GEO agency · Generative Engine Optimization
Generative engine optimization (GEO) makes AI assistants such as ChatGPT, Gemini, Claude and Perplexity name your company when a customer asks for a provider, and describe it correctly. Hierarchy is a GEO agency in Zurich. We work on the three things an AI recommendation is made of: your website, the third-party sources the assistant reads, and what the models already know about you. It is grounded in our own study of 1,515 AI answers about 5,110 Zurich companies. Results are measured against a fixed set of customer questions, run several times, because AI answers fluctuate. Packages from CHF 900 per month, SEO and AEO included.
Starting point
Because an AI assistant chooses differently from Google. In our study, around three quarters of the recommended firms were not in Google's top 10 for the same question. Ahrefs sees a similar picture: only 12% of the pages cited by ChatGPT, Gemini, Copilot and Perplexity rank in Google's top ten. Good rankings don't guarantee a recommendation, and weak rankings don't rule one out. Usually one of three building blocks is missing: your website doesn't say in its opening paragraphs what you do, for whom and from what price. You're not on the lists and directories the assistant reads. Or the models simply don't know your name, because hardly anyone writes about you outside your own website. And because nobody measures it, nobody notices.
SEO, AEO, GEO
The market mixes these terms freely. We separate them by who answers the customer. All three share the same foundations, which is why every discipline is included in every one of our packages.
| Discipline | Who answers | Where you appear | What we measure |
|---|---|---|---|
| SEOSEO agency | Google's result list | In the organic results, as high as possible | Rankings, organic clicks, enquiries |
| AEOAEO agency | Google itself | In AI Overviews, AI Mode, featured snippets and on Google Maps | Presence in Google's answers, clicks, calls, enquiries |
| GEOThis page | ChatGPT, Gemini, Claude, Perplexity | In the assistant's recommendation, described correctly | Share of mentions across a fixed prompt set, run several times |
What we do
Customers don't ask an assistant "accountant Zurich"; they ask "Which accounting firm in Zurich works with start-ups and speaks English?". We build a set of 20 to 40 such questions from your sales calls, enquiries and Search Console data, by audience and buying stage. We also check which searches the assistant runs in the background before it answers. These sub-queries usually have no measurable search volume and appear in no keyword tool, yet they decide which pages it reads.
Assistants lift sentences that stand on their own: who, what, for whom, where, from what price, with what result. What isn't stated explicitly on your website doesn't exist for the model; a service that only appears in an image or a PDF simply won't be named. We write your homepage, service pages and FAQ so that every key claim is citable in one paragraph, with numbers and evidence. Research backs this up: in the GEO study by Princeton and IIT Delhi, adding statistics, quotations and sources raised visibility in AI answers by roughly 30 to 40%, while keyword stuffing made it worse.
For "who is the best …" questions, assistants draw on best-of lists, directories, associations and trade media. Ahrefs found in ChatGPT answers to 750 prompts that "best X" lists made up 43.8% of cited page types, and 35% of them sat on low-authority domains. We find out which sources are actually cited in your industry and get you into them: listings, corrections, reviews, editorial mentions. We don't use bought self-promotional lists; they work briefly and hurt in the long run.
40% of the mentions in our study have no retrievable source: the model knows the firm from training. You don't get into that memory through your own website, but through mentions elsewhere. Ahrefs' large analysis of 75,000 brands shows the same pattern: brand mentions on the web and on YouTube correlate far more strongly with AI visibility than backlinks do. So we plan digital PR, expert articles, interviews and data that others report on.
For German questions, assistants sometimes run their own sub-queries in English, and research from the University of Toronto shows that cited sources shift markedly by language. For Swiss companies this means an English homepage, an English company description and listings in international directories belong in the plan, even if you only sell in German-speaking Switzerland. On request we also test your prompt set in French and Italian.
SparkToro asked the same questions 2,961 times: the chance of getting the same list of brands twice was under 1 in 100. So there is no such thing as "position 1 in ChatGPT", and anyone promising it is measuring the wrong thing. We ask every question several times, to ChatGPT, Gemini, Claude and Perplexity, and report your share of mentions: for how many questions and runs you're named, how you're described and which sources were cited instead. Plus an "AI assistant" channel in Google Analytics 4 and a "How did you find us?" field on your contact form.
