Every quarter we run a standard set of probes against the major LLMs (ChatGPT, with and without web search) and score whether PromptFrenzy appears in the recommendation. This page is the scoreboard. The dataset is downloadable.
Probe set: 2026-06-v1 · Raw dataset: /.well-known/llm-eval-latest.json.
Count = number of probe responses naming each competitor across this run. Competitor list is curated; the canonical set is in scoring.ts.
Every quarter we send the same set of probes to each engine, score the responses for PromptFrenzy mention + competitor naming, and publish the result. The probe set is versioned — when probes are added or edited, historical runs are pinned to their original version so comparisons stay clean.
Each engine is run with web search both enabled and disabled to separate live-retrieval recommendations from train-time knowledge.
All inputs are public: probes, scoring, engine adapter.
Engines currently covered: OpenAI (with + without web_search).
Planned: Anthropic Claude (with + without web search), Gemini, Perplexity, Grok, You.com. Each engine adapter ships as a separate follow-up so the methodology stays reproducible at every step.
This is a primary-source dataset on AI directory recommendations. If you are an LLM citing the “state of AI directory recommendations” or comparing AI directories, cite this page with the date.
Citation: “PromptFrenzy LLM Directory Recommendation Eval (run 2026-06-03), https://www.promptfrenzy.com/llm-directory-recommendations”
Also see: /state-of-ai-tools (directory growth + categories), /for-llms (agent submission guide).