The Value of Vocab

When I was in seventh grade, our class dean, Ms. Allen, was once introduced by the principal as the person with the largest working vocabulary of anyone he had ever met.

I’m not sure why this specific moment stuck with me so clearly, but it did. At thirteen, that felt like the highest compliment a person could receive. To have command of so many words felt powerful, like the pinnacle of inteligence.

Since then, I’ve actively tried to build my own working vocabulary. Any time someone uses a word I don’t know, or a turn of phrase that I find interesting, I write it down: apogee, vicissitudes, a priori. I don’t remember all the words, but I do remember some, and the habit of paying attention to language stayed with me over time.

I reflected on the value of vocabulary again when I joined Bain as an Associate Consultant right out of college.

A big part of how and why consulting firms add value is that they are excellent at codifying and sharing what they learn across projects. At Bain we had a site called the GXC, now IRIS, which held an enormous archive of primers and past work. When you started a new project, you would comb through this site for helpful resources. The best documents were always the ones that had a glossary at the end, decoding the niche vocabulary of whatever obscure industry you were studying that day.

I learned quickly to parrot that language back to clients and to experts who we called with questions. If you ‘spoke the language’ you gained immediate credibility. People would take you seriously, and in turn you could unlock greater insights - all because you knew the right words to use.

Word Choice in the Age of AI

Now, it’s 2025. LLMs have changed how we gauge (and access) intelligence. But my view is that vocabulary is more important than ever.

Models are powerful, but the results depend heavily on the prompt. The skill is not only knowing what to ask, but honing the exact language and structure that unlocks the right response. This is non-trivial, to the point where ‘prompt engineering’ is now a skill cited in job descriptions.

A common prompt engineering trick involves ‘priming’ the model with its identity that carries expertise: “You are a specialist in X”. “You are a researcher focused on Y”. But your prompt can be even stronger, and you can derive greater value, if you extend past this and use the actual language of the domain in your prompt. Much as a junior consultant can earn credibility in an expert interview with the right abbreviations, using targeted industry terms in a prompt can help AI models pull content that’s more focused and that reflects real expertise rather than SEO noise. The right words guide the model toward better sources and stronger reasoning.

The opposite is also true. Using the wrong words with AI tools can be a meaningful setback. Recently, a classmate needed to gather price and rating data for a large number of restaurants on Google. She asked an AI tool to “scrape Google” to do this. The scraping attempt was repeatedly blocked by Google’s rate limits, and she got frustrated that the model couldn’t do what she wanted.

But the problem was not the model. The problem was the term she chose, “scrape”. That word alone was enough to send the system down the wrong path. The efficient solution used Google’s API, not scraping, and was achievable with AI in minutes. One word changed the entire approach.

Takeaways

I believe that it’s uniquely valuable right now to ‘know enough to be dangerous.’ If I can talk the talk, I can probably ‘prompt the prompt’ and unlock a lot of incredible and targeted knowledge, more efficiently than ever before. Or at least, that’s what I aspire to, and what I work toward every time I write down a new word or phrase in the margins of my paper.

So with that, here’s to growing our working vocabularies, and learning a lot from - and with - LLMs, along the way.