Building a GEO Program From Scratch: The Kentico Playbook

Starting With a Clear Goal 

It was clear that Kentico needed to be visible in AI-generated answers and in traditional search to stay visible. What wasn't clear was where to begin, which metrics would tell us whether it was working, or what best practices for GEO even looked like. 

We knew we couldn't set a numeric target going in. Partly because we didn't yet know which metrics would matter, and partly because GEO had none of the reference points that make a target meaningful: no established benchmarks, no historical data, no trends to read. When a discipline like GEO is still being defined, picking a target number is mostly guesswork. 

What we did have was a directional goal: grow Kentico's visibility in AI-generated answers to non-branded prompts with an increased AI citation frequency. When someone asks an AI “what are dxp vendors” or “what is a hybrid headless CMS," they're in discovery mode. That's exactly where we wanted to show up. 

More specifically, we wanted to track how frequently Kentico was being mentioned in answers to those prompts, and whether that frequency was growing over time. The trend mattered as much as the starting point: we needed to see whether our GEO actions were actually getting Kentico surfaced in AI search. 

That goal became the filter for decisions that followed. 

Measure Before You Build

The first real task was figuring out how to measure visibility. 

AI search visibility tracking is still hard to quantify. One of the differences between GEO vs. SEO is that there are no standardized rankings, no keyword positions, and no click-through metrics. Even intent is harder to read: a search query signals what someone is looking for; but an AI prompt can be conversational, exploratory, or deliberately open-ended. 

We evaluated several tools built specifically for this space (platforms like PromptWatch, Otterly, and Profound) alongside manual testing approaches where we systematically queried AI platforms and logged what we observed. 

We set up two monitoring tracks: 

• Primary: visibility on non-branded prompts. Tracking how visible Kentico is in AI responses to high-relevance, non-branded queries about digital experience platforms, CMS platforms, and related topics. This is the discovery metric: the one tied directly to our goal. 
• Secondary: visibility on branded prompts. Tracking mentions and citations for prompts that include Kentico or competitors' brand names. This tells us about share of voice and helps catch meaningful shifts in the competitive landscape.

The two metrics we settled on as the core signals: 

• Brand mentions: how often Kentico is referenced in AI-generated answers to relevant queries, including our rank relative to competitors
• Citation frequency: how often those AI answers include a direct link to kentico.com

Citations are particularly valuable, though less as a traffic source than you might expect, since people rarely click links in AI answers, especially in ChatGPT. Their real value is that they indicate AI systems are selecting kentico.com as a trusted source in relevant answers. 

Building the Content Map: Topic Clusters for GEO

With measurement in place, we moved to planning what to actually build. 

We chose a topic clustering approach: organizing all planned content into thematic groups, each covering a broad topic through a structured set of questions and answers. In traditional SEO, one page typically addresses one or two closely related keywords. In GEO, one page can answer multiple questions within the same theme, with each question-and-answer pair functioning as a discrete, citable content chunk that AI models can reference independently. 

For example, one cluster centered on What is a Digital Experience Platform (DXP), covering questions like: 

• What is a digital experience platform?
• How is a DXP different from a traditional CMS?
• How does a DXP support omnichannel customer experiences?
• What are the key features to look for when choosing a DXP?
• What are the main business benefits of using a DXP?

Another cluster covered CMS with digital marketing, connecting content management to campaigns, personalization, and marketing automation. A third focused on CMS vs DXP: the comparison questions that commonly appear when buyers are orienting themselves in the market. 

Each thematic page followed the same structure: substantive answers to the cluster's questions, plus a compact FAQ section at the bottom. We used the existing glossary hub page as a starting point. Many cluster themes already had a corresponding page; where gaps existed, we had a prioritized list of what to build next. 

For theme prioritization, there's no GEO equivalent of search volume: no tool tells you how often a specific question is asked inside ChatGPT or Perplexity. We relied mostly on intuition and industry familiarity: what topics our audience actually cares about, what questions come up in real conversations about DXPs and CMS platforms, what someone needs to understand before making a buying decision in this space. Traditional search volume analysis came in as an occasional sanity check, useful for confirming that a topic had real audience interest, but not what drove the decisions.

What Didn’t Work

Practitioners rarely publish their dead ends, but we're hoping to help, so here are ours. In a discipline this young, knowing what to skip is worth as much as knowing what to do.  

