AI is useful when messy information needs to become structured data. Notes, book lists, resource cards, project metadata, image paths, titles, summaries, and categories can all be cleaned up faster when the structure is clear.
The rule is simple: once the schema works, preserve it. AI should help fill, clean, sort, compare, and review the data. It should not redesign the structure every time you add one more entry.
A static site can still use structured data. LaunchShell uses plain HTML, shared CSS, local assets, small JavaScript helpers, and JSON data files. That keeps the site simple while making repeated content easier to maintain.
A few cards can be written by hand. But once a site has many guides, projects, resources, book entries, image files, summaries, categories, and links, repeated manual edits become harder to track.
Small inconsistencies start to matter: one image path is wrong, one category name is different, one summary is too long, or one object is missing a required field.
Structured data keeps repeated information in predictable fields. AI can help clean that data, but only if the project has a stable schema and clear rules.
That turns AI into a cleanup assistant instead of an architect that redesigns the data model every time.
Structured data is not only a backend idea. Even a static site benefits when repeated content has a clean source of truth.
Use AI after you know what the data should look like. The workflow should move from messy input to stable, reviewed output.
Decide what each object needs: title, URL, image, category, summary, type, level, time, tags, or other metadata.
Show AI a correct object before asking it to add, clean, reorder, or normalize more data.
Tell AI not to rename keys, add fields, nest objects, or redesign the structure unless you explicitly request it.
Check JSON syntax, preview the rendered page, inspect paths, and confirm the public data is safe to publish.
AI works well with repeated data when the desired output is clear and reviewable.
Titles, summaries, page URLs, image paths, categories, levels, time estimates, and homepage sections can be normalized into consistent objects.
A rough reading list can become public data with title, author, topic, format, notes, and safe display fields.
Tool names, links, categories, summaries, and student-use notes can be made consistent without hand-editing every card.
Project status, difficulty, time estimate, topic, image path, and summary can be cleaned for portfolio pages and indexes.
AI can turn messy planning notes into bullets, sections, tables, JSON drafts, or README handoff blocks.
Filenames like card-ai-data-normalization.png can be checked against page
IDs, folders, and JSON references.
Before asking AI to clean data, decide the rules. Without rules, AI may “improve” the format in a way that breaks the rest of the project.
summary to description without updating the renderer.
The LaunchShell homepage can stay cleaner when repeated card information lives in
data/site-index.json. AI can help add or adjust entries, but the field names
should stay consistent.
{
"id": "ai-data-normalization",
"title": "AI Data Normalization",
"type": "Guide",
"level": "Beginner+",
"time": "45-60 min",
"category": "AI / Learning",
"summary": "Use AI to turn messy notes, card metadata, book lists, resource entries, and project data into stable structured content while preserving the schema.",
"url": "guides/ai-data-normalization/",
"image": "assets/card-ai-data-normalization.png",
"homeSection": "none",
"featured": false
}The prompt should tell AI what to preserve, what to change, and what not to invent.
Good JSON cleanup prompt:
Clean up the following JSON entries.
Rules:
- Preserve the existing field names
- Do not add new fields
- Do not remove fields
- Do not rename keys
- Do not convert strings into objects
- Keep URLs and image paths in the same style
- Keep summaries concise and public-facing
- Return valid JSON only
- Point out any entries that look incomplete or unsafeGood metadata prompt:
Here is one correct object:
[paste example]
Now convert these rough notes into matching objects:
[paste rough notes]
Rules:
- Follow the example exactly
- Keep the same keys
- Use short student-friendly summaries
- Use category names consistently
- Do not invent links or filenames if missing
- Mark unknown values as TODOAI is useful when you have human notes that are meaningful but not structured.
Rough notes often mix titles, ideas, side comments, possible links, image names, categories, and private thoughts in one place.
That is fine during brainstorming. It becomes a problem when the content needs to render on a public page.
Clean data separates fields clearly. Public summaries are polished. Private notes are removed. Missing values are marked. Links and filenames follow the project pattern.
This is where AI can save a lot of time, as long as you still review the result.
Messy note:
AI data cleanup guide - probably AI / Learning
Talk about site-index json, book recs, messy notes, preserving schema.
Image should be card-ai-data-normalization.png.
Do not make homepage card maybe branch page only.
Clean object:
{
"id": "ai-data-normalization",
"title": "AI Data Normalization",
"type": "Guide",
"level": "Beginner+",
"time": "45-60 min",
"category": "AI / Learning",
"summary": "Use AI to turn messy notes, card metadata, book lists, resource entries, and project data into stable structured content while preserving the schema.",
"url": "guides/ai-data-normalization/",
"image": "assets/card-ai-data-normalization.png",
"homeSection": "none",
"featured": false
}AI-generated data should be treated like code. It needs review before it becomes part of the public project.
Make sure the JSON is valid: commas, quotes, brackets, and braces all need to be correct.
Confirm every object has the required keys and uses the same field names.
Open the page that uses the data and confirm the cards, filters, or list items render correctly.
Remove private notes, credentials, sensitive URLs, account details, and anything not meant for public view.
# Basic JSON validation with Python
python3 -m json.tool data/site-index.json > /tmp/site-index-check.json
# If no error appears, the JSON syntax is validMost AI data mistakes come from letting the model be too creative with structure.
Do not let AI change field names, nest simple values, or invent a new structure unless you are intentionally redesigning the data model.
AI may invent image names or URLs that sound correct. Check that the actual files and folders exist.
Brainstorming notes may include private context. Public JSON should contain only safe, reviewed display content.
AI is most useful after you already know the shape of the data. That is why this guide comes after project planning and project building.
Data cleanup often starts from private notes. Review carefully before publishing.
Do not publish API keys, tokens, passwords, private URLs, private hostnames, private logs, unrevised class material, or account details.
If data starts from private exports, trim it down to the public fields needed for the page. Do not publish the raw source history.
Keep useful summaries, categories, safe project notes, public links, image paths, and explanations that help someone else learn.
AI is excellent at cleaning repeated data after the structure is clear. Define the schema, give examples, preserve the fields, validate the output, and review the public content. The data should become cleaner, not more complicated.