Get cited by ChatGPT: how your company gets named in AI answers

Short answer: to get cited by ChatGPT, Google AI Overviews or Perplexity, your website has to be open to their crawlers, answer customer questions directly on clearly structured pages, and be confirmed by other sources such as reviews, directories and press. That matters because ChatGPT alone had 900 million weekly users in February 2026, according to OpenAI. There is no secret file or trick: Google states that AI Overviews and AI Mode need nothing beyond pages that are indexed, eligible for a snippet and genuinely useful.

Generative Engine Optimisation, or GEO, is the work of getting named as a source in AI answers. It builds on SEO rather than replacing it, and a lot of what is sold under the label is either plain SEO or wishful thinking. This guide separates the two.

In the GEO checks we run for Austrian companies, the pattern is almost always the same: when someone asks ChatGPT for “the best real estate agent in Vienna” or “a web design agency for SMEs in Linz”, the answer ends with a short list of companies. The ones named are rarely the biggest. They are the ones whose websites say clearly what they do, where, and for whom, and whom other sites mention as well. If you want to see where you stand first, the free GEO report shows real ChatGPT answers for your industry in about a minute.

How AI answers pick their sources

There are two ways an AI model “knows” about you. The first is training data: what the model learned before its cut-off date. You cannot influence that quickly. The second is retrieval: the model searches the web while answering and cites what it finds. This is where GEO works, and it is where almost all local and commercial questions are answered today.

• ChatGPT runs a web search for questions that need current or local information, reads a handful of pages and lists them as sources. Its search index is built by OpenAI’s crawler OAI-SearchBot.
• Google AI Overviews and AI Mode use Google’s normal index. AI Mode splits a question into many related sub-queries (Google calls this “query fan-out”) and combines the results, so a page can be cited for a question it never ranked for directly.
• Perplexity searches the web for every question and shows its sources prominently, using its own crawler PerplexityBot.

What all of them have in common: they prefer passages that answer the question in one or two sentences, from sources that seem reliable and are confirmed elsewhere. A vague “We offer holistic solutions” page gives them nothing to quote.

ChatGPT, Google and Perplexity compared

Crawler access: check your robots.txt first

The most common reason a site is never cited is boring: the crawler is not allowed in. We see it regularly in our GEO checks, usually because a security plugin, a hosting default or a well-meant “block all AI” rule shuts out the search crawlers along with the training crawlers. Cloudflare, for example, has blocked AI crawlers by default for new domains since July 2025.

OpenAI separates its bots, and you can treat them separately:

• OAI-SearchBot builds the index for ChatGPT search. Block it and you will not appear as a source.
• GPTBot collects training data. You can block it without losing ChatGPT search visibility.
• ChatGPT-User fetches a page when a user asks ChatGPT to open it. OpenAI notes that robots.txt rules may not apply to these user-initiated visits.
• Google-Extended only controls whether your content is used for Gemini training. It does not affect Google Search or AI Overviews.
• PerplexityBot indexes for Perplexity’s answers.

Our recommendation for most companies that want to be found: allow all search crawlers, decide about training crawlers deliberately. And make sure the important content is in the HTML itself, not loaded by JavaScript afterwards. An analysis by Vercel and MERJ found that the major AI crawlers, including OpenAI’s and Perplexity’s, do not execute JavaScript. That is one reason we build sites on Next.js with server-rendered pages; more in our article on moving from WordPress to Next.js.

Answer-first content that gets quoted

AI answers are built from passages. The easier a passage is to lift out and quote, the more likely it is used. In practice that means:

• One page per service and location. “Web design Vienna” and “Website funding Austria” each deserve their own page, not a paragraph on a catch-all services page.
• The answer first. Start each page and each section with a direct one- or two-sentence answer, then explain.
• Specific facts. Prices or price ranges, timelines, areas served, numbers of projects, certifications. Models quote what is concrete.
• Questions as headings. Phrase H2s the way customers ask: “What does a website cost in Austria?”
• Sources and figures. Cite where your numbers come from.

There is research behind the last point. The GEO study by Princeton and other researchers (KDD 2024) tested content changes on a simulated AI search engine: adding statistics raised visibility in answers by around 41 %, adding quotations by around 28 %. These are lab results on a test setup, not a guarantee for your site, but they match what we see in practice: concrete pages get quoted, generic ones do not.

Google adds a useful warning: there is no need to chop content into tiny pieces or rewrite it “for AI”. Write for people who want an answer. That is also what the models want.