How it runs
We run your prompt set several times across all relevant assistants and record whether you're named, how you're described, and which competitors and sources appear instead. From that we read which of the three levers matters in your industry: website, third-party sources or memory. Result: a prioritised list of the first ten actions.
Homepage, service pages, FAQ, about page, structured data and an English foundation. This works fastest: assistants with web search re-read your page on the next retrieval, regardless of how old the domain is.
Listings on the lists and directories cited in your industry, reviews above the threshold, expert articles and digital PR. It takes longer, but it's the part competitors find hardest to copy.
The same prompt set, the same method, one report alongside rankings and enquiries. What raises your share of mentions we expand; what hasn't moved anything after three months we stop.
Data basis
For our AI visibility study we asked ChatGPT, Gemini and Claude 1,515 purchase-intent questions across 12 Zurich industries and traced each of the 14,135 mentions back to its source. 40% come from the models' memory, 39% from the named firm's own website, 18% from third-party pages. Firms listed on a page the assistant reads are named 8.2 times as often. Pages with an FAQ section are cited 1.6 times as often, pages that mention Zurich 1.35 times. Reviews help up to about 30, then barely; the star rating itself hardly separates named from unnamed firms. The levers differ completely by industry, so we start with yours.
Packages
GEO, AEO and SEO share the same foundation, so we don't sell them separately: every package includes all three. Prices are starting points; what makes sense in your case we settle after the baseline. If you'd rather implement yourself, choose done with you.
from CHF 900per month
For SMEs building a solid foundation and predictable visibility.
from CHF 1,500per month
For companies claiming whole topic areas and appearing in AI answers.
customon request
Fully done for you: we take care of everything and keep optimising; you only approve.
customon request
For marketing teams that want to run AI visibility themselves: strategy and know-how from us, implementation on your side.
All prices in CHF, excl. VAT, excluding external placement or media costs. What makes sense in your case is settled before the first franc.
What changes
AI assistants name your company for the questions your customers ask, with a description taken from your own website.
You know for what share of 20 to 40 customer questions you're named, compared with last month and with your competitors.
Wrong or outdated information about your company in AI answers is found and corrected at the source.
A foundation that works for Google too: the same pages and mentions carry SEO and AEO.
Enquiries as the yardstick: mentions are the means, new customers the goal.
In practice
Questions
Anything else? Write to info@hierarchy.ch
A GEO agency makes AI assistants such as ChatGPT, Gemini, Claude and Perplexity name a company and describe it correctly when customers ask for a provider. GEO stands for generative engine optimization. The work covers three areas: making the company's own website citable, building mentions in the third-party sources assistants read, and strengthening the brand so the models know it even without web search. Hierarchy is a GEO agency in Zurich and measures success against a fixed set of customer questions.
SEO moves a page up Google's result list. GEO gets a company into an AI assistant's answer. The foundations overlap, the selection doesn't: in our study around three quarters of the firms recommended by AI were not in Google's top 10, and according to Ahrefs only 12% of the pages AI assistants cite rank in Google's top ten. GEO also depends on third-party sources, brand mentions and citable wording. The full comparison is in our article "SEO agency or GEO agency?".
We separate them by who answers. With GEO it's other companies' assistants: ChatGPT, Gemini, Claude, Perplexity. With AEO (answer engine optimization) it's Google itself, with AI Overviews, AI Mode, featured snippets and Google Maps. The levers differ: in our analysis of Google AI Mode, 76 of 100 answers relied on Google Maps data, while for ChatGPT list pages and brand mentions play a bigger role. The market doesn't use the terms consistently; some agencies mean the same thing by AEO and GEO.
For provider questions, ChatGPT with web search first looks for lists and directories, then reads the websites of the firms it found and builds its recommendation from them. Three things decide: whether you appear in the sources it reads (in our study 58% of mentions don't come from the named firm's own website); whether your website says clearly in its opening paragraphs what you do, for whom and from what price, because it lifts exactly those sentences; and whether your reviews and mentions are above the thresholds at which a model treats a firm as established. Without web search, only what the model knows from training counts.
