Skip to content
AI for SMBs

GEO: How to Get Found by ChatGPT, Gemini, and Perplexity

Wouter·6 February 2026·11 min read
TL;DR

OpenAI, Google and Perplexity each write down which names they read in robots.txt and what those are for. OpenAI and Perplexity also recommend allowing their IP ranges. For a visit a user sets in motion, robots.txt may not apply according to OpenAI, and generally does not according to Perplexity. How ChatGPT and Perplexity then choose who to name is not in the documents we read. What you can do to your text comes from one study in a test setup and from Google's guide. Part of it is our own advice, and we say so. How many clients it brings, we do not know.

Someone types into ChatGPT: "Which agency can automate my bookkeeping?" The answer may name a few agencies, with links. Are you one of them?

This post sets out what the makers of ChatGPT, Gemini and Perplexity publish about that themselves, what one study measured, and what you can do. Where a piece of advice is our own choice, we say so.

Corrected on 14 November 2026: this passage said "That's not a future scenario. That's now." and that millions of people use AI search engines daily. We had no source for that number.

What Is GEO?

GEO stands for Generative Engine Optimization: work meant to raise the chance that an AI answer uses and cites your page as a source.

With SEO you work on your place in the list of search results. With GEO the question is whether an AI service uses and names your page in an answer it writes itself. That split is our own.

Google sees it differently for its own search engine. Its guide to generative AI in Search says: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."

Corrected on 14 November 2026: this section said that being named in an answer is "more powerful" than a place in the search results and "a stronger endorsement than position 7 in Google". We had no source for that. It also said that AI models read your content, understand the context and "decide whether your information is valuable enough". What the makers publish themselves, and what is missing from it, is below.

What Does the Research Say?

The term GEO comes from a study by researchers from, among others, Princeton University and the Indian Institute of Technology Delhi. They changed web pages in nine ways and looked at how much of an AI answer then drew on that page.

These are three scores from the study's main table, on the first of two metrics. Higher is better.

  • Adding quotations: 27.2, against 19.3 with no change
  • Adding statistics: 25.2
  • Citing sources: 24.6, about level with writing more fluently (24.7)

Adding quotations and adding statistics were the top two on both metrics. This was measured in a test setup with the top five Google results and the language model GPT-3.5 Turbo. The study measures a source's share of an answer, plus a language model's rating of that source, including how likely a user is to click it. Real visitors or clients were not measured. The full table and the limits are in our GEO guide.

One more result, from a separate test in which all sources were changed at once: with three of the five methods in that test, the page ranked fifth in Google gained the most, while the page ranked first lost visibility with those methods. The researchers write that "factors such as backlink building should not disadvantage small creators". That is their reasoning in a test setup. It promises nothing for your site.

Corrected on 14 November 2026: this section said plus 40 percent for statistics, where the main table works out at about 31 percent (25.2 against 19.3), and plus 30 percent for citing sources (then listed as "Adding citations"), where it works out at about 27 percent. It also said that small websites benefit the most. The authors' institutes included Georgia Tech and the Allen Institute. Both were listed in the first version of November 2023; in the version of June 2024 those two authors are listed as independent.

What the Three Services Publish Themselves

Each of the three has its own names in robots.txt and writes down what they are for. What refusing costs is stated most clearly by OpenAI.

ChatGPT (OpenAI)

OpenAI describes its robots in an overview for site owners. For search there is OAI-SearchBot: "OAI-SearchBot is used to surface websites in search results in ChatGPT's search features." If you block that robot, OpenAI says: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links."

OpenAI recommends allowing OAI-SearchBot in robots.txt and allowing requests from its published IP ranges. For search results it can take about 24 hours after a change to robots.txt, according to OpenAI, before its systems follow it.

GPTBot is a different robot, for training models: "Disallowing GPTBot indicates a site's content should not be used in training generative AI foundation models." The two settings are independent. So you can allow search and refuse training. OpenAI gives that example itself.

Another name is ChatGPT-User, for a visit that a ChatGPT user sets in motion. About those OpenAI writes: "Because these actions are initiated by a user, robots.txt rules may not apply." For appearing in search results OpenAI points to OAI-SearchBot. The page lists a fourth name, OAI-AdsBot, for pages submitted as ads.

Gemini and Google Search (Google)

With Google these are two different things.

For the AI features in Google Search (AI Overviews and AI Mode) the regular search index applies. The documentation on AI features (updated 10 December 2025) says: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." Your page has to be in the index and allowed to show with a snippet. Google's guide is newer (updated 10 July 2026) and names one more condition: the site has to be included in the generative AI features in Search Console. More on that is in our guide to AI Overviews.

