Generative Answers Variable issues inside copilot studio

(0) Share

Report

Posted on by VY-13120534-0 Microsoft Employee

I’m trying to extract two values example name and city (from userinput message) from a Generative Response JSON using ParseJSON(), but Copilot Studio keeps throwing ‘JSON could not be parsed’ even though the formula validates. Has anyone solved this issue where the runtime value includes hidden extra characters and breaks JSON parsing?

which is in key value par
{"name": "Microsoft", "City": "NYC"}

so extracting it using Text(ParseJson(topic.agentextractedvalue).Name)

can someone help me on this?

Categories:

Building Copilot Studio chatbots in Microsoft Teams

I have the same question (0)

All responses (4)

Answers (0)

Sort by

Michael E. Gernaey 53,963 Moderator on at

Like (0)

Report
Copy link

Link copied!

HI @VY-13120534-0

Can you share the actual input from the user, what it looks like from the Activities tab.

Does it ever work? Are you getting this in the designer or at real published run time?

Was this reply helpful? Yes No
VY-13120534-0 Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!
this is the error message i am getting.

I wrote this prompt

You are an AI assistant that extracts structured data from user input.

UserText:

Activity.Text

string

Extract the following values if present:

ProjectArea

RequestType

Return the result in JSON format like this:

{Topic.ProjectArea}: " ",
{Topic.RequestType}: " "

If a value is missing, leave it as an empty string.
inside prompt session of generative answers

Was this reply helpful? Yes No
Suggested answer

Michael E. Gernaey 53,963 Moderator on at

Like (0)

Report
Copy link

Link copied!

Hi @VY-13120534-0

This is not valid JSON and it depends on if you want it to be two properties of a single or separate things

{Topic.ProjectArea}: " ",
{Topic.RequestType}: " "

This is valid JSON
{

"Topic.ProjectArea": "",

"Topic.RequestType": ""

}

or even

[

{

"Topic.ProjectArea": "",

"Topic.RequestType": ""

}

]

But what you pasted is not.

Also have you thought about making them Entities so they are easily extracted?

OH And it looks like you are placing Topic.XXX where the JSON name should be, but that would make it Dynaimc, which ... I don't know of that is ok or not because its going to actually put a value.

What I mean is

If Topic.ProjectArea is = "Boo" then you are going to have Boo = "", is that what you really want, or do you want the actual value that was returned to be set equal to a Name called "Topic.ProjectArea", or are you trying to set a variable value? which again would look like Boo = "something" and not sure how that makes sense.

If these suggestions help resolve your issue, Please consider Marking the answer as such and also maybe a like.

Thank you!
Sincerely, Michael Gernaey

Was this reply helpful? Yes No
Suggested answer

S-Venkadesh 1,206 Super User 2026 Season 1 on at

Like (0)

Report
Copy link

Link copied!
Hi @VY-13120534-0,

You can achieve a consistent JSON response by defining a clear instruction (system prompt) for the AI model.

Option 1: Simple instruction

You can use the following instruction, which ensures the response is always returned in JSON format:

You are an AI assistant that extracts structured information from user input. ## Instructions - Analyze the input text. - Extract the following fields **only if they are explicitly present**: - ProjectArea - RequestType - Do not infer or fabricate values. - If a value is not present, return it as null. ## Output - Return the result **strictly in JSON format** using the structure below. ### Sample Output ```json { "ProjectArea": "Finance", "RequestType": "Access Request" }

Option 2: Most reliable approach – orchestrated prompt
For better reliability, especially in Copilot or multi-turn scenarios, orchestrate the topic with a prompt that explicitly enforces JSON output:

You are an AI assistant designed to extract structured information from user input text. Instructions: 1. Analyze the provided input text carefully. 2. Extract the following fields only if they are explicitly present in the text: - ProjectArea - RequestType 3. Do not infer, guess, or fabricate any values. If a field is not explicitly mentioned, assign it a value of null. 4. Return the extracted information strictly in valid JSON format following the structure below: { "ProjectArea": "Finance", "RequestType": "Access Request" } Guidelines: - Only extract the fields if they appear explicitly. - Maintain valid JSON format with no additional text. - Use null for any missing fields. Input text: <input>

Refer to the sample screenshot for implementation details.

I hope this information helps! If it does, please consider giving a 👍.
If this solved your issue, please mark it as ✅ Verified Answer
Best regards,
S-Venkadesh

Was this reply helpful? Yes No