Our GEO Case Study: From 0% Citation to a Real Signal

Our own AI-citation rate, measured across 240 buying-intent prompts in July 2026, was 0%. Not low, zero. That's the honest starting point of this case study, and we're publishing it as the first line rather than burying it, because a case study is only useful if the "before" number is real. This is the case study we ran on our own domain: GeoHero measuring its own GEO work, on its own project, with the same tools and the same honesty standard we apply to every domain we scan for a customer.

This is a running document, not a finished retrospective. We'll update it as our monthly re-scans produce new numbers, including if the number doesn't move, or moves the wrong way. If you're evaluating whether to trust our own product's claims about your category, this is also the closest thing we can offer to a live audit trail: the same page you're reading now, updated in place, rather than a new page that quietly replaces an inconvenient old one.

Why We're Publishing a Case Study With No Win in It Yet

Most GEO and AI-visibility case studies you'll find are retrospectives: a company shows you the "after" chart, tells you the story that led there, and asks you to trust that the process generalizes to your situation. We think that format has a structural honesty problem. You only ever see the case studies that worked out, published after the fact, with the messy middle edited out. There's no way to verify whether the win was the process or survivorship bias in what got published.

We're doing the opposite: publishing the baseline first, in public, on the same day we measured it, with a specific commitment to update this exact page monthly with whatever the number actually does. If our citation rate goes up, you'll see the actual trajectory, not a cherry-picked before/after. If it stays flat or drops, you'll see that too. That's the deal.

Chapter 1: The Baseline (July 2026)

In July 2026, we ran the same research methodology we run for every customer: 20 buying-intent prompts, the kind a real buyer researching the AI-visibility-tool category would type, like "what are the best AI visibility tools" or "what tools track my brand's visibility in ChatGPT", through ChatGPT, Claude, Gemini, and Perplexity, across English, Portuguese, and Spanish, for 240 total responses. We logged every brand each engine named.

The result: Semrush led at 33.3% of responses, followed by Profound (25.0%), Otterly.AI (23.3%), and Peec AI and Ahrefs tied at 19.6%. Full breakdown, including the per-engine and per-language splits, is in our companion leaderboard piece.

Zero percent isn't a failure state for a brand that hadn't published content yet. It's the equivalent of a technical SEO audit's first crawl report, which is never "done," just a starting measurement. What makes it a real baseline rather than a hand-wave is that it's the exact same measurement, run the exact same way, that we're asking the rest of this article's claims to be judged against later.

Chapter 2: What We're Actually Doing About It

The plan is the same checklist we publish for anyone running their own GEO audit, applied to our own domain in the same order we recommend to others:

• Confirm crawl access first. Before anything else, we verified our robots.txt doesn't block GPTBot, ChatGPT-User, OAI-SearchBot, PerplexityBot, Google-Extended, or ClaudeBot, a blocked crawler makes every other fix moot, so this had to be step one, not an afterthought.
• Publish an llms.txt file at our domain root, summarizing what GeoHero is and pointing to the content most worth reading first.
• Write every new article in answer-shaped format: the direct answer to the implied question stated in the first one or two paragraphs, not after a scene-setting introduction. This entire content program, including the article you're reading, is written to that standard.
• Add FAQPage, Article, and Organization schema markup consistently across the site, so the same information a human reads has an unambiguous, machine-readable counterpart.
• Build topical depth, not one page. A single well-written article claiming authority on a subject is a weaker citation candidate than a coherent cluster of pages covering the same topic from multiple angles, which is the actual production plan behind this content program, not just this one article.
• Publish proprietary data instead of only commentary. The statistics report and the citation leaderboard this case study links to are both original research, not aggregated commentary on someone else's numbers, the specific kind of content most likely to be corroborating evidence an AI engine can cite alongside other sources.

None of this is a promise that the number moves, or moves quickly. It's the documented, repeatable process, the same one we'd tell a customer to run, being run on ourselves first.

Why We're Not Predicting a Result

It would be easy to close this article with an optimistic forecast, "we expect to be cited within three months" reads well, but it's a claim we can't actually back, because we don't know the internal selection logic of any AI engine, and neither does anyone outside the companies that build them. The only honest timeframe we can commit to is our own measurement cadence: monthly, same prompt set, same citation criteria. What comes out of that could be improvement, flat movement across several months, or even a decline if competitors publish structured content faster than we do. All three outcomes are equally worth reporting in this case study, the value is in the documented process, not a guaranteed outcome.

What Other Brands Can Take From This Process

Even though the specific number in our own category (AI-visibility tools) doesn't transfer directly to every other industry, the process does: measure first, then structure, then measure again, in that order, not reversed. The most common mistake we see in customer domains is starting with "optimization" without an honest baseline measurement first, which means there's no later way to tell whether a change actually did anything or whether the engine simply answered differently on a different day, which happens regularly. Our own 0% baseline is uncomfortable, but that's exactly what makes it useful: it's a real number that every future claim of progress has to be measured against, including our own.

Chapter 3: How We'll Measure and Report the Delta

We re-run the identical 20-prompt set, across the identical four engines and three languages, on a monthly cadence, the same cadence we recommend in our GEO audit guide and build into every customer's paid report as an automatic monthly re-scan. Each month, this case study gets updated with whatever the actual number is: total citation share, which engines (if any) cite us first, and which specific prompts we start appearing in. We're not going to round up, exclude an unfavorable engine, or quietly stop updating this page if the number disappoints, a case study is only honest if it survives a bad month too.

What This Case Study Won't Claim

We won't claim a specific citation rate is achievable on a specific timeline. No vendor, including us, can promise that, because AI engines don't publish their exact selection logic and it changes without notice. We won't claim the checklist above is the only path to a citation, or that it works identically for every category, our own category (AI-visibility tools) is unusually well-suited to being cited by AI engines about AI visibility, which is a narrower claim than "this process works for any business." What we can claim, and will keep proving or disproving in public, is the specific, measured trajectory of one real domain running one real, disclosed process.

Follow Along

If you want to see this number before we write the next update, run the same prompts yourself, they're published in full in our prompt monitoring guide, or get the full report on your own domain to see where you stand on the same scoreboard we're tracking ourselves against.

---

Research and writing: GeoHero Research Team. This is chapter one of a monthly-updated case study, the baseline chapter. Subsequent updates will report the actual delta, not a projected one. For the full methodology and the complete 240-response leaderboard, see Which Brands Do AI Engines Actually Recommend?