Update an AI website with content changes from detect-changes API.
Uses THREE parallel Gemini calls (same pattern as onboarding):
- Update
llms.txt
- Update homepage + Q&A pages
- Update
data.json (Schema.org)
This is the second step in the 3-API update flow:
detect-changes - Find what changed- update-ai-site (this endpoint) - Update the AI website
discover-products-from-changes - Find new products
How It Works
1. Fetch current AI site files (llms.txt, homepage HTML, data.json)
2. Run 3 PARALLEL Gemini calls:
- llms.txt: "Here's current + changes, update if needed"
- Q&A pages: "Here's current + changes, update if needed"
- data.json: "Here's current + changes, update if needed"
3. Check if any product source URLs changed → regenerate affected product llms
4. Generate files via generate_ai_site (only new + changed replica pages)
5. Deploy to Vercel
6. Store updated site_map + page_hashes
7. Return changed URLs for IndexNow submission
Product LLMs Regeneration
When pages change, this endpoint checks if any existing products have product_source_urls that overlap with the changed URLs. If so, those product llms files are regenerated.
Products in DB:
- Sandals: product_source_urls = ["/products/sandals", "/categories/footwear"]
- Boots: product_source_urls = ["/products/boots", "/winter-collection"]
Changed URLs from detect_changes:
- ["/products/sandals", "/about"]
Result:
- Sandals /llms/sandals.txt → REGENERATE (overlap: /products/sandals)
- Boots /llms/boots.txt → SKIP (no overlap)
Efficient updates: Product llms files are only regenerated if their source content changed. This scales well for sites with hundreds of products.
Partial content limitation: When regenerating product llms, we only use content from the changed pages. If a product has 3 source URLs and only 1 changed, the regenerated file will be based on that 1 page’s content. For a complete regeneration with all source content, use regenerate-fresh-website.
Request Body
Takes the output from detect-changes directly:
| Field | Type | Required | Description |
|---|
business_id | string | Yes | Clerk org ID |
new_pages | array | Yes | Pages that didn’t exist before (have markdown) |
changed_pages | array | Yes | Pages with content changes (have markdown) |
removed_urls | array | Yes | URLs that no longer exist |
unchanged_pages | array | Yes | Unchanged pages (NO markdown) |
updated_site_map | array | Yes | Current URL list |
updated_hashes | object | Yes | Current hash map |
business_info | object | Yes | Business entity and AI site info |
Key difference from old API: unchanged_pages replaces all_pages.
Unchanged pages don’t have markdown because they weren’t batch scraped.
Response
{
"status": "success",
"files_updated": 42,
"product_llms_regenerated": 2,
"deployment_url": "https://example.searchcompany.dev",
"project_name": "example-searchcompany-dev",
"source_url": "https://example.com",
"changed_urls": [
"https://example.com/new-page",
"https://example.com/about"
]
}
changed_urls array should be submitted to IndexNow after deployment.
Three Parallel Gemini Calls
Each call receives the current file + changes and decides what to update:
1. Update llms.txt
Current llms.txt: [current content]
Changes: [new/changed/removed pages]
Task: Update if needed, preserve Boosted sections
2. Update Q&A Pages
Current homepage HTML: [current structure]
Changes: [new/changed/removed pages]
Task: Update homepage + Q&A pages if needed, preserve Boosted sections
3. Update data.json
Current Schema.org: [current JSON-LD]
Changes: [new/changed/removed pages]
Task: Update if needed
Boosted sections are preserved - The Gemini prompts explicitly instruct NOT to modify any “Boosted” sections, as those are managed separately by the boosted pages cron.
Database Updates
ai_sites.page_hashes - Updated hash mapai_sites.site_map - Updated URL listai_sites.last_content_check_at - Timestamp
External API Calls
- Gemini 3 Flash Preview (x3 parallel) - Content updates
- Gemini 3 Flash Preview (per affected product) - Product llms regeneration
- Vercel API - Deployment
