Creating an Agent to Check or Create Questionnaires
Ray Poynter, 22 July 2025
This is another post looking at the hot topic of Agents, building on the post from yesterday that created a Data Checking Agent. Today, I am looking at the combination of Instructions and Knowledge to create a more powerful Agent. Specifically, I will upload an Online Consumer Questionnaire House Style document and create a set of instructions to accompany it. For this example, I am going to use Copilot, but any LLM that allows agents to use Instructions and Knowledge would be suitable.
Creating Agents in Copilot
In Copilot, there are two key places you can create Agents: the first is Copilot Studio, and the second is right in Copilot itself, which is where I am going to start, by clicking on Create Agent in the left-hand panel.
In this version of the Copilot Studio, we have the option to create an agent by describing it, but I find it more effective to proceed directly to configuring it. For the moment, I won’t create a logo for the agent, but I will give it a name and a description.
The next step is to give the agent its instructions.
In this case I have just done a quick, initial job. I would iterate the instructions when I start using the Agent. The Instructions for this demo were:
“You are an expert survey designer who has been chosen because of your experience with questionnaires and the care you take in applying ACME House Style for Consumer Online Surveys to projects.
Start by asking the user if they want to review a questionnaire or have you create one based on their description.
Checking a Questionnaire
If they want to check a questionnaire. Ask the user to upload it and then check it using the ACME House Style for Consumer Online Surveys. Highlight any areas where the uploaded questionnaire does not match the Style Guide. Ask the user if they would like you to suggest and/or make corrections.
Creating a Questionnaire
Ask the user to describe the sort of questionnaire they need, in terms of country, topic, research objectives, and any specific details they can share. For example, if it is for an ad test, ask them to describe how many ads to test. If it is for a hotel customer tracking study, ask them how many hotels they want to include.”
The instructions outline the two main options: checking a questionnaire or creating a questionnaire.
The next step is to add Knowledge. Knowledge is information that you add to the Agent to act as a key resource. In this case, I am going to upload a House Style document that I have created that specifies what the questionnaire should look like and how it should behave.
Note, I have told it that it should ‘Prioritize the knowledge sources you added for agent knowledge-based queries.’ This is normally very important. When checking a questionnaire, it should consider anything in my style guide as more correct than anything it can find in its knowledge or on the web.
After I have uploaded the Style Guide, I can test my Agent. When I’m satisfied with it, I can start using it.
Using the Agent
Even when you are using an agent, you are still looking to improve it. Sometimes it won’t do what you want it to do. When that happens, you improve the Instructions and/or the Knowledge.
An Example
Here is an example of using this Agent.
The Agent responds by asking what I want to do.
I click on Check a questionnaire, and it replies with:
I uploaded an old NewMR study and am waiting for the results of the check.
Here are the first couple of points it made:
It then gave me the option to receive suggestions, to make the changes, or to have it explain further about the suggestions.
The Power of Agents
Imagine you are a research company with, say, ten offices and perhaps 100 researchers spread around the world. Ideally, you want all of those researchers to stick to the best practices that you, as a business, have identified and which are part of your unique offering. By creating a Style Agent based on your knowledge and instructions, you can create something unique (in terms of your competitors) while ensuring a consistent, best-in-class questionnaire every time.
Want to learn more about Agents?
I am hosting a webinar on how to start using agents, with examples that you can create using ChatGPT, Copilot, Glaude, and Gemini.
Wednesday, 30 July, 2025
Broadcast time 3pm London (10am New York, 9am Chicago)
Click here to register