📋 Table of contents
Key points at a glance
- GEO targets mentions, recommendations and source citations in AI responses. A website visit is a separate outcome.
- GEO complements traditional SEO. Ensure your content is easy to understand, clearly structured and technically accessible; however, this does not guarantee inclusion in AI responses.
- Answer specific customer questions concisely first, then go into more detail about conditions, decision options and limitations.
- Support specialist claims with original sources and ensure that professional accountability can be verified.
- Check your visibility using a fixed set of questions for each platform. Individual favourable responses do not constitute reliable proof of success.
What GEO is intended to achieve and why SEO remains important
Generative Engine Optimisation, or GEO for short, refers to the work carried out on content and its discoverability with the aim of becoming visible in generative search results. In doing so, you should distinguish between different types of results: a brand mention simply names your company; a recommendation suggests your offering as a solution to a need. A source citation refers to a piece of content. Only a click leads to a visit to the website.
Traditional SEO focuses on discoverability in search engines and search result lists. GEO also considers whether and how your company appears within a generated response. Use comprehensible content, a clear structure and technical accessibility as common working principles.
This covers ChatGPT with web search, Google AI Overviews and Perplexity. Not every ChatGPT response is a search result: OpenAI describes web search as a function for up-to-date information and source links. Perplexity also documents access to websites for search results and responses, as explained in the technical section.
The following tips are verifiable working approaches, not proven universal GEO ranking factors. Good content and sound technical implementation do not, in themselves, demonstrate a causal influence on AI citations. Nor do providers’ definitions of GEO constitute independent proof of effectiveness. Google expressly states that even meeting its requirements does not guarantee indexing or display.
How do you answer your target audience’s questions specifically?
Answer a specific question directly and then explain the prerequisites, alternatives and exceptions. This allows you to check whether your content actually supports a decision, rather than merely touching on a topic.
Tip 1: Collect real-world questions about information, comparisons and purchasing decisions
Use consultation meetings, sales enquiries and support cases as a starting point. Alongside the question, note down which contact it came from and what decision lies behind it. Internal ‘wish list’ terms are not proof of actual demand.
Hypothetical example: A solar energy company in Lucerne categorises documented questions by decision-making stage. ‘Is my roof generally suitable?’ relates to the orientation stage. ‘How do the proposed systems differ?’ relates to the comparison stage. ‘Are scaffolding, registration and commissioning included in the quote?’ relates to the actual commissioning stage.
These questions require different content. For the orientation stage, an explanation of the assessment criteria is appropriate. For comparing quotes, clear definitions of what is included are needed. A general page about solar energy does not automatically address both.
Your assessment criterion is: Can you link each question to a clear customer need? If this link is missing, treat the question initially as a hypothesis and clarify it in the next appropriate conversation.
Tip 2: Put the short answer before the detailed explanation
Don’t start with a general description of the industry. The first paragraph should answer the question without omitting any prerequisites.
Hypothetical example for the question ‘How is a quote drawn up for flat cleaning in Basel?’:
‘We’ll draw up a bespoke quote for your flat cleaning. The price depends on the floor area, the condition of the flat and the scope of services required. We’ll itemise window cleaning and any satisfaction guarantee separately. The quote will show the total price in CHF and the VAT status.’
This is followed by details on additional work, site visits and exclusions. Without a reliable calculation, this introduction makes more sense than a made-up average price.
Check the reply paragraph in isolation from the rest of the text: are the services, terms and any outstanding points still clear? Brevity must not result in a personalised quote appearing to be a standard, fixed price.
How do you make content transparent and verifiable from a technical perspective?
Tip 3: Back up verifiable statements with original sources
Link figures, legal statements and technical claims directly to the relevant original source. What matters is not the number of links, but whether a reference precisely supports the statement next to it.
Hypothetical example: A trust office in Bern states that the standard Swiss VAT rate is 8.1 per cent and provides a direct link to the Federal Tax Administration’s tax rate overview. To determine the tax classification of a specific service, it would also need to examine the relevant technical basis. The general tax rate alone does not answer this specific question.
In this case, the FTA confirms the tax rate, not the quality of the trust office. This distinction is part of source research. Check the publisher, scope and currency of the information. A source link is no substitute for expert verification or accurate reproduction.