For the Gemini app there is a separate name in robots.txt: Google-Extended. Google fetches your pages with its regular robots and uses the name as a switch. The overview of Google's robots says you use it to manage whether what Google fetches from your site may be used for training future Gemini models and for grounding in the Gemini apps and in Vertex AI. Google describes grounding there as "providing content from the Google Search index to the model at prompt time to improve factuality and relevancy". For your place in Google Search it makes no difference: "Google-Extended does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search."

Our reading: if you refuse Google-Extended, you stay in Google Search and you tell Google not to use your pages for training future Gemini models or for answers in the Gemini app.

Perplexity

Perplexity has its own page about its robots. It says: "PerplexityBot is designed to surface and link websites in search results on Perplexity. It is not used to crawl content for AI foundation models." The advice resembles OpenAI's: allow the robot in robots.txt and allow requests from the published IP ranges. A change can take up to 24 hours, according to Perplexity.

Perplexity also has a second name, Perplexity-User, for a visit that a user sets in motion: "Since a user requested the fetch, this fetcher generally ignores robots.txt rules." The same description also says that Perplexity-User controls which sites these user requests can access. The page does not explain how those two sentences fit together.

If you use a firewall for your site, you may need to allow Perplexity's robots there separately. That is on the same page.

What Is Not in There

The pages from OpenAI and Perplexity are about access. How ChatGPT and Perplexity choose between the pages they are allowed to reach is not in them.

Google says a little more. The AI features in Search are, according to the guide, "rooted in our core Search ranking and quality systems". A list of factors and weights is not in the three Google pages we read.

What We Know About Clicks

Being named is not yet a visit. Pew Research Center tracked the browsing of 900 US adults in March 2025 and looked at 68,879 unique Google searches. 18 percent of those searches produced an AI summary. When there was an AI summary, people clicked a regular search result on 8 percent of visits to such a results page. Without a summary that was 15 percent. A link inside the summary itself was clicked on 1 percent of visits to such a page.

That study is about Google, about 900 US panel members and about one month. For ChatGPT, the Gemini app and Perplexity we have not read a study of clicks. The limits of the Pew study are in our guide to AI Overviews.

Google itself writes, without a figure: "people are increasingly gravitating to generative AI experiences to help them find information".

Corrected on 14 November 2026: this section gave three figures: more than 200 million weekly ChatGPT users ("source: OpenAI, 2025"), more than 100 million searches per week on Perplexity, and AI Overviews in 47 percent of searches in the US. None of the three had a link, and we have not read the publications. It also said that more and more people start with ChatGPT and not with Google, and that with SEO alone you miss "a growing segment" of your clients. We had no source for that.

How AI-ready is your business?

Take the free AI Readiness Quiz: 7 questions, 2 minutes.

Take the quiz →

7 Practical Tips to Get Found by AI

Seven things you can do. Each one says where it comes from: the study, Google's guide or our own choice. What is under "Do this" is always our own translation into practice. None of the seven is a guarantee.

1. Answer From Your Own Work

Google's guide is clearest on this: "Creating content that people find unique, compelling, and useful will likely influence your website's presence in generative AI search in the long run more than any of the other suggestions in this guide." Google advises writing it yourself, based on what you know about the topic, and considering what experience you can bring.

Do this: Write pages that answer a client's question from your own work, with your own numbers and the choices you make for clients. Putting the answer at the top is our own choice.

Corrected on 14 November 2026: this tip said that AI models look for content that answers questions, and that for a question about invoicing an AI wants to cite a step-by-step answer. We had no source for that.

2. Use Numbers and Statistics

In the study's main table, adding statistics scored second highest. A number is specific and can be checked, as long as you say where it comes from.

Do this: Replace a vague sentence like "many businesses save time" with a number you measured yourself or one from a source you name. Do not make a number up. In the study a language model wrote the statistics in; on your own site every number has to be true.

3. Cite Sources and Experts

Citing sources scored 24.6 in the main table, against 19.3 with no change. In the separate test where all sources were changed at once, it gave the page ranked fifth the largest gain of the five methods in that test. The page ranked first lost 30.3 percent on average with this method, and in the Perplexity test citing sources came out below no change on the second metric (19.0 against 24.7).

Do this: Reference studies, reports, and recognised experts. Name names, years, and specific findings. A sentence with the name of the study, the year and a link is in our judgement stronger than "research shows that..."

4. Add Direct Quotes

Adding quotations scored highest on both metrics in the main table. The study describes the method as adding quotations from credible sources.

Do this: Interview clients, partners, or industry experts. Add their quotes to your content, with their permission. Or cite recognised sources with direct quotations. Only use quotations that were really said or written.

