How can I improve my AI visibility?

1. What AI visibility actually means

A customer doesn't necessarily ask for your company name. He describes a problem: “I would like to install a solar installation and a heat pump in Aarau. Who will take care of both from one source?" You are visible if the answer mentions your company in an appropriate context, recommends it or uses your website as a source.

These three outcomes are different. A provider can be recommended without a link to its own website. Conversely, an AI may cite your guide and recommend another provider. A name can even appear in a critical statement. A single combined score hides these differences.

SEO, GEO and AEO: What's the difference?

SEO refers to optimization for search engines. GEO and AEO are used for visibility in generated responses. The terms do not describe a uniform, certified method. For its generative search features, Google emphasizes that basic SEO work is still relevant. This is a statement about Google, not a complete description of the rankings of all AI providers. [1]

Some answers are based on web searches, others on existing model knowledge or other data sources. You control your publications, but not the selection decision. Therefore, you need to consider ChatGPT, Google's AI search, and other systems separately.

2. What numbers show - and what they don't show

AI search deserves systematic attention. It does not justify invented promises of reach. For usage figures, see our separate regional overview. Here, we focus on what they mean for improving your visibility.

Visibility is not the same as website traffic

Pew evaluated the browsing of 900 adults in the United States. Search activity dates back to March 2025; Results were collected in April. On AI summary pages, a link within the summary was clicked in 1% of visits. The observed 8% and 15% are not experimentally proven effects of AI and cannot be transferred to every Swiss industry. [2]

Our practical conclusion: Don't judge your work based on traffic alone. Also check the accuracy of your company representation and the quality of incoming inquiries. At the same time, you should not replace clicks and orders with an “AI reach value” that is difficult to verify.

Why “up to 40% more visibility” is not a promise

The GEO research paper, accepted for KDD 2024, reports improvements of up to 40% in its studied setup. This affects the visibility metrics of generated responses defined there. It is not proof of 40% more customers, sales or current ChatGPT recommendations. The authors examine, among other things, references, statistics and quotations; Effects vary depending on the topic. [3]

Use such research as an opportunity for clean testing. Complete a number if it answers a customer question and is supported. Don't add random statistics just because you expect it to have a ranking effect.

3. First, find out why you are missing

Lack of visibility can have several causes: the question does not match the offer, the region is wrong, your information is not accessible, important services are missing from the website or other sources are more helpful for the system. None of these causes can be determined with certainty from a single answer.

Establish a small, usable baseline

1. Collect real phrases from sales conversations, search queries and support. Remove personal customer information.
2. Cover different needs: the task, location, comparison, costs, suitability and a deliberate question about the brand.
3. Test open-ended questions without your company name separately from questions with brand mention.
4. Document system, date, language, location preference, web search and conversation context. Save the complete answers including sources.
5. Repeat selected questions. A practical starting size, for example, is 10-20 different questions with several repetitions. This is a working convention, not a statistically proven minimum number.

Translate observations into verifiable tasks

4. Make sure important content is accessible

Before you rewrite texts, have your five most important public pages checked. Ask your web agency for specific URLs and findings instead of a blanket “SEO is green”. A test in your own logged in browser does not prove that a crawler will receive the same content.

• Accessibility: Does the site provide a successful status and real content - even without registration?
• Indexing: Is there an accidental noindex, an incorrect canonical URL or a redirect to an inappropriate page?
• Internal links: Is the service page accessible via navigation or appropriate overview pages?
• Display: Is the central information available as text, even if a large image or video fails?
• Protection systems: Are legitimate search crawlers blocked by a firewall, bot protection or a challenge? Make targeted exceptions only for verified crawlers.
• Up to date: Do the sitemap, links and public prices match the actual offer?

