group. It collects visits from assistants like ChatGPT, Gemini, Claude or Copilot, with the medium "ai-assistant". Note: visits from AI Overviews and AI Mode don't go there; they count as Organic Search.
- ChatGPT: adds utm_source=chatgpt.com to the links in its answers, so you'll see it in GA4 under that source.
- Search Console: the generative AI performance report shows your pages' impressions in AI Overviews and AI Mode (impressions, not clicks). Google says it has been available to all websites since 31 August 2026, although some properties don't see it yet.
- Bing Webmaster Tools: the AI Performance report (launched in preview in February 2026) counts how often Copilot and Bing's AI summaries cite your pages, and for which queries.
If you want your own definition, for example to separate an assistant GA4 doesn't recognize yet, create a custom channel group: one channel with the condition "source matches regex", placed above Referral. Google applies these groups to reports retroactively, so you'll see earlier months too. For example:
chatgpt|openai|perplexity|claude|copilot|gemini|deepseekAnd a fifth, the most direct: ask yourself. Write down the questions your client would ask an assistant before buying from you, put them to ChatGPT, Perplexity, Gemini and Claude every month, and note who gets named. We track in ChatGPT the questions someone would ask when looking for an agency like ours.
Common mistakes that make AI skip your website
Avoid these so you don't ruin your chances:
- ❌ Generic copy with no substance: "We're a leading company in the sector" doesn't cut it anymore. Bring data, examples, solutions.
- ❌ Disorganized content: A block of text with no subheadings or lists is a maze for AI (and for your readers).
- ❌ Ignoring mobile: If your website looks bad on a phone, you're missing a lot of opportunities.
- ❌ Blocking AI crawlers by accident: A robots.txt copied from another site or a firewall that blocks "bots" can keep OAI-SearchBot or PerplexityBot out. Check both.
- ❌ Counting on llms.txt alone: It's a useful file, not a shortcut.
- ❌ Manufacturing mentions: Paid reviews or articles to get people "talking about you" aren't what Google rewards; the mentions that count are the ones you earn.
A checklist for your website
If you only do one thing after reading this, go through this list:
- Your robots.txt lets in Googlebot, OAI-SearchBot, Claude-SearchBot and PerplexityBot, and you have decided on purpose what to do with GPTBot, ClaudeBot and Google-Extended.
- Your CDN or firewall doesn't block the IP ranges OpenAI and Perplexity publish.
- Each service has its own page, and its first sentence says what you do, for whom and where.
- Every important page has at least one figure that can be checked, with its source linked.
- There are FAQs written in your clients' own words.
- Articles carry the byline of a named person and a link to their profile.
- Structured data (organization, services, author, FAQs) matches what's visible on the page.
- Your Google Business Profile and your Bing Places listing are up to date.
- Search Console and Bing Webmaster Tools are verified, and IndexNow tells Bing about every change.
- In Search Console, the Search generative AI setting keeps your site included (it's the default).
- There's an llms.txt at the root of the site.
- Every month you look at GA4's AI Assistant channel, Search Console's generative AI report and Bing's AI Performance.
What we do on our own website
We don't recommend anything we don't do on haki.marketing. Our robots.txt names the AI crawlers and tokens of OpenAI, Anthropic, Perplexity, Google and Apple one by one and lets them all in, training included: we want the models to know us. We have an llms.txt that sums up who we are, what we do and the results published in our case studies. Every time we change a piece of content, the site tells Bing through IndexNow. And every article is signed, with its author's structured data.
For clients we do it as part of our GEO service, always on top of their SEO.
Conclusion: get to work!
Optimizing your website for AI isn't black magic: it's applying common sense and the good practices that, in many cases, you should already be applying for Google. Focus on original content that is well structured and signed, on a website AI crawlers can read, and measure what arrives. That way you make things easier not only for your clients but also for the artificial intelligences they're already asking.
robots.txt and llms.txt are the easy part, and the most talked about. What decides whether an assistant names you is the same thing that decides whether a client picks you: having something of your own to say and published proof that it works.
If you want to know what ChatGPT, Perplexity and Google answer today when someone asks about what you sell, we'll look at it with you on a discovery call.



