I spent about a year building an essay chatbot. In the process, I learned a lot about the potential and limitations of the technology. My main takeaway: my students typically use this tool in the opposite way they should. 

​

Most use LLMs like ChatGPT to "polish" their essay but this usually smooths over whatever distinct voice is in the writing. The result is an essay with recognizable stock phrases and syntax that make the reader realize that what they're reading has been AI-generated. I've had the experience, more and more in the last couple of years, of reading a student's essay and thinking, "I've read that before..."

​

Where I think LLMs can be most helpful is not in "polishing" but in structuring the writing process so that it resembles a conversation. Rather than staring at a blank page, a chatbot can ask you a series of questions that prompt you to introspect more deeply, and push you to produce insights and connections you might not have found on your own. I call this process conversational writing, and it's the product of simply observing what works best for my students: whenever they're stuck with writing, I just ask them to talk through their thoughts, and they're able to express themselves more freely and vividly than on the page.​​

​

Some points we covered during the webinar:

​

• Why "polishing" an essay using an LLM does not address the central concern of admissions officers: they're looking for content, not style.

• How to prompt an LLM so that it doesn't write for you 

• How you can still harness the power of conversational writing without using AI

​

In the last 10 or so minutes, I answered some questions from attendees, including:

​

• Can your techniques also help with academic papers?

• How can "conversational writing" be used in email correspondence?

• Should my son use humor in his college essay?