Decide search and training separately

OpenAI distinguishes OAI-SearchBot for searching from GPTBot for potential model training. The settings are independent. To be included in ChatGPT search results, OAI-SearchBot should not be blocked; however, this does not force placement. The technical documentation also provides IP information to verify legitimate requests. [4]

Give your agency a clear brief: “Keep public service pages accessible to the intended search services; keep the admin area, customer portal and confidential documents protected.” A robots.txt controls crawler access, but is not access protection. You need real authentication for confidential content. Google also explains that a blocked URL may be indexed without crawled content. [5]

How to verify the work

Use URL checking in Google Search Console and the corresponding checks in Bing Webmaster Tools. View rendered content, indexing status, and prominent server responses. A successful live test does not automatically mean that the page already appears in the index or in an AI response.

The current settings for inclusion in generative search functions are also relevant for Google. The specific documentation and interface may change; check them directly in the linked official guide. [1]

5. Build pages that help customers make a decision

“Innovative solutions from a single provider” says little about whether a business can solve a specific problem. A useful service page explains the problem, audience, service, area covered, limitations and next steps. This is not a secret AI format: it helps people assess an offering.

The building blocks of a strong service page

1. A direct answer: Explain in a few sentences what task you are taking on and for whom.
2. Scope of services: Separate included services, options and work of other parties involved.
3. Suitability and limitations: Say when your offer fits and when it doesn't. This saves inappropriate inquiries.
4. Process: Describe the enquiry, assessment, quote, implementation, handover and support.
5. Price logic: If a fixed price is not possible, explain what drives the cost. Only give price ranges with clear assumptions, a date and a factual basis.
6. Evidence: Show a real project, a suitable certificate or documented results.
7. Customer questions: Answer the actually recurring questions precisely. Avoid ten variations of the same question.
8. Next step: Explain what information you need for an initial assessment and what happens afterwards.

Which contents first?

Choose topics based on customer relevance and economic importance. For an installer, “renovation in an inhabited house”, “coordination of several trades” or “maintenance after installation” can be more valuable than the hundredth general solar guide. For a B2B service provider, integrations, responsibilities, handovers and procurement criteria are often crucial.

Google recommends helpful, reliable content with recognizable experience and traceable authorship. [6] Our editorial recommendation: Have the person with practical experience check the draft professionally. Add details that only your company can provide: typical stumbling blocks, decision-making questions or an honest project evaluation. AI can help with structuring; Responsibility for facts and promises remains with you.

Compare without declaring yourself the winner

A good comparison site explains criteria and conflicting goals. Example: One provider coordinates everything, another works with partners; different models can suit different customers. Disclose that you are a provider yourself. Back up concrete statements about competitors and keep them up to date. A made-up “top 10 list” with you in first place is not a reliable sign of trust.

6. Make your experience verifiable

A source is more convincing when readers can check where the statement comes from. This is a quality principle, not proof that every certificate brings a fixed AI ranking bonus. Start with evidence to help your customers make their choices.

Project example instead of a general success story

Document the initial situation, requirements, chosen solution and result. Separate measurement from estimation. When you mention an energy saving, it includes the time period, comparison basis and relevant changes. If no reliable numbers exist, describe qualitative results and their basis. Only publish photos, customer names or quotes with the necessary rights and approvals.

A central overview of company data

Maintain company name, trade name, address, public telephone number, domain, contact route, services and regions served in an internal reference list. Use this list when updating profiles. Document previous names and moves so that contradictory information can be corrected. If there are several locations, it should be clear what is actually offered at which location.

Improve external sources in a targeted manner

Check trade associations, manufacturers’ partner directories, reputable industry portals and relevant local media. Aim for accurate, useful information. Ask for incorrect entries to be corrected; do not invent memberships, awards or editorial recommendations. Paid listings must be treated as paid listings, not as independent evidence.

Ask real customers neutrally for honest feedback. Do not buy reviews or write supposed customer experiences yourself. For Google, the official local documentation lists, among other things, awareness, links and reviews as relevant factors. This does not result in an identical mechanism for ChatGPT. [7]

7. Region, language and target group must match

A company from Aarau that only installs in Aargau needs different content than an international software provider. Don’t write for “the AI of the world,” but rather for the markets where you can actually sell or deliver.