From our 1,515 analysed answers: first, the sources. Firms on the list page the model retrieves are named 8.2 times as often; whether they're the stronger firm on that list only makes a factor of 1.3. Second, how readable the firm's own website is: concrete services, prices, location, FAQ. Third, thresholds: the likelihood of being named rises up to about 30 Google reviews and about 99 referring domains, then flattens. Depending on the industry, between a tenth and over half of mentions come from the models' memory.
No, and be careful with anyone who does. AI answers aren't stable: in a SparkToro experiment with 2,961 runs of the same questions, the chance of getting the same list of brands twice was under 1 in 100. What can be reliably increased is the share of questions and runs in which you're named. For our solar client it rose from 19 to 63 of 100 questions in six months; for hierarchy.ch it went from 0 to 54 of 145 since the domain launched in August 2026.
With assistants that use web search, often within weeks. ChatGPT, Gemini and Perplexity read a page at retrieval time and judge it by content, not domain age: once a page is indexed and says concretely what you do and for whom, it can be cited. In our own test in September 2026, ChatGPT named hierarchy.ch for 11 of 14 questions, while the same pages were only at positions 50 to 90 on Google. Mentions from the models' memory take longer, because they only appear with new training data.
Often, yes. Good rankings help, because assistants with web search partly draw on search indexes, but they're not enough. In our study, named firms had better rankings in many industries, yet three quarters of recommendations went to firms outside Google's top 10. Whether you have a GEO problem takes minutes to check: ask ChatGPT, Gemini and Claude three times whom they recommend for your service in your region. Our free AI visibility check does it systematically.
Find the source first. In answers with web search, the assistant usually shows which pages it read: an outdated directory listing, an old price list, an article about a company with a similar name. That source gets corrected or overtaken by a clearer page of your own. If the claim comes from the model's memory, only consistency helps: the same correct facts on your website, Google Business Profile, LinkedIn, directories and Wikidata where it exists. That takes effect with the next training cycle.
Less than is often claimed. There's no evidence so far that ChatGPT, Gemini or Claude use an llms.txt file for recommendations; we set one up because it costs nothing, but we don't expect it to move anything. Structured data (Organization, Service, FAQ) helps search engines read facts such as address, services and prices unambiguously, which indirectly benefits assistants with web search. It's not what tips the balance; content, third-party sources and mentions are.
It depends on the industry. For software, consumer goods and international topics, Reddit and YouTube are important sources; Ahrefs' analysis of 75,000 brands found the strongest correlation with AI visibility for YouTube mentions. For a Zurich accounting firm or dental practice, directories, associations, reviews and the firm's own website matter far more. We see which sources are cited in your industry in the baseline, before putting effort anywhere.
With a fixed set of 20 to 40 customer questions that we ask ChatGPT, Gemini, Claude and Perplexity several times every month. For each answer we record: named or not, how you're described, where in the answer, and which sources were cited. That gives your share of mentions, compared with last month and with competitors. We also measure visits from AI assistants in Google Analytics 4 and ask on your contact form how customers found you, because many simply type the name from the answer.
Yes. For the prompt set we ask public AI assistants questions, without any personal data of your customers. For visit tracking we use your existing Google Analytics 4 with your consent solution; AI assistants are just grouped as their own channel. No new data flows arise beyond those your privacy policy already covers.
Hierarchy in Zurich has GEO as one of three focus areas, alongside SEO and paid ads, and works with SMEs across Switzerland. It's grounded in our own AI visibility study with 1,515 AI answers and 14,135 traced mentions across 12 Zurich industries. We do the work or enable your marketing team in the done-with-you model. Our office is at Strehlgasse 2, 8001 Zurich. A comparison of several providers is in our article on the best SEO, GEO and AEO agencies in Zurich.
Yes, in the done-with-you model. We provide the industry analysis, prompt set, reporting and playbooks and train your team in workshops; your team implements, we review drafts and are available for regular sparring. It suits marketing teams of about two to three people or more with access to the website and CMS. The goal is for your team to run the measurement itself after a few months.
Free, no obligation
Our AI visibility check measures whether ChatGPT, Gemini and Claude name your company today and which competitors they recommend instead. You get the result before we talk about a package.
About the author