Power Automate

AI prompt using wrong input

(1) Share

Report

Posted on by MM-05061025-0

Hi,

When the following flow runs, it is supposed to do the following:

- get files from a sharepoint folder

for each of those files

- get file content

- put content through an AI prompt to extract data

However, when I run the flow, the prompt returns the same output for every file which is wrong

"get file content" outputs the correct information and the prompt is set to use the output from "get file content" but it doesn't, for some reason it keeps using the same file over and over.

Screenshot 2026-0...

Your file is currently under scan for potential threats. Please wait while we review it for any viruses or malicious content.

Categories:

AI Builder

Building flows

I have the same question (0)

All responses (6)

Answers (0)

Sort by

trice602 16,294 Super User 2026 Season 1 on at

Like (0)

Report
Copy link

Link copied!

can you please share your flow? Paste in here. Attachments rarely show up/

------------------------------------------------

⭐ If this was helpful, please like and check the box below "Does this answer your question" to help others find this too!

Always glad to help! 💯💯💯💯💯

Tom

Follow me on LinkedIn - Thomas Rice, PMP | LinkedIn

Was this reply helpful? Yes No
David_MA 14,840 Super User 2026 Season 1 on at

Like (1)

Report
Copy link

Link copied!

In addition to sharing the flow within the post as @trice602 requested, please also share the AI prompt especially if you configured a custom prompt. Below is an example of a custom prompt I have. It is doing what I believe you have described you are going. Going through a document and extracting content. You'll see in my prompt I explicitly tell the AI how I want the response formatted when it replies. This has worked out very well so far.

You are an AI system that extracts structured data from legal contracts. Analyze the provided contract text [text input] and return ONLY valid JSON. Do NOT include any explanation, formatting, or extra text.

Return this exact JSON structure:

{

"Summary": string,

"ContractingParty": string,

"CounterParty": string,

"StartDate": string,

"EndDate": string,

"ContractType": string

}

Rules:

- Summary must be less than 150 words and written in clear business language

- Dates must be in ISO format: YYYY-MM-DD

- If a value cannot be found, return null

- ContractingParty = the external entity (not your company)

- ContractType = e.g., MSA, NDA, Service Agreement, Supply Agreement, etc.

- Output MUST be valid JSON that can be parsed programmatically

I also have a question. The title of your post says it is using the wrong input, yet in the body of your post you says the issue is with the output. How do you know it is using the wrong input?

Was this reply helpful? Yes No
MM-05061025-0 6 on at

Like (0)

Report
Copy link

Link copied!

Was this reply helpful? Yes No
MM-05061025-0 6 on at

Like (0)

Report
Copy link

Link copied!

@David_MA @trice602 here

Was this reply helpful? Yes No
David_MA 14,840 Super User 2026 Season 1 on at

Like (0)

Report
Copy link

Link copied!

In my experience, you need to use the recognize text in image or document before you use the Run a prompt action. Withtout that, you are probably passing binary data to the run a prompt action. For your text input in the Run a prompt action, use the Full text output from the recognize text action. If the files include images or image-based pdfs, the recognize text action will perform OCR on the file in order to extract the text.

Was this reply helpful? Yes No
Suggested answer

chiaraalina 2,348 Super User 2026 Season 1 on at

Like (0)

Report
Copy link

Link copied!

Hi @MM-05061025-0

Can you please verify the dynamic content binding in the Run a Prompt is correct. Maybe also add a screenshot of the binding here.

Also, if Concurrency Control is enabled, please disable and try again.

Is the repeated output correct for one specific file (always the same file)?
Or is it generic/incorrect for all files (no file-specific content)?

Let me know!

Was this reply helpful? Yes No