5. Structure Your Content With Clear Headings

Google names headings as a help for readers: "People generally appreciate it when web pages are organized by paragraphs and sections, along with headings that provide a clear structure to navigate content." Cutting your text into pieces for AI is not needed for Google Search, according to the same guide: "There's no requirement to break your content into tiny pieces for AI to better understand it."

Do this: Use headings that say what is underneath. "Our Process" says little. "How Lead Automation Works in 5 Steps" says what you are about to read.

Corrected on 14 November 2026: this tip said that AI models scan content for structure and that headings help the model understand where information lives. We had no source for that. Google names headings as a help for readers.

6. Make Your Expertise Explicit

In its guide Google puts two kinds of text side by side. A first-hand account gives a perspective of its own, Google writes, "whereas a summary of existing content simply restates information already available elsewhere".

Do this: Our advice, without a source: state on your site who you are and what your experience is, and put an author on your blog posts. Publish examples from your practice with numbers you measured yourself. Only name a client with their permission.

Corrected on 14 November 2026: this tip said that AI models assess whether a source is trustworthy, "partly based on E-E-A-T". That abbreviation does not appear in the documents we read, and how an AI model assesses trustworthiness is not in them either.

7. Be Precise on Your About and Services Pages

According to Google you do not have to put every search variant literally in your text for its search engine: "AI systems can understand synonyms and general meanings of what someone is seeking, in order to connect them with content that might not use the same precise words." For local businesses the guide names the Google Business Profile. According to Google that can help you be visible in AI responses and in the regular Google results.

Do this: Our advice, without a source: write in plain words what you do, for whom, and where. "AI automation for SMBs in the Netherlands" is specific. "Digital solutions" is vague. If you run a local business, keep your Business Profile up to date.

Corrected on 14 November 2026: this tip said that AI models use your business information to determine whether they can recommend you. We had no source for that.

For Google It Is the Same Work

According to Google the rules of thumb of SEO still hold for its AI features: "The best practices for SEO continue to be relevant because our generative AI features on Google Search are rooted in our core Search ranking and quality systems." That is about Google. ChatGPT and Perplexity have their own robots, and you manage those separately.

Corrected on 14 November 2026: the heading above this section was "SEO Isn't Dead, But It's Not Enough". It said that many GEO optimisations also improve your traditional SEO and that better content "scores better everywhere". We had no measurement of that.

What you can do today:

  1. Open your robots.txt and read what it says for OAI-SearchBot, PerplexityBot, Googlebot and Google-Extended. Googlebot covers Google Search including its AI features, Google-Extended covers the Gemini app. If you use a firewall, check there too.
  2. Choose three questions your clients often ask.
  3. Write a page for each question from your own work, with clear headings and with numbers and quotations that are true.
  4. Put an author on it.
  5. Then ask the three questions yourself in ChatGPT, Gemini and Perplexity and write down who gets named. That is one sample. For Google there is also the generative AI report in Search Console.

What We Did on mindsora.io Ourselves

Our own robots.txt names OAI-SearchBot, GPTBot, ChatGPT-User, PerplexityBot, Perplexity-User and Google-Extended one by one and allows them. On 14 November 2026 we requested the address of this page three times, with the name of OAI-SearchBot, of PerplexityBot and of GPTBot in the request, and got the page back three times. That was measured from our own address. Whether a request from the addresses of OpenAI or Perplexity gets the same answer, we cannot tell from that.

Whether ChatGPT, Gemini or Perplexity name us because of it, we have not measured.

One thing on our own side is not in order yet. Our Agent-Ready Grader reports that your robots.txt blocks AI crawlers as soon as one name on its list is refused completely, even when that is only GPTBot or only Google-Extended. According to OpenAI, GPTBot is about training and is independent of search in ChatGPT. According to Google, Google-Extended does not affect your place in Google Search; that name is about training and the Gemini app. The grader does not make that distinction yet.

This post is itself an example too. Until 14 November 2026 it had three figures without a link and a series of sentences about what AI models look for, scan and assess. We put the post next to the documents from OpenAI, Google and Perplexity, next to the study and next to the Pew study. What had no source is gone or is now marked as our own advice. The lines under "Do this" and the list of steps are our own translation into practice. The sources were read on 14 November 2026.

Need Help?

Start with your robots.txt: you can check that yourself. What blocking costs for the robots of OpenAI and Anthropic is in our post on the technical audit. For Perplexity and Google-Extended it is above.

Want us to look along with you? View our services or book a free consultation.

Corrected on 14 November 2026: this section said that most small businesses are not working on GEO yet and that those who start now have "a head start". We had no source for that.

Ready to automate?

Book a free consultation and discover what AI automation can do for your business. Or take the 2-minute AI Readiness Quiz first.

Book a call

More articles