In short
Colombian Cacao Experience is a hands-on chocolate workshop in Medellín, Colombia, where visitors make their own chocolate from bean to bar. The goal was to be named in answers from ChatGPT, Perplexity, Gemini and Google AI Mode, including broad questions such as “what to do in Medellín”. At the baseline in May 2026, that happened on none of the broad questions. At the re-test in August 2026, Google AI Mode names the client as the only provider in a category of its own; ChatGPT, Perplexity and Gemini do not yet, and that is what I am working on now.
Facts at a glance
| Client | Colombian Cacao Experience |
|---|---|
| Sector | Tourism and activities (B2C) |
| Period | May 2026, ongoing |
| Disciplines | GEO, SEO, Content, Strategy |
| Technology | WordPress, Elementor, Rank Math, Semrush, Search Console, JSON-LD schema |
| Status | In progress |
About the client
Colombian Cacao Experience is based in the Laureles neighbourhood of Medellín. Visitors roast, grind and pour their own bar, in English or Spanish. There is also an evening session pairing rum with chocolate. Most guests are travellers already in the city, looking for something to do.
The brief
Travellers don’t search for chocolate. They ask an AI assistant what to do in Medellín. The client wanted to be in that answer, next to well-known activities such as coffee tours. A second aim: when an AI system mentions the client, the link should point to their own website, not to a booking platform that takes a commission.
Where things stood
I started with a baseline on 19 May 2026. I asked four AI platforms ten broad questions about things to do in Medellín. Result: zero mentions on all ten questions, on all four platforms. Chocolate did not exist as a category in the answers; coffee tours did.
I also delivered an audit report and a NAP check (making sure name, address and phone number match everywhere). At the end of July 2026 the site moved to WordPress; I did not do that migration myself. My interim audit after the move found, among other things, old schema code (structured data for search engines) with outdated prices, two schema blocks side by side, empty titles and descriptions, test pages in the sitemap and booking pages without an H1 (main heading). The site had a translate button, but not a single real Spanish page and no hreflang (code that tells search engines which language version is for whom).
What I did
Measuring before changing anything
I first recorded where the client stood, using fixed questions, platforms and languages. That way every later measurement can be compared fairly.
Strategy aimed at the broad question
The client is the only bean-to-bar workshop in the city, so winning “best chocolate workshop” is worth little. I built the content strategy around the broad questions travellers ask instead. I also wrote a style guide so every text uses the same terms and facts.
Technical groundwork
I wrote schema markup and page copy in English and Spanish. On 23 September 2026 I published the Spanish copy as real pages, with the language set to Spanish and with hreflang. I am now working through the twelve technical points from the interim audit (status 23 September 2026).
A choice I deliberately did not make
I did not treat the translate button as the answer for Spanish. Visitors see Spanish text, but search engines and AI systems only see English. That is why separate Spanish pages are now live.
How I checked it
For the re-test on 29 August 2026 I used the same platforms, languages and ten broad questions as the baseline. I added two questions, making twelve in total. I report the broad questions first, with every outcome, zeros included. Only then do I report the narrow question about the best chocolate workshop, always noting that the client is the only provider in the city, so a mention there proves little.
I also check what my tools tell me. Semrush flagged ten broken links. They turned out to be WhatsApp buttons; WhatsApp blocks crawlers (programs that read websites), but the buttons work fine. Semrush also reported no schema on the site. That was wrong: the JSON-LD was there, Semrush just didn’t detect it. Without that check I would have spent time on problems that didn’t exist.
In the source code, every page had its language set to English (en-US). So crawlers only see English. That explains why the Spanish broad questions returned nothing.
Results
Broad questions, re-test 29 August 2026 (checks on all four platforms):
- Google AI Mode: for “best tours in medellin” a separate category “Bean-to-Bar Chocolate Workshops” appears, with Colombian Cacao Experience as the only provider named.
- ChatGPT, Perplexity and Gemini: still zero mentions on all broad questions.
- Spanish broad questions: zero on every platform. As expected, because no Spanish pages were live at that point.
Narrow question: when asked for the best chocolate workshop, all four platforms name the client. It is, however, the only provider in that category.
What the client says
Great working with Julian. Very professional. went above and beyond. thank you so much
These reviews were posted by verified clients on Fiverr. View them on my Fiverr profile. See all reviews on Fiverr
What’s next
I am finishing the technical fixes. After that comes the content for the broad questions. I check every article for real experience and correct facts. The next re-test uses the same twelve broad questions, on the same four platforms, in both languages. That will also be the first measurement of the Spanish pages.
Frequently asked questions
Is it worth ranking for “best chocolate workshop” if you’re the only one?
Not much. All four platforms already named the client there. Travellers start with a broad question, so that is where the gain is. It is also where three of the four platforms still say nothing.
Is a translate button enough for Spanish-speaking visitors?
No. In this case search engines and AI systems only saw English, despite the button. Every Spanish broad question returned zero. That is why real Spanish pages are now live.