For local service providers

Maintain an authorised Google Business Profile with a suitable category, correct opening hours and service information. Google describes relevance, distance and prominence as key factors in local results. A wording trick cannot remove distance. [7] Explain the real operational region, possible access conditions and existing locations on your website. Use local project examples if you can provide evidence.

Do not create a series of almost identical location pages that only swap the town name. A separate location page makes sense when it offers distinct information: the local team, access, specific services and local projects. Search spam policies cover manipulative doorway pages and mass-produced content with no value. [8]

For multilingual and international companies

Don't just translate words. Adapt currency, service area, delivery conditions, technical terms and contact methods to the market. A French text about a service only available in Aarau should not seem like an offer for the whole of France.

Google documents hreflang for identifying language and regional variants. The corresponding pages must be consistently linked to one another; the signal does not replace a good translation. [9] Have specialist texts checked by a person who knows the language. Measure open customer questions separately for each market and language, instead of mixing all results into a supposed global market share.

8. Use structured data and media sensibly

Structured data describes content in a machine-readable format. Depending on the page, an organisation, local business or product may be appropriate. Google uses this information for certain search features. The markup must match the visible content and does not guarantee a particular appearance. [10]

Our approach: First correct the public information, then derive the markup from it. Recheck name, URL, hours, price and availability for any changes. Do not invent star ratings or mark other people's reviews as your own. A technical test can confirm the syntax, but not the truth of your statement.

For shops: Product data is more important than advertising adjectives

A helpful product page explains variants, compatibility, dimensions, what is included, available sizes, price components and restrictions. Keep shop data and product feeds in sync. Images should show the actual product. Different products need distinct URLs and information. Check which shopping integrations each AI service supports; a Google feed does not automatically register you with every AI system.

For images and videos

Show processes, results or differences that can be better explained visually. Add meaningful image descriptions and a textual summary of important video content. Don’t write central services exclusively in a picture. A graphic should create understanding, not be another area for keywords.

Do I need an llms.txt?

According to Google's own documentation, it is neither necessary nor a ranking advantage for Google's generative search functions. Special “AI schema” labels or a prescribed text length are also not required there. [1] For other services, you should check the specific intended use. Invest in proven access and content issues first before maintaining an additional file that has no proven benefit to your goal.

9. Measure changes without deceiving yourself

Before revising, determine what you want to observe. If you subsequently select new questions, test a different system, or only show the best answer, you won't be able to compare the results fairly.

Three separate measurement levels. There is no fixed conversion rate from a recommendation to an order.

A simple measurement protocol

Record question, category, system, search mode, market, language, date, repetition and complete answer. Note whether your company is mentioned and recommended, which URL is quoted and whether the statement is correct. Save failed or unclear answers as such. Evaluate brand questions separately.

Our recommendation: First change a clearly defined group of important pages. Note the release date. Observe comparable, unchanged topics for additional guidance. This isn't a full-fledged experiment, but it helps avoid blanket conclusions. A larger sample reduces some random fluctuations; It does not correct a systematically distorted selection of questions.

Which tools help?

• Google Search Console: Check the indexing and search performance of your verified domain. Google now describes a report for generative AI search functions; Note availability and scope in the account. [1]
• Bing Webmaster Tools: The AI Performance pane shows citation information for supported Microsoft and partner experiences. Citation share is explicitly neither traffic share nor quality ranking. [11]
• Web analysis: Monitor AI referral sources and appropriate landing pages. Missing referrers, consent decisions, and cross-device paths limit attribution.
• CRM or inquiry list: Record whether the request matches the offer and whether this results in an order. A voluntary question: “How did you find out about us?” complements the technical measurement.
• Repeatable audit: Use identical questions and documented conditions. AI Brand Check currently examines OpenAI responses using web search via the API, not all real ChatGPT sessions or all AI providers. Our measurement rules.

