When ChatGPT came online a few years ago, it became the default model for most people’s tasks. Need help planning a vacation? Workshopping a proposal? Researching pet insurance? “I’ll ask ChatGPT,” was—and to some extent, still is—the most common response.

But generative AI has come a long way since 2023. In addition to the thousands of purpose-built tools for functions like data analysis and coding, there’s also the question of what general-purpose model to choose for everyday queries. ChatGPT now competes with Claude, Gemini, Copilot and others, each with different strengths, pricing structures and ideal use cases.

For most people, the choice has become genuinely confusing—and that confusion comes with a cost. Teams waste time on tools that aren’t right for the job; meanwhile, leaders are greenlighting subscriptions without a clear sense of the problem they’re solving.

Picking the right AI tool may seem daunting, particularly because the landscape is changing so quickly. But with just a little bit of legwork, it is possible to find the best fit.

In order to think through which model will best suit your needs, it’s important to first understand what your needs actually are. Different models have different strengths—some are faster than others, some excel at deep research, some in visual outputs. It’s unlikely that you’ll use the same tool for creating a presentation that you would to draft a meeting summary.

It doesn’t help that updates are being released all the time, meaning the playing field is constantly changing. It pays to stay updated on the latest releases, since what may have been Open AI’s domain one day may be Anthropic’s the next.

MIT Sloan professor Rama Ramakrishnan recommends breaking tasks up into small, discrete parts, and then considering the potential costs of using AI to do them—factors like the difficulty and margin for error. “Some tasks, like writing ad copy, product descriptions, or a movie plotline, have slightly more room for error. Uses that require logical reasoning or factual knowledge; encompass cause-and-effect relationships; or have high stakes, like medical care, demand more accuracy,” he explains. Because LLMs are improving all the time, it’s worth re-running these audits semi-regularly.

Say you’re learning to bake. You start with the basics: a mixing bowl, a whisk, a baking sheet. Those items handle most of what you need. But as you get more serious, you start seeing the value of specialized equipment—a bench scraper for portioning dough cleanly, a pastry brush for glazing, a digital kitchen scale for down-to-the-sugar-granule measuring accuracy.

AI works the same way. Think of general-purpose models like ChatGPT or Claude as your mixing bowl—an essential that you use all the time, but not for everything.

In light of that, consider the tasks you want AI to perform. Can your base model handle them? For example, I was recently toying around with a new headshot, and asked ChatGPT to blur the background. The results were not ideal—the model somehow erased parts of my face, and created a strange, ethereal glow behind my ears. I realized I was trying to use a generalist for a specialist’s job. ChatGPT excels at many things, but for this, I realized I’d be better off turning to a dedicated tool like Canva.

Obviously, it’s impossible to try every offering on the market. But once you identify a base model and use it consistently, pay attention to where it struggles. When you hit a consistent wall, find out whether a specific tool exists for that job, and whether it’s worth the added cost and learning curve.

The Widening AI Training Gap

Not all AI use is created equal. According to a joint study from the professional services firm KPMG and University of Texas Austin, there are certain behaviors that set “sophisticated” users apart—one of which is switching between models or tools depending on the task at hand.

The problem is that AI skills don’t develop on their own . One 2,000-person survey from the The Adecco Group found that only ten percent of C-suite leaders feel their companies are “future-ready;” at the same time, 60 percent of leaders expect employees to update their AI skills on their own, and 34 percent of companies have no policy for how AI should be used at work.

It should not come as a surprise, then, that not everyone is on the same page. Ill-formed (or non-existent) AI training leaves some further behind than others, with women and older workers most at risk. If leaders want their teams to use AI effectively, it’s incumbent on them to create the environment that allows it to happen.

Start small. Run internal workshops where people share what’s working and what isn’t. Create a shared document where teams can log the tools they’ve found useful for specific tasks. Normalize the idea that experimentation is part of the job.

The AI landscape will keep shifting. New models will emerge, old ones will improve and the right choice today may not be the right choice six months from now. What won’t change is the underlying skill: knowing what you need, understanding your options, and being willing to keep iterating.