Guides ChatGPT published aug 12, 2026
In this guide, you will learn how to turn customer feedback into fresh ad angles in bulk with ChatGPT Work. You will create two CSVs, generate visual ads, and turn your best process into a simple reusable app.
You will build a local ad factory that turns customer comments into a structured pain-point library, ad copy, image prompts, and visual mockups. The same workflow can process a small test batch or scale across dozens or hundreds of comments.
*This local ad factory turns customer language into reviewable concepts and queues the strongest ideas for image generation.*
Keep the CSVs as the durable layer of the system. You can add new feedback, reuse the image prompts in another tool, or switch models without rebuilding the process from scratch. Because the files and images stay in the local project folder, you can always inspect and reuse them.
If the app works well for you, publish it for your team with ChatGPT Sites (https://app.therundown.ai/guides/deploy-a-mini-saas-in-10-minutes-with-chatgpt-sites). This example keeps the workflow in CSVs, so you can publish a first internal version without adding a database.
Create a new folder on your computer before you open ChatGPT. Give it a clear name such as customer-feedback-ad-factory. This will hold your source CSVs, generated images, app files, and exports.
Open the ChatGPT desktop app, select ChatGPT, then switch from Chat to Work. Open the folder you just created as a new project and start a Work chat inside it. ChatGPT Work is designed for longer tasks and finished deliverables, while the project keeps the files and conversation together.
If Work is missing, confirm that your paid plan is eligible and ask your workspace admin to check the role's Work permission. The feature is still rolling out across eligible accounts, and local desktop access can appear later than cloud access.
Pro tip: Create the folder yourself before starting the project. You can then use the project's three-dot menu and Reveal in Finder to find every CSV, image, and code file without hunting through ChatGPT's default folders.
Gather an initial batch of 10–20 comments from places where customers already explain what is frustrating them. Useful sources include:
Use only information you are authorized to analyze. Remove unnecessary personal details before uploading private messages. You can paste or upload the feedback yourself. Work can also gather it from an approved source when you are logged in and grant the required permission.
Add one example CSV to the project if you already have a preferred structure. Then use this prompt:
Create a CSV that gathers pain points about [BRAND OR PRODUCT]. Use the attached example as the structure when one is available.
Include these columns:
- source
- customer or persona
- pain point
- frustrating moment
- exact phrase
- current workaround
- desired outcome
- primary objection
Preserve the customer's wording in the exact phrase field. Do not invent details that are not present in the feedback. Start with [10 OR 20] comments from [APPROVED SOURCES].Review the resulting CSV before moving on. The pain point needs enough detail to support a useful ad idea, and the exact phrase should still sound like the customer rather than polished marketing copy.
*The first CSV keeps the original customer language beside the pain, desired outcome, and objection.*
Ask Work to create a second CSV instead of modifying the first one. Keeping the evidence and the ad concepts separate makes both files easier to inspect.
Create a second CSV that turns each pain-point row into an ad concept.
For every row, include:
- the source row or ID
- pain point
- ad headline
- CTA
- an idea for a square feed image for Facebook or Instagram
Keep every concept traceable to its source row. Save this as a separate CSV.Once the concept CSV exists, type /side to open a temporary side chat in the same Work project. Use it to build the reusable folder structure while the main chat keeps working with the customer feedback.
Scaffold this folder as a factory for creating ad variations from the pain-point CSV and the ad-concept CSV.
Create clear folders for inputs, generated images, and exports. Keep the output tied to the source row so each file can be traced back to the original customer feedback.Work may create the script and folder structure automatically after this instruction. Review the plan and files it proposes. You do not need another prompt for work that is already underway.
Test the workflow on a small batch before asking for 100 outputs. In the main Work chat, use this prompt:
Generate the first five ad concepts and create two visual variations of each.
Save them in the correct export folder. Give every file a descriptive name that includes its source spreadsheet row or ID so I can match the image to the concept.Work will generate the images one at a time. Open them in the canvas view to compare the batch, comment on an individual image, or select several images for the same revision.
*A small batch makes it easy to compare the creative direction and correct weak concepts before scaling it.*
Look for whether the visual actually communicates the pain point. If it does not, give a direct correction such as:
Create five more variations with more visual emphasis on the pain point. Keep the source concept and customer language unchanged.Pro tip: Automating 100 things at once is usually the worst way to get good output. Do the task manually, fiddle with it until you get what you like, then turn that into a repeatable automation.
Repeat the CSV and image batches until you reach the number of usable ideas you need. Treat the larger sheet as a bank of hypotheses, not proof that every concept is ready to run as an ad.
After you have corrected the manual workflow, Work has enough context to turn it into a simple local app. Ask for the app in the same project so it can reuse the CSV structure, filenames, and creative decisions you already approved.
Build a simple local web app that accepts my customer-feedback CSV and an OpenRouter API key.
Use the CSV to create pain points, ad angles, hooks, ad copy, and image prompts. Let me review and select concepts before generating image mockups. Save the enriched CSV and generated images into organized local output folders.
Make the app easy to use. Let me enter the OpenRouter key in the interface, but do not hard-code it or save it in the CSV, source files, or output folders.Open the app, import the CSV, enter the key in the interface, and process the feedback. The app should automatically create the pain, angle, hook, copy, and image-prompt fields. Select only the concepts you want to visualize, then generate the images.
The app reduces the tested workflow to a few clicks. You still review the source language and creative direction before scaling the output.
*This is one visual ad concept generated after the customer-feedback workflow was reviewed and refined.*
Keep going
Members get all guides, courses, live sessions, and $1,000+ in partner perks.