Tip 4: Structure content according to questions and decision-making criteria
Subheadings should make it clear which question a section answers. ‘What services does the quotation cover?’ is more specific than ‘Our services’. The answer and any limitations should be grouped together in the paragraph below.
Hypothetical example: A joinery firm in Aargau explains material options for a built-in wardrobe. It could compare the variants actually on offer using the same criteria: finish, care instructions, repair options and intended use. A table is suitable if reliable information is available for each variant. Generalised assessments such as ‘high-quality’, on the other hand, require an explanation of what the judgement is based on.
Your test criterion: Are the differences understandable without further interpretation? If a choice of material depends heavily on design or intended use, an explanatory paragraph is better than a seemingly unambiguous ranking.
Do not create a separate page for every similar phrase. Instead, check whether several questions relate to the same decision and can be answered together.
Tip 5: Indicate who is technically responsible
Name the person or specialist department responsible, their relevant qualifications and a means of contact. Describe qualifications in sufficient detail so that their relevance to the topic remains clear. A marketing role alone, for example, does not in itself imply responsibility for technical installation instructions.
An update date should reflect the actual status of the content. Document internally which statements, sources or services have been checked. A newly set date does not replace a substantive revision.
Also, ensure that the company name, location, scope of services and contact details are consistent across the website and managed profiles. In the case of multilingual content, any conflicting areas of responsibility or service descriptions should be clarified.
The key criterion is verifiable responsibility: can you determine who is responsible for the content and why that person is the right fit in terms of expertise? An author profile is a starting point for this, not a substitute for expertise.
Technical accessibility and structured data: what needs to be checked?
The platforms document different access methods. The following overview describes their own technical specifications; it does not constitute independent proof of effectiveness.
| Platform | Documented basis | Specific checks |
|---|---|---|
| Google AI Overviews | A linked page must be indexed and approved for a search snippet. Google also documents participation via the generative AI setting in Search Console. | Check the indexing status, snippet rules, Googlebot access and the relevant Search Console setting. Sources: Google documentation, Search Console Help |
| ChatGPT with web search | OpenAI uses OAI-SearchBot for searching. Authorisation is independent of GPTBot for model training. | Check that the required search access is permitted in robots.txt and that access is granted via the published IP ranges. Source: OpenAI crawler documentation |
| Perplexity | PerplexityBot is used to link to websites in search results. Perplexity users support user-initiated page requests. | For PerplexityBot, check the access permissions in robots.txt and the firewall rules. For Perplexity users, check accessibility and firewall permissions based on the documented IP ranges. Source: Perplexity crawler documentation |
Ask your technical support team to check not only the configuration but also actual accesses in the server logs. An intended authorisation and a successful retrieval are different points to check.
Structured data should accurately describe the visible content. Google does not require any specific Schema.org markup or an additional llms.txt file to appear in its AI search functions. Therefore, do not treat llms.txt as a blanket requirement or a substitute for accessible content. Technical accessibility is a prerequisite for access in each case; it is not synonymous with inclusion in search results.
Tip 6: Build a relevant professional presence outside your website
Treat external expert presence as verifiable communication work, not as a collection of as many mentions of your name as possible. Suitable venues may include specialist media, relevant association profiles or editorially managed industry publications.
Hypothetical example: A heating company in the canton of St. Gallen wishes to offer an article to help prepare for a boiler replacement. First, it narrows the topic down to specific decision-making questions. It then checks whether a regional trade publication reaches this target audience and exercises editorial control over specialist articles. The text provided explains the prerequisites and conflicting objectives, rather than simply promoting the company’s own services.
Key selection criteria include suitability for the target audience, editorial quality and technical value. An association profile requires accurate company details; a specialist article requires an independent perspective and expert oversight. Linked service information should complement the article, not overshadow its actual purpose.
Your assessment criterion is: Is the publication factually accurate, accessible and relevant to potential customers, even independently of AI search?
Editorial mentions are not entirely within your control. Neither paid placement nor a published specialist article guarantees recommendations or citations by AI systems. You should therefore base your decision on the communication benefits, not on promised AI rankings.
Tip 7: Measure your AI visibility with repeatable tests
Define a fixed set of questions and comparable test conditions
Use the documented customer queries as a fixed test set. Distinguish between information-seeking queries and provider-related queries: ‘How do I evaluate an offer?’ examines something different from ‘Which company is right for my project?’.