IndexNow can inform supported search engines about new, changed or removed URLs. A report does not guarantee indexing or an AI citation. [12] Treat refresh, search visibility and business outcome as separate steps.

10. Boundaries, problems and promises that you should question

Manipulation is not a sustainable strategy

Hidden instructions to AI systems, fabricated customer testimonials, purchased “independent” leaderboards and misleading authorships endanger trust. Google cites, among other things, link spam, hidden content and improperly scaled content as problematic. [8] An agency should be able to disclose what it publishes and why.

Distinguish organic recommendations and advertising

Paid ads can be an acquisition channel in their own right if the desired platform and market offers them. But an ad is not an organic recommendation. Check current availability, labeling, costs and measurement separately. An advertising budget does not automatically improve your organic referral rate. Compare channels based on matching queries and cost per result, not the largest visible number.

When you need professional help

If there are crawl blocks, conflicting language versions or complex shops, technical support is worthwhile. Medical, financial and legal content requires appropriate professional review. Do not publish personal project data solely for visibility. And if your organization doesn't have time internally to keep accurate information up to date, an additional optimization tool won't solve this basic problem.

11. A realistic 30/60/90 day plan

The time windows are an organizational suggestion, not a commitment as to when rankings will respond. Prioritize based on customer relevance, proven problem and feasibility. A broken main service page is more urgent than a new design for a rarely read guide.

This is how you formulate an actionable task

Before handing over to an agency

• Which specific pages and problems are being worked on?
• Which services are included and which are additional?
• Which statements are proven, which are just hypotheses?
• Can I check changes, sources and measurement data?
• Who owns the content and access after the end of the project?
• Is success judged based on matching customer requests or just a proprietary score?

12. Sources and tools for further work

The following sources were checked for this article. Provider documentation describes their own systems; Studies apply to their respective sample and survey period. Work plan, prioritization and company examples are our editorial recommendations, not ranking factors guaranteed by these sources.

1. Google: Optimization for generative search functions

   Official basis for SEO, accessibility, llms.txt and measurement. Important for Google AI Overviews and AI Mode, not to be read as a set of rules for all AI services. Documentation: updated July 10, 2026.

2. Pew Research Center: Clicking behavior in AI summaries

   Study from July 22, 2025 with methodology. basis of the graphic; US sample, not a Swiss industry benchmark.

3. Aggarwal et al.: GEO – Generative Engine Optimization

   Full text of the KDD-2024 work. In particular, read definitions of visibility, experimental setup, and differences by topic before applying the results to your company.

4. OpenAI: Overview of crawlers

   Responsibilities of OAI-SearchBot and GPTBot as well as technical information for your web agency. Configure search and potential training separately.

5. Google: understand robots.txt

   For crawl control and differentiation from indexing. Particularly important before changing a blanket bot ban.

6. Google: helpful, reliable content

   Questions for editorial self-examination: benefits, experience, sources and responsibility.

7. Google: improve local ranking

   For company profiles and the limits of relevance, distance and awareness.

8. Google: Spam guidelines

   Check supposed shortcuts: buying links, doorway pages and manipulative mass-produced content.

9. Google: Language and regional versions

   Technical guidance for hreflang and cross-referencing between appropriate versions.

10. Google: structured data

    Basics and limits of special search representations. Additionally: Rich Results Test for technical control.

11. Bing: AI performance, topics and citation share

    Official description from June 16, 2026. Helps to classify citation data correctly. Getting started: Bing Webmaster Tools.

12. IndexNow: Report changes to URLs

    Technical guidance for supported search engines; no promise of indexing or ranking.

Further helpful introductions

• Google Search Console – verify your own domain, examine indexing and search performance.
• PageSpeed Insights – check technical loading and usage experience; no AI ranking score.
• Our guide to measurement – Key figures and fair comparisons in detail.
• Our methodology – How the audit works, its evidence base and its limitations.