LLM Integration

Design reliable card-price assistants with grounded retrieval and citation-safe outputs.

GEO Setup Checklist (2026)

Allow search/user AI crawlers in robots.txt and keep private routes disallowed.
Publish and maintain `llms.txt` plus a clean XML sitemap index.
Use answer-first sections with explicit market/timestamp evidence.
Keep technical docs update cadence visible (`last updated`, changelog links).

Recommended Retrieval Flow

Classify user intent (`lookup`, `compare`, `trend`, `grade spread`).
Resolve entity with `/v1/cards/search` using strict filters where possible.
Fetch exact card with `/v1/cards/:id` and optional history endpoint.
Ask the model to summarize only retrieved fields, never infer prices.

Grounding Template

System: Answer only from tool data.
If data is missing, say "not available".
Always include source fields used (market, condition, timestamp).
Never output fabricated prices.

Caching + Freshness

Cache resolved `card_id` by normalized query for fast repeated lookups.
Cache card detail responses with a short TTL (for example 5-15 minutes).
Display timestamp context when presenting trends or price changes.

Bot Coverage

Keep these agent families allowed for discovery: OpenAI (`OAI-SearchBot`, `ChatGPT-User`), Anthropic (`ClaudeBot`, `Claude-SearchBot`, `Claude-User`), and Perplexity (`PerplexityBot`, `Perplexity-User`).