GEO Case Study: How We Make Webwavers Visible in ChatGPT, Perplexity and Google AI Overviews

By Edip Kilinc, founder of Webwavers. Published November 1, 2026, last updated November 1, 2026. Reading time: about 8 minutes.

Key points at a glance

We apply GEO (Generative Engine Optimization) to ourselves first before we recommend it to clients. This case study shows our approach step by step and the numbers from the first measurement. It is deliberately a baseline: the new content had not yet been published at the time of measurement. We will update the numbers once it is live and enough days of data are in.

• Collected 804 real search questions, clustered them into 148 topics, one strong page per topic instead of one page per question.
• Wrote 53 guides, 21 specialist pages and 3 calculators, researched, backed by sources and independently reviewed, plus an English version.
• Measure 350 prompts daily in ChatGPT, Perplexity and Google AI Overviews.
• First measurement, as of November 1, 2026, across 1,398 AI answers: Webwavers visible in 17.7 percent, in ChatGPT in 28.7 percent.
• When Webwavers is mentioned, it is on average at position 2.0 and named as the first agency in 62 percent of cases.

Table of contents

• Why we test this on ourselves
• Step 1: Collect prompts
• Step 2: Cluster instead of single pages
• Step 3: Content AI can cite
• Step 4: Tech and structure
• Step 5: Sources outside the website
• Step 6: Measure, but properly
• The numbers from the first measurement
• What we learned
• What you can take away for your company
• FAQ

Why we test this on ourselves

Decision-makers increasingly research vendors through AI assistants. According to Gartner, 45 percent of B2B buyers used generative AI in their last purchase, mainly for vendor and product research (Gartner, May 2026). 67 percent prefer to buy without a sales rep (Gartner, March 2026). If you don't show up in those answers, you don't even get asked.

But there are no guarantees in GEO and hardly any reliable benchmarks. That's why we test every method on our own website first. What works here, we use for clients. What doesn't, we drop.

Step 1: Collect prompts

GEO doesn't start with keywords, it starts with questions. First, we collected what potential clients actually ask AI assistants. To do that, we combined prompt data from an AI visibility tool, keyword data from SEO tools and real questions from initial consultations.

The result was 804 prompts. They range from "Which Webflow agency in Munich can you recommend?" to "What does a website relaunch cost for a mid-sized company?" to very specific questions like "We're a machinery manufacturer and need a product catalog with data sheets in several languages. Which agency can build that reliably?"

We assigned every prompt to a stage: understanding the problem, comparing solutions, looking for vendors or vetting vendors. That matters because a comparison question needs a different page than a recommendation question.

Step 2: Cluster instead of single pages

One page per prompt would be the fastest route to thin content. Google treats mass-produced pages without their own value as spam, and AI systems also prefer to cite one thorough page over ten similar ones.

So we grouped the 804 prompts into 148 topic clusters. For each cluster, we decided:

Outcome	Number of clusters	What we did
New guide or comparison	62	in-depth article with sources, tables and FAQ
New industry or service page	14	specialist page with examples from the industry
Calculator	3	interactive calculator with an open methodology
Existing page	45	only added internal links
Outside the website	16	directories and review platforms
Careers, glossary, FAQ and other	8	own pages, one placeholder, two discarded

We merged some clusters into a single guide, which is why we ended up with 53 guides. Besides industry and service pages, the 21 specialist pages also include careers and FAQ pages. The table also shows what we deliberately didn't do: about a third of the topics are already covered by an existing page. Better internal links are enough there.

Step 3: Content AI can cite

AI assistants like to pick up clear, verifiable statements. So every new page followed fixed rules:

• The answer is in the first paragraph, not at the end.
• Every number has a source, and the source is linked.
• Key takeaways sit as a short list at the very top.
• Comparisons are neutral. We offer Webflow, WordPress, TYPO3 and headless and recommend what fits the project.
• We state costs as market ranges with a source, not as bait.
• Every page has an FAQ with direct answers.

Every text was researched, written and then checked against these rules in a second, independent review. Only then did it go into the CMS. That's how we created 53 guides, 21 specialist pages and 3 calculators, for example the cost per lead calculator or the guide to choosing a GEO agency.

Step 4: Tech and structure

Good content only helps if crawlers can find and understand it. For us, that includes:

• Every page is readable without JavaScript, the text is in the HTML.
• Structured data (JSON-LD) for articles, FAQ, breadcrumbs and the organization.
• A table of contents with working jump links, so answers can link to individual sections.
• A clean sitemap and redirects for old URLs. The backlink analysis alone surfaced 33 old addresses that we redirected to matching new pages with 301s.
• An English version with its own URLs and correct hreflang tags.
• An llms.txt as a small extra. According to Google's John Mueller, no AI system was actively using this file in June 2025 (Search Engine Roundtable), but it takes very little effort.

Step 5: Sources outside the website

For recommendation questions, AI assistants often don't cite the agency itself but comparison portals and best-of lists. In our measurement, review platforms and agency directories regularly show up as sources, frequently more than the agencies' own websites.

That's why maintaining these profiles is part of GEO work: complete details, current services, genuine reviews from clients. The source analysis in our tracking shows which portals matter. They differ a lot by industry.

