# Project 64: Test AI project generation prompts

## Prompt quality

#### Paste the exact custom prompts you used for all 3 projects.

1. Please generate a new API-related project using the `kitchen.yaml` file. The project should be built around this file.
2. Create a project for my restaurant's kitchen. I want to be able to integrate with food delivery tech, so I need to have API docs.
3. Could you create a project for Airbnb based on the currently attached resources? The project should reproduce Airbnb's philosophy.


#### Which prompt was easiest for the AI to interpret, and why?

- The project created with the second prompt was the best fit because it was the most detailed and included attached resources.


#### Which prompt was too vague, too broad, or missing important context?

- The Airbnb project was the worst. We only linked the Airbnb site, using the same resources as the other projects.


#### Which prompt details most clearly influenced the generated project?

- The first prompt.


## Output quality

#### Did the generated project feel usable as a first draft?

- The docs were generated well, but the API Reference tab with the OpenAPI spec is not working.


#### Was the structure clear and appropriate for the intended audience?

- Yes.


#### Was the generated content accurate, specific, and free of obvious filler?

- Project 2: Yes. The generated content is close to the description in the prompt.


#### What would you need to fix before sharing the project with someone else?

- Project 1 had broken links, which caused the AI build to fail.


## Feature controls

#### Did Add features feel predictable?

- It was not clear to us. Adding the word "Airbnb" enabled all features.


#### Were feature suggestions relevant to the typed prompt?

- Not sure.


#### Did adding and removing features behave as expected?

- Yes.


#### Which features did you choose for each project?

- OpenAPI Reference


## Resource controls

#### Did Add resources clearly separate links, OpenAPI files, logos, images, and Markdown files?

- Yes.


#### Did the generated output visibly reflect your uploaded resources and links?

- Not for Project 3.


#### Did the OpenAPI file produce a useful API reference or API-focused structure?

- Yes.


#### Did the brand challenge use the chosen brand color palette?

- No. It created a correct logo, but the overall styles did not match the brand.


## Overall

#### What made you smile?

- Project 2: After adding "redocly.com" as a link, it created similar sidebar navigation.
- All projects follow the same UI structure.


#### What felt confusing or slow?

- The first build runs with empty resources and always fails.
- After a project is generated, the message says "Build complete.", but the build is still in progress on the Deployments page.


#### What should we change before wider rollout?

- Fix the first build that always fails.