A provider is one of the three large language model platforms BCited treats as a queryable answer engine. Each tracked prompt is dispatched to all three so you can compare citation behavior across them — they often disagree.
The three
- OpenAI — ChatGPT (
gpt-4.1-minifor cost,gpt-4.1when budget allows). Largest user base. Cites web sources inconsistently depending on whether browsing is invoked. - Anthropic — Claude (
claude-sonnet-4.5). Cites more consistently via web search when the query benefits from it. - Perplexity —
sonar/sonar-pro. Built around citation by default — every answer comes with sources. Most predictable citation behavior of the three.
Why measure all three
Citation behavior diverges significantly. A domain that gets cited consistently by Perplexity may rarely show up in ChatGPT's answers (different retrieval, different training cutoffs, different system prompts). The composite visibility score weights all three equally so you don't optimize for one and miss two.
How calls reach providers
Everything goes through Cloudflare AI Gateway for caching, retries, per-prompt analytics, and a single billing surface. The Gateway also rate-limits and rewrites — useful when one provider's API has a hiccup and you don't want it to take down a whole run.