In the AI era, people often ask bots to recommend service providers and follow their advice. So, we decided to test how competent these bots are. We asked several of them to recommend a Ukrainian translation agency and analyzed their answers.
First, we asked a simple question: “Recommend a Ukrainian translation company.”
We tried ChatGPT, Claude, Gemini, and DeepSeek. Since we know the leaders of the Ukrainian translation industry, we expected to see familiar names on the list, especially Technolex Translation Studio, which has consistently ranked among the top three Ukrainian LSPs, sometimes even first. Technolex has been recognized by Nimdzi, Slator, and Common Sense Advisory, all respected consulting agencies in the language industry.
Well, things turned out differently. Chatbots produced random results. Even worse, the same chatbot gave different answers each time. Sometimes they mentioned real leading Ukrainian translation companies. But other times, we saw names we had never heard before. “Who is that?” we wondered. One response seemed especially suspicious, so we asked the AI where it found these companies. The answer was shocking:
“I’ve just made up the list. Do you want real companies?” So yes — it fabricated the data.
Of course, hallucination wasn’t always the case. Sometimes, chatbots did list actual translation companies, but not necessarily Ukrainian ones. Some were based in Poland or were MLVs from the USA. Others weren’t companies at all but portals like ProZ.com or TranslatorsCafe, which are marketplaces, not translation providers, and certainly not Ukrainian.
In short, it was chaos. Chatbots could not give a reliable recommendation for Ukrainian translation services. Or, most likely, for any other type of supplier.
Why does this happen?
The simple answer is that AI is either not knowledgeable enough or misinterprets the question.
To get a more accurate result, one must ask a very specific question, such as: “Please list the top translation companies of Ukraine. Only companies. Only located in Ukraine. Only the ones with proven recognition by Nimdzi, Slator, Common Sense Advisory, or other reputable rating agencies. Only ISO certified.” This phrasing produces better, though still imperfect, results.
But most people won’t ask such detailed questions. They don’t know about industry ratings, standards, or how these systems work. People ask simple questions, and receive simple, randomized, or even invented answers.
So, we’re still far from the day when AI can recommend service providers based on objective expert knowledge rather than crowd guessing.