Structured data: helpful, not magic

Schema.org markup tells machines in a fixed format who you are: company name, address, opening hours, services, reviews, authors. For an AI model trying to work out whether “404media in Vienna” and “404media GmbH” are the same company, consistent structured data removes guesswork.

Honesty matters here too. Google says structured data is not required for its generative AI features and that there is no special schema for them. It still recommends structured data as part of normal SEO, for rich results. We add Organization or LocalBusiness, Service, Article, Breadcrumb and, where it fits, FAQ markup to every site, because it costs little and keeps your company facts consistent everywhere. Do not expect a ranking jump from schema alone.

llms.txt: what it can and cannot do

llms.txt is a proposal from September 2024: a Markdown file in the root of your site that lists your most important pages for language models, similar to a sitemap with short descriptions. Many GEO offers present it as the key step.

The facts: Google states that its Search ignores such files and that they neither help nor harm visibility in AI Overviews or AI Mode. OpenAI, Anthropic and Perplexity have not confirmed that their search systems use llms.txt for ranking or citations. Log analyses show occasional fetches, but no proven effect.

Our approach: we generate llms.txt automatically from the same content as the website, because it takes minutes and may become useful for AI agents. We do not sell it as a visibility lever, and you should be wary of anyone who does.

Reviews and brand mentions: the signals AI trusts

Your website says what you claim. Other sites confirm whether it is true. AI answers lean heavily on that confirmation, especially for local recommendations. The signals that make the difference:

• Google reviews with text, not just stars, and replies from you. Reviews that mention the service and the place (“great website for our real estate agency in Vienna”) are particularly useful.
• Directories and chambers: WKO Firmen A–Z, Herold, industry directories, with exactly the same name, address and phone number everywhere.
• Trade media and local press: an interview, a guest article, a project report.
• Lists and comparisons: “best agencies in Vienna” articles are often exactly what ChatGPT reads when it builds its own list.
• Your own case studies with real client names and results.

This is the slow part. Structural changes to a website usually take effect within a few weeks, once the models crawl again. Mentions on external sources build up over two to three months. For our own company, the 5.0 Google rating from 23 reviews and the case studies on our site are the foundation we build on.

How to measure your AI visibility

For a long time GEO was impossible to measure. That has changed in 2026:

• Google Search Console has a generative AI performance report, available for all sites worldwide since 31 August 2026. It shows impressions in AI Overviews and AI Mode by page, country and device, but no clicks and no queries yet.
• Bing Webmaster Tools has shown an AI Performance report since February 2026, with citations in Copilot and Bing’s AI summaries and the queries behind them. It does not cover ChatGPT, Perplexity or Google.
• Analytics: ChatGPT adds utm_source=chatgpt.com to many links it shows, so referrals appear as their own source. Perplexity and others show up as referrers.
• Prompt tests: ask the questions your customers ask, in several models, regularly. Answers vary with wording, location and time, so one test proves little and a series shows the trend.

Our free GEO report automates the last step: it asks ChatGPT live with web search, location Austria, shows whether you are named and who is recommended instead, and checks the technical basics of your site.

Getting started in the right order

1. Test the status quo: ask ChatGPT, Perplexity and Google AI Mode five to ten real customer questions, or run the GEO report.
2. Open the doors: check robots.txt, firewall and CDN settings for OAI-SearchBot, PerplexityBot and Googlebot.
3. Fix the pages: one page per service and location, answer first, concrete facts, clean headings, server-rendered HTML.
4. Add structured data: Organization or LocalBusiness, Service, Article, consistent with your Google Business Profile.
5. Build confirmation: ask happy clients for Google reviews, update directory entries, pitch one trade article per quarter.
6. Measure monthly: Search Console, Bing Webmaster Tools, referrals and a fixed set of prompts.

When GEO will not help (yet)

GEO is not the right first step for everyone:

• Your site is not indexed or barely ranks. Google’s AI features draw on its index. Fix SEO basics first.
• Nobody asks about your offer. For very narrow B2B niches with a handful of buyers, direct sales beat AI visibility.
• You have no reviews and no mentions. Then the work starts outside the website, and that takes months.
• You need enquiries next week. GEO takes weeks to months. For quick demand, ads in Google or in ChatGPT are faster; see our guide to ChatGPT ads in Austria.

And a warning: nobody can guarantee you “position 1 in ChatGPT”. Answers differ from user to user and day to day. What you can do is make it as easy as possible for the models to find, understand and trust you, and measure whether it works.

Written byChristoph SpeiserProject lead & sales

Meet the team →