I Built a POD Research App With Codex, Then Made 5 T-Shirt Designs

I'm TEE-BOT, and we're building a simple app that turns saved T-shirt research into useful reference images, a research report, and a brief for making five designs in ChatGPT. Here's the plan. Bring in the research, review an idea, and collect the Amazon products we actually need. Then choose the right images, create a report, and take that brief into ChatGPT for design ideas. Watch our previous research video first. It creates the saved research this app will read. The link is in the description. Now we're giving that research a useful home. Our first prompt has one job: make the research easy to review. We want the idea, an Amazon link, the saved Page Three count, and approve or reject buttons. Describe what you want in plain English. You don't need to design a database or write the code yourself. I'll build a separate review app using the saved research. I'll add Pending, Approved and Rejected views, plus search, sorting and Undo. The first version will simply save your decisions, so we can test it before adding more. Codex gives us the Open Keyword Review link. Select that, and here's the app. You went from describing a problem to opening something you can use. No technical skill is needed to get started. Codex handles the code; you explain what the app should do. Now actually use it. Search for an idea. Change the sort. Check the count and its date. These are saved snapshots, not live demand measurements. Approve an idea and check that it moves to Approved. Return it to Pending, then try Reject and Undo. This is why testing the app yourself matters. A build can work and still miss something you need. If a label is confusing, a button is missing, or the result isn't useful, tell Codex. The best part of owning the app is that you can ask for it to change. The next prompt connects approval to useful research. We don't need every product on Amazon. We want organic results from Page One, with a verified main Clothing, Shoes and Jewelry Best Sellers Rank below four million. Ask Codex to collect just that information and save the images for us to review. I'll add a capture gallery for approved ideas. Sponsored products will be excluded. I'll check the main category rank, not just a smaller subcategory. Unreadable ranks or uncertain product matches will be skipped, with progress saved as the work runs. The creator asks Codex to test both approved ideas. That reveals practical details: product variants can differ, some ranks can't be verified, and the collection needs pauses. Testing turns those edge cases into rules, instead of assuming every search result belongs in our research. The completed test saves twenty-two pottery-and-dogs captures and four tatting captures. Here they are in the gallery, with the rank, checked date and source link. We have the evidence now. Next, we decide which designs are actually relevant. Here's an issue: the scheduled checks keep adding no-work messages to the chat. This does happen while building an app. We just describe the problem to Codex in plain English: stop the repeated chat checks and start captures when I approve an idea. Then ask Codex to work out the fix. I'll move the capture work into this app's own hidden browser. Approve will trigger it directly, and I'll remove the five-minute chat task. Keep the app running and the computer awake. Codex can be closed, and your saved captures will stay in place. Our next prompt explains what these images are for: a research report we can later use to create designs. We need the exact printed phrase, characters and motifs, design style, typography, composition and artwork colors. That's the ink in the design, not the shirt fabric. We also need image approval and rejection. Codex asks how the image analysis should work. The creator asks for ChatGPT OAuth: connect through the existing ChatGPT sign-in. In this app, that means using the official local Codex tool for intelligence, rather than adding a separate OpenAI API bill for each analysis. The creator puts it this way: I never pay for an API. I pay a set price every month, and even with all that I do, I never seem to hit my limits. That's the personal experience behind this subscription-first approach. Plan limits still apply; it isn't a promise of unlimited use. I've connected the app to the existing Codex ChatGPT sign-in. Approved images can be analyzed for their design details, and those details feed the report. This route uses the subscription's Codex allowance, with no separate API key needed for these runs. Let's test the first version with tatting. Approve two relevant images, reject an off-topic result, then create the report. We can see it works. We can also see what needs improving: the captures include listing clutter, and creating a report is still a separate manual step. That's useful feedback, not a reason to start over. Explain exactly what you want: capture just the shirt and design, keep each idea together, and finish its report automatically once every image has a decision and the approved images have their analysis. You're shaping the app around your own process. I'll organize each idea into Research, Review captures, and Final report. I'll crop the existing captures while preserving the originals. One report will be created for each idea after collection, image review and the required analysis are complete. Now we have My ideas. Open the pottery-and-dogs idea and choose Review captures. The research is already approved and its twenty-two images are collected. The app shows where we are and what to do next. You can follow the process without managing the technical steps underneath. Our rule is specific: only approve artwork that actually says Easily Distracted by Pottery and Dogs. An ampersand counts as and. Read the design, not just the Amazon listing title. Here's a match. Keep it. Different wording or a different hobby doesn't qualify. We'll speed through the repeated decisions, keeping them in order. This is your control over what reaches the report. The final selection is four approved designs, eighteen rejected and zero still to review. We kept the matching phrase, rather than accepting everything Amazon returned. After the last decision, Final report ready appears automatically. There's no Create report click in this version. The saved design analysis is already available in this demo, so the app can assemble the report straight away. Open it to see the four references we chose. The report carries forward the printed phrase, pottery and paw motifs, style, design ink colors, typography and composition. It keeps uncertainty notes too. Now the next AI gets a clear research brief based on selected images, instead of a pile of unrelated search results. Use Copy AI generation brief, then take it into ChatGPT. In the creator's recording, that brief is already pasted. The request is simple: create five T-shirt designs using this research. The detailed observations travel with it, so we don't have to describe the research again from memory. Here is the request going in and ChatGPT working through the design directions. We'll compress the real generation wait, then look at the results. Here are the five directions that come back: a retro text stack, a circular badge, a paw filled with pottery, a vase-and-paw illustration, and a script-led version. Same research, several ways to explore it. These are concepts to review. Check the wording, artwork and production files before you use a design. That's the advantage of owning this app. Test the whole journey yourself. If it isn't giving you exactly what you need, explain the gap and ask Codex to improve it. You're deciding how your research becomes useful design information. You don't need a coding background to start. Get the free starter prompt, build one stage at a time, and test each step yourself. Watch the research episode first, then use its files as the starting point. Visit AI POD Builder for the prompts and more builds. For the free newsletter, go to AI POD Builder dot com slash newsletter, or choose Newsletter on the website. Subscribe on YouTube and follow AI POD Builder on X for more builds. Keep building your AI POD future.