Step 6: Measure, but properly

We measure 350 prompts daily in three systems: ChatGPT, Perplexity and Google AI Overviews. We separate four metrics that often get mixed up:

Metric	Meaning
Visibility	share of answers in which Webwavers is named or webwavers.de is linked as a source
Mention rate	share of answers that name Webwavers in the text
Citation rate	share of answers that link webwavers.de as a source
Position	where Webwavers ranks among all vendors named

Important: AI answers fluctuate. The same question can get a different answer today than tomorrow. So only values across many answers and several days are meaningful, never a single screenshot.

The numbers from the first measurement

The first measurement across all 350 prompts is as of November 1, 2026 and covers 1,398 AI answers. 50 core prompts have been running since October 30, the other 300 had their first run on November 1. At that point, the new guides and specialist pages had not yet been published. So the values show how visible Webwavers was with the previous website.

Metric	Value
Overall visibility	17.7%
Mention rate	15.9%
Citation rate webwavers.de	15.6%
Visibility in ChatGPT	28.7%
Visibility in Google AI Overviews	14.0%
Visibility in Perplexity	10.2%
Average position when mentioned	2.0
Named as the first agency	62% of mentions
Named among the first three	83% of mentions

The distribution is more interesting than the average:

• For all 21 questions about Webwavers itself ("Is Webwavers a good agency?"), we are visible, mostly with sources straight from our website.
• For industry questions such as websites for aviation or energy, we are named in about 57 percent of answers, for "Webflow agency in Munich" in 86 percent.
• For topics without a dedicated page so far, such as ongoing SEO support, GEO or conversion optimization, we were close to zero. Exactly these topics are covered by the new content.

Among all brands named in the text of the answers, Webwavers has the highest mention rate, partly because 21 of the 350 prompts ask about Webwavers directly. Comparison portals are partly ahead on visibility because they are often linked as a source but rarely named in the text.

What we learned

1. No page, no visibility. Where a matching page exists, we get named. Where none exists, practically never. It sounds trivial, but it is the most important finding.
2. The systems differ a lot. We are almost three times as visible in ChatGPT as in Perplexity. If you only measure in one system, you only see part of the picture.
3. Portals are gatekeepers. For broad recommendation questions, directories and best-of lists have a say. Without well-maintained profiles, you miss out on part of those answers.
4. When named, then near the top. With position 2.0 and 62 percent first mentions, the problem isn't placement, it's the number of topics we appear in.

We will run the next evaluation once the new pages have been live and indexed for a few weeks. Then we'll compare the same 350 prompts against this baseline.

What you can take away for your company

• Start with 20 to 40 questions your customers really ask and measure them in several systems. That's your baseline.
• Check for each question: is there a page on your website that answers it directly?
• Build one thorough page per topic rather than many thin ones.
• Back numbers with sources and put the answer at the start.
• Maintain the portals that show up as sources in your industry.
• Measure over weeks, not individual answers.

If you don't want to do this yourself: we offer GEO as an ongoing service, from the baseline to content to measurement. How to choose an agency for it is in the guide to choosing a GEO agency, and whether a retainer or a one-off project makes more sense is in the comparison SEO and GEO: retainer or one-off project.

FAQ

What is GEO?

GEO stands for Generative Engine Optimization. The goal is for your company to be named and linked as a source in answers from AI assistants such as ChatGPT, Perplexity, Gemini or Google AI Overviews.

How does Webwavers measure AI visibility?

With 350 fixed prompts that are queried daily in ChatGPT, Perplexity and Google AI Overviews. We evaluate separately whether Webwavers is named in the text, linked as a source and at which position we appear.

How visible is Webwavers in AI answers?

In the first measurement (as of November 1, 2026, all 350 prompts), Webwavers was visible in 17.7 percent of 1,398 AI answers, in ChatGPT in 28.7 percent. When we are named, it is on average at position 2.0.

Why isn't the new content in the numbers yet?

Because the first measurement deliberately ran before publication. That gives us a clean baseline, and later we can show what the new pages change.

Can mentions in ChatGPT be guaranteed?

No. Nobody controls what a language model outputs. You can only increase the likelihood, with good content, clean tech and well-maintained external sources.

How long does it take for GEO to work?

That depends on how quickly new pages are crawled and indexed and how often the systems refresh their sources. First changes are often measurable after a few weeks. We'll show our own values as soon as they are in.

Conclusion

GEO isn't a trick, it's diligent work with a system: collect real questions, cluster them cleanly, build backed-up content, maintain tech and portals and measure honestly. Our baseline shows that where there's a matching page, we get named near the top. Now it's about increasing the number of those topics. We'll keep this case study up to date with new measurements.

Sources

1. Gartner: 67 percent of B2B buyers prefer a rep-free experience, March 9, 2026
2. Gartner: 69 percent of B2B buyers turn to sales reps to validate AI-generated insights, May 20, 2026
3. Google: Spam Policies for Google Web Search
4. Google: AI features and your website
5. Search Engine Roundtable: Google on llms.txt, June 2025
6. Webwavers' own measurement: 350 prompts, ChatGPT, Perplexity and Google AI Overviews, as of November 1, 2026 (50 core prompts since October 30), 1,398 answers

Change log: Published November 1, 2026 with the baseline measurement. Next update after the new content is published.