How to give expert feedback on your AI-generated document

You're reading this because we've delivered a workflow. You've been asked to give expert feedback, and you want to know how. This article explains how to give feedback that improves future results. Your feedback is used only for your specific workflow and never shared with other customers.

In short, use the following rule of thumb: give feedback as if you're training a new colleague:

The more specific, the better;
Explain what you want to see, and why;
Explain as if you want your new colleague to write dozens more of these documents. Give feedback that's generally applicable, meaning: useful for all future runs of this workflow;
It's not about just feedback to improve this plan, but for all future plans. Share your expert feedback with the workflow, and your results will get much better, saving you a lot of time in the future.

The more specific the feedback is, the fewer feedback-loops we need.

What to use as input: two testing approaches

The preferred option: upload all input a consultant would usually use to write a grant. We accept PDF, Word and Excel-files. Please don't attach e-mails as we don't support that.
The second option is to upload a fully written project plan. Please be aware: this is an unfair comparison. We are trying to write towards the result that you upload. Also note: don't upload a plan we used as an example to train the workflow.

Three ways to provide feedback

You can give feedback in one of two ways:

Directly in the draft result, by adding comments or edits in Word or Google Docs.
During a meeting with Twin AI. We will process your feedback and adjust the workflow settings accordingly. This is perfect for in-depth feedback.

Multiple meetings or iterations are often included, depending on the document’s complexity and amount of feedback.

How to recognise what needs your feedback

Scenario	How to recognise it	Solution	Example
It's the input	Output is repetitive, vague, or stretched. You see notices above chapters indicating missing input.	Steps: Provide the missing input. Re-run the workflow with all previous and new input.	The "goal" section is vague and repetitive. Add a mission statement from the client.
It's the prompt: your expertise	Text is okay, but could be improved. You see writing that needs your expert touch: tone, structure, strategy, clarity. Requirements are (more or less) met, but it lacks your added value.	Provide expert feedback: specific, explain what you want to see and why, explain as if you're training a new colleague and useful for all future runs. The more specific the feedback is, the fewer feedback-loops we need.	Example A: Where possible, replace abstract descriptions with concrete examples or scenarios that demonstrate how the process works in practice. Link each step in the explanation to the intended outcome so the reader understands why each stage matters. Highlight any potential challenges, decision points, or dependencies, and describe how these are managed or resolved. Example B: Support the explanation with relevant references, such as benchmark data, case studies, or best-practice guidelines, to provide credibility. Consider grouping related information together to help the reader see connections, and use consistent terminology to avoid confusion.

Scenario

How to recognise it

Solution

Example

It's the input

Output is repetitive, vague, or stretched.
You see notices above chapters indicating missing input.

Steps:

Provide the missing input.
Re-run the workflow with all previous and new input.

The "goal" section is vague and repetitive. Add a mission statement from the client.

It's the prompt: your expertise

Text is okay, but could be improved.
You see writing that needs your expert touch: tone, structure, strategy, clarity.
Requirements are (more or less) met, but it lacks your added value.

Provide expert feedback: specific, explain what you want to see and why, explain as if you're training a new colleague and useful for all future runs.

The more specific the feedback is, the fewer feedback-loops we need.

Example A:

Where possible, replace abstract descriptions with concrete examples or scenarios that demonstrate how the process works in practice. Link each step in the explanation to the intended outcome so the reader understands why each stage matters. Highlight any potential challenges, decision points, or dependencies, and describe how these are managed or resolved.

Example B:

Support the explanation with relevant references, such as benchmark data, case studies, or best-practice guidelines, to provide credibility. Consider grouping related information together to help the reader see connections, and use consistent terminology to avoid confusion.

How to give feedback in scenario 1 and 2

Scenario 1: It’s the input

When input is missing or insufficient, AI will become repetitive. It may use filler language to meet length requirements.

How to recognise it: The draft shows a warning above the section.

Solution: Add the missing input like slide decks, notes, data, or attachments and run the workflow again. Always include all previous input as well.

Screenshot of information missing in a document

Scenario 2: It’s the prompt

You know it's the prompt if the writing can be improved. It requires your expert-feedback on how a chapter should be written. It's your knowledge, your consulting to the client, the experience that brings a plan from "great" to "superb".

Please see the table above for examples of great feedback, and the table underneath for the criteria your feedback should match.

Not useful	Instead, write this
We don't want A to be written	Avoid A, we want to have this: B. This is because of [explain shortly].
This is not useful	Instead of writing A, write as following: XYZ. This is because of [explain in shortly].
This is vague	Give more context on this subject, specifically on topic X, subject Y and innovation Z.
There's too little emphasis on X.	During a recent evaluation, an evaluator stated it's important to support statement X with data Y. Therefore; always try to explain this topic.
This can be explained more briefly	Don't give feedback just yet. Please check again if the section has enough input. If not: run the workflow again with more input. Please check scenario 1 (above) to see which input is missing. This should improve the output. If yes: then it's the prompting. Provide your expert feedback.

Known limitations (for now)

Interlinking between chapters. Content may be repetitive between chapters;
Layout or formatting of tables;
Automatically generated images, diagrams or flowcharts.

These are currently outside the scope of our workflows. However, they can be requested at support@twinai.nl. Also other requests are strongly encouraged!