Prompt engineering
Prompt engineering is writing the input to an AI model so it gives the output you want. In AI search tracking, it means writing prompts the way customers ask.
Prompt engineering is the craft of writing the input to an AI model so that you get the output you want. Developers use it to make a model follow instructions, use a format, or stay on topic.
For marketers it means something a little different. When you track how AI platforms talk about your brand, the prompts you choose decide what you measure. Writing them well is the difference between numbers that reflect your customers and numbers that reflect your own guesses.
Two kinds of prompt engineering
The first kind is the developer's: long, careful instructions, examples of good answers, and rules about format. That's how products like chatbots are built.
The second kind is the tracker's. Here the goal is to sound like a real customer, because you want to see what a real customer would see. Clever instructions get in the way. A prompt like "List the top 10 coffee brands in a table with prices" tells you how the model handles your instructions, and very little about what a shopper hears.
Writing prompts the way customers ask
A few habits help:
- Write a whole question, the way someone would put it to a friend.
- Leave your brand out of most prompts, so you see whether the platform names you unprompted.
- Add the detail real buyers add, like who it's for, a budget or a place.
- Cover the stages of buying, from learning about a category to comparing brands to choosing one.
For example, "coffee subscription" is a keyword. "What's the best coffee subscription for someone who drinks two flat whites a day?" is a prompt a customer might really type, and the answer tells you far more.
Prompts in AnswerOrbit
When you add prompts, we suggest writing each one as a whole question, the way a customer would ask a person. Generate prompts in Topics lets you choose the mix of learning, comparing and buying questions, and the Demand tab shows how often people ask about a word or phrase. See Topics and prompts.