From Google to AI: The Skill Gap in Finding the Right Answers

A few years ago, when I was still actively leading my team of developers, I had an intern with whom we experimented with new technological solutions for the company using Symfony and WIX. Since these were new technologies, much information had to be found online while performing tasks. There weren’t always clear guidelines or ready-made implementations.

By the end of each day, the intern would often conclude that “no such information” was available. He would tell me that it was impossible to find anything. However, unwilling to believe that, I would take the initiative and, within five minutes, find the answers on Google—maybe not in the first search result, but certainly by the third or fourth. This led to an unpleasant realization: the information was there all along, but an entire day had been wasted with no results.

What does this story have to do with ChatGPT?

The intern in question had two significant issues. First, he lacked basic skills in using Google effectively. He couldn’t correctly phrase what he was looking for, often entering search terms in an unclear or incorrect manner, which prevented Google from guiding him in the right direction. Second, he lacked the knowledge and problem-solving skills to refine his search based on partial results. He didn’t understand how to adjust his queries or even recognize whether the information he found was what he was looking for. In other words, he didn’t know what to ask to get the correct answer.

This reminds me of many inexperienced users who struggle with ChatGPT or other AI tools. For instance, our current development team lead says that he rarely writes code anymore—he simply tweaks and corrects what AI generates. Meanwhile, junior developers complain that AI results are unsuitable for programming tasks.

The same fundamental problem arises: you need domain knowledge and context to use AI effectively. You must be competent enough to ask the right questions to get the desired results—and even better, you should know how to utilize those results properly.

For example, the first time I encountered Python, I spent about half a day programming scripts that eventually helped automate our project managers’ routine tasks. I didn’t know Python, nor had I ever used the ChatGPT API. The AI-generated results were poor and low-quality, but they were sufficient for me to build a working solution. My years of programming experience allowed me to recognize AI’s mistakes and correct them accordingly.

Beyond understanding how to use the tool itself (whether it’s Google or ChatGPT) and having the necessary domain knowledge, another key factor is realizing where AI can be applied. Since AI is a relatively new tool, many people struggle to identify its practical applications. We are so used to our usual workflows that we don’t immediately think of AI as a way to optimize them.

For example, people may not realize that ChatGPT can process and filter an email list, automatically removing unwanted contacts. We are accustomed to using ChatGPT for conversations and answering questions, but it’s also an excellent tool for routine text formatting, processing, and transformation—especially in its paid versions.

What surprises me the most is how AI adoption has spread. Teaching my parents how to use Google took a long time before they fully embraced internet searches. But today, almost everyone uses ChatGPT—from a bartender with no IT background looking for gift ideas to a marketing guru or developer solving complex problems. Its widespread adoption is happening at an incredible pace.

The key takeaway is that to use AI more effectively, we should invest a little time learning how to use it properly. When forming a query, for example, it helps first to instruct AI on what role it should assume, describe the situation in detail, and specify the expected result. You might even be able to outline the methodology it should use to generate that result.

For example:

“You are the CEO of a 15-person development company looking to implement a salary system based on the Hay Group methodology. Provide recommendations and generate a competency matrix with salary ranges. The output should include competency titles, descriptions, and a possible breakdown of positions.”

This is similar to using Google: you start with a core keyword (e.g., “PrestaShop”), then specify what you’re looking for (e.g., “PrestaShop carrier”), and finally describe the issue (e.g., “PrestaShop carrier not working with range selection”). If you want a simple yes/no answer, you should phrase it as a question (e.g., “Is an apple a fruit or a vegetable?”). To check the current time in a specific location, you can type “current time in New York.” Most of the time, you’ll find the answer within the first five search results.

In summary, the main barrier to broader AI adoption today is ourselves. On the one hand, we don’t always know how to ask the right questions to get the results we expect, and on the other hand, we sometimes lack the necessary knowledge even to know what to ask.

So, how do you deal with these challenges?