Test the same phrasing repeatedly on each platform. Document the date, language, Swiss location context, visible model or search mode, as well as login details and relevant personalisation. Where possible, start with a new conversation and record the response along with its sources.
With ChatGPT, it must be clear whether a web search was used. With Google, ‘no AI Overview displayed’ should also be included in the log: Google does not display AI Overviews for every search query. Evaluate German-, French- and Italian-language services separately.
Separate mentions, source citations and business results
Hypothetical example: A Swiss cleaning company is testing its documented quotation enquiries. The following template records observations separately. The date, language and test context form part of the metadata for each run.
| Query | Platform and search mode | Brand mention | Recommendation | Source link | Linked landing page |
|---|---|---|---|---|---|
| Wording of the test question | Mode actually used | Yes / no | Yes / no, with text excerpt | Specific URL or ‘none’ | Own page, external page or ‘none’ |
A mere mention of a name does not automatically count as a recommendation. A source link to a specialist article, on the other hand, is not the same as a recommendation from your organisation.
Record attributable website visits and enquiries separately. A source citation without a visit and a visit without an enquiry remain distinct results. Your test criterion is consistent documentation across multiple runs.
Make decisions based on patterns rather than promises of success
If a relevant question is repeatedly answered with external sources, but your relevant page never appears, this warrants a closer examination of content, accessibility and the focus of the question. This does not, however, constitute a diagnosis of a specific GEO error.
Record changes to your pages separately from the tests. Changing answers, platform updates, new competitor content and varying test conditions may also provide an explanation.
A single additional mention is therefore insufficient to prove an effect. Use recurring patterns to prioritise specific checks and next steps, not to set fixed deadlines for success.
Summary
With GEO, your aim is to ensure your content appears in AI responses: you answer relevant questions specifically, make your statements verifiable and ensure your website is technically accessible. Use SEO as a foundation for this, whilst GEO additionally takes into account how response systems pick up on content and incorporate it as sources.
It is crucial that your content matches the actual question and provides a comprehensible, technically sound answer. Clear structures, identifiable responsibilities and robust evidence facilitate classification. A relevant professional presence outside your website complements this work, but does not replace compelling content on your own pages.
You can assess whether your measures are moving in the desired direction by conducting repeatable tests in ChatGPT, Google AI Overviews and Perplexity. Individual mentions are less meaningful than patterns observed over several runs; changes in answers can have various causes.
Start with your target audience’s most important questions and assess the relevant pages for the quality of their answers, clarity and accessibility. In this way, you can derive concrete improvements from your observations, rather than basing your next steps on individual AI responses.
Frequently Asked Questions (FAQs)
What is Generative Engine Optimisation (GEO)?
Generative Engine Optimisation (GEO) refers to optimising your content for visibility in AI-generated answers, such as in ChatGPT, Google AI Overviews and Perplexity. The aim is to secure relevant brand mentions, recommendations or source citations. To achieve this, you present information in a way that is clear, comprehensible and tailored to the questions asked by your target audience.
How does GEO differ from SEO?
SEO focuses on making your website visible in traditional search results, whilst GEO aims to ensure your content is used in AI-generated responses. For both approaches, use clear content, a logical structure and technical accessibility as the foundations of your work. When planning your strategy, GEO therefore complements search engine optimisation rather than replacing it entirely.
How do you prepare your content for AI searches?
You prepare your content for AI searches by answering specific questions directly and structuring verifiable statements in a clear and understandable way. To do this, gather questions from your customer support interactions, formulate short answers and include original sources, appropriate subheadings and details of professional responsibility. Then check whether each answer remains understandable even without the surrounding text.
How do you measure your visibility in ChatGPT, Google AI Overviews and Perplexity?
You measure your AI visibility by repeatedly testing a fixed set of questions on each platform and documenting the results. Record the date, language, location context, search mode, brand mention and linked landing page. Distinguish between a mention, a recommendation, a source citation and a website visit. Test German- and French-language questions separately for your Swiss offering.
Can you guarantee a mention in AI responses with GEO?
No, you cannot guarantee a specific mention, recommendation or source link in AI responses using GEO. You should therefore focus your efforts on verifiable improvements: precise answers, substantiated specialist information, consistent company details and regular visibility tests. Assess your progress based on repeated observations and relevant website visits or enquiries, not on the basis of a single test response.
