This is the second installment of "Prompting 101,” a series to help everyday users make the most out of AI applications. To read the first installment, click here.
The best AI results often come from an interactive conversation between the human user and the chatbot. Collaboration, fact-checking, role playing, and even using AI to write prompts can all be part of "going deeper" and gaining true insights, according to Bryant experts Poojan Patel, Ph.D., assistant professor of Finance and co-author of the newly released textbook, Prompt Engineering for Accounting and Finance: Theory and Practice; and Marketing Professor Stefanie Boyer, Ph.D., co-founder of RNMKRS, an AI platform for training and assessing sales talent, and co-author of Selling with GenAI: Practical Insights for Sales Leaders.
Their intermediary tips on AI prompting include:
Make AI show its work
Ask AI to think questions through step by step (instruct it to “show your reasoning”) before answering your prompt.
“The answers tend to be better, and when something is off, I can see exactly where the logic broke,” Patel says. “Skip this, and you get a confident conclusion with no way to tell whether it holds up.”
Questions first, draft second
Prompting isn’t a one-sided conversation. Let AI ask you questions by typing in: "Before drafting, ask me what information is missing.”
By asking questions, AI can identify things you may have left out of your prompt.
“Questions fill the gaps before AI fills them with guesses about things like budget, audience, and timing,” says Boyer.
“Questions fill the gaps before AI fills them with guesses about things like budget, audience, and timing."
Have AI play your audience
Role-playing with AI is a cheap rehearsal, says Patel.
“Before I send anything important, I ask AI to read it as the end user: a skeptical manager, a busy client, a confused student. It points out the questions and objections I didn’t see coming,” he says.
Let AI write your prompt
When Patel is confronted with a large problem, he sometimes asks AI to help write the prompt.
“I explain my thinking to the AI the way I’d explain it to a colleague: what I’m trying to do, how I’d approach it, and what a good result looks like,” he says. “Then, I ask it to turn that into a prompt. It’s surprisingly good at organizing my thoughts and spotting what I left out.”
“I explain my thinking to the AI the way I’d explain it to a colleague: what I’m trying to do, how I’d approach it, and what a good result looks like. Then, I ask it to turn that into a prompt."
Another way to get AI involved in prompt engineering is during the review process.
“Sometimes I build something with AI one piece at a time, correcting and refining as I go,” Patel says. “By the end, the conversation holds everything I learned along the way. I ask AI to write a single prompt that would produce the result, so next time I can start from that instead of repeating the whole process. The thinking you put in shouldn’t disappear when the chat ends.”
Start a new chat when things go sideways
The back and forth of a detailed AI prompt can make for long conversations, which can get messy.
“Ideas you dropped hang around, and AI starts repeating mistakes or forgetting what you told it,” Patel observes. “When I notice that, I ask for a quick summary of where we landed, open a fresh chat, and paste it in. A clean start usually fixes what 10 more corrections won’t.”
If your initial prompt is getting lost in the weeds, try this one: “Summarize the key decisions from this chat in five bullets so I can continue in a new conversation.”
Keep a file of good prompts
“When a prompt gives me a great result, I save it,” explains Patel. “Over time, this becomes a small library I reach for every week.”
“When a prompt gives me a great result, I save it. Over time, this becomes a small library I reach for every week.”
Many AI tools let you save standing instructions, so you don’t have to repeat yourself. Patel suggests writing (and reusing) a prompt like this: “Meeting summary: Turn the notes below into decisions, action items with owners, and open questions. Notes: [paste new material].”