Query fan-out tools (the tools that predict which sub-queries an AI engine runs behind a prompt) did not earn a place in our workflow. They offered mostly irrelevant prompts, and when they were relevant, they were obvious. And as noted above, we never found a usable GEO equivalent for search volume. We would have loved a number to lean on, but intuition informed by real customer conversations turned out to be the better guide. 

Developing the Content

With the clusters mapped, execution followed two tracks. 

On the page, we focused on three things: 

• Publishing glossary pages for core terms under kentico.com/discover/glossary. Each page was designed to contain 10 to 20 structured Q&A pairs, plus concise FAQ summaries at the bottom. The goal was to give AI models multiple structured entry points into each topic: clear, citable answers to specific questions rather than long-form content that requires interpretation.
• Writing blog posts that directly addressed the highest-priority cluster questions
• Adding structured FAQ sections to existing platform pages to answer topical questions in a citable, machine-readable format

Off the site, we updated Kentico's Wikipedia page and became more intentional about participating in relevant Reddit discussions, particularly in threads where our topic clusters were actively being discussed. 

This method probably sounds familiar: the tactics that build authority in traditional search (quality content, question and answer structure, authoritative sourcing, relevant engagement) are the same tactics that build visibility in AI-generated answers. GEO is not a separate discipline invented from scratch. It's still content optimization for engines, just extended to a new distribution channel. The channels are different; the underlying logic of relevance and authority is the same.

Closing the Loop: What Gets Cited

Tracking brand mentions and citation frequency tells us how often we show up and how prominent our appearance is. But to understand why, and what to build next, we needed to go one level deeper: analyzing which specific pages are being cited, how frequently our domain shows up in responses to the non-branded prompts we monitor, and how it all trends over time. 

In the early months, when our pages appeared at all, they appeared deep in the reference lists. Both metrics have improved meaningfully over the tracking period: our pages now rank higher within reference lists, and our domain appears in them more and more often. 

This is where the loop closes. The strongest performers are the structured formats we bet on: our hybrid headless CMS glossary page has become one of our most frequently cited URLs, climbing into the market-wide top 10 of cited sources in June, and our comparison-style alternative pages earn citations month after month. Long-form content plays a supporting role; structured, citable answers are what gets picked up. This finding now drives our content plan. 

Results: So Far, So Good

Back to where we started: grow Kentico's visibility in AI-generated answers to non-branded prompts. 

The chart below tracks brand mentions and citation frequency across the non-branded monitor from October 2025 through June 2026. The two metrics operate on different scales, so each has its own y-axis: brand mentions (orange) on the left, citation frequency (purple) on the right.  

Brand mentions have trended steadily upward, with only small month-to-month dips along the way, a trend we have seen among our competitors as well. Citation frequency declined through the first several months, then reversed in the spring and has climbed every month since.  

One more pattern worth calling out: because we track competitors on the same monitors, we can see that some month-to-month dips happen across the whole vendor set at once, not just for us. It looks remarkably similar to the way SERP data used to shift after major Google algorithm updates. So before reacting to a drop in your own numbers, check to see whether the whole market moved with you. 

The Path Forward

We're still early. GEO as a discipline is still being defined, and anyone working in it today is building practices that will be refined significantly in the years ahead. But the foundation is in place: clear metrics, a structured content architecture, and a measurement loop that tells us when something is working and when it isn't. 

Scaling the Loop With an Agent

Since we launched this program, one thing has changed on our side: as part of our Agentic Marketing Suite, we built an agent to simplify this work.  

Xperience by Kentico now includes the SEO & GEO Specialist, an agent that evaluates pages for SEO and GEO readiness and returns a scored report with prioritized recommendations. It now takes over parts of the workflow we used to do m. If you are starting today, you can run the same playbook with less manual work. 

If you are starting to implement your GEO strategy next quarter, this is what we suggest. Treat it as a starting point: 

1. Think about a directional goal first: where you want to show up and for whom. A numeric target can come later, or not at all
2. Think about measurement before creating content: you can choose your own signals; brand mentions and citation frequency on non-branded prompts worked for us
3. Map out your topic clusters, letting audience relevance guide prioritization more than tool output
4. Explore on-page (glossary pages, structured Q&A, FAQ sections) and off-page (Wikipedia, communities) tactics; we ran both in parallel 
5. Watch which pages actually get cited and let that feed back into your content plan

We're continuing to expand cluster coverage and will keep sharing what the data shows about our GEO efforts. 

If you're looking for advice on how to get started with your own GEO strategy, take a look at our free Marketer's Guide to SEO & GEO for additional tips and a 30-day action plan to help you hit the ground running.