How these leaderboards are built

Every question is asked to ChatGPT, Claude, Gemini and Grok through their standard consumer products — the same interfaces real users ask — with no custom system prompts beyond the formatting instructions below. What you see is what a user asking that model would be told.

Each model gets the identical prompt, which forces a ranked answer with reasons:

Rank the top 5 <category> for <audience>, best first.
For each, one line:
<rank>. <product> || WHY: <its real strengths> || FIX: <the one change that
would most improve its ranking>
Then one MISSED: line — a product you'd add and why.
Real product, brand, or company names only. No commentary.

The WHYis the model's stated reason for the rank. The FIXis what that model says would most improve the product's standing — shown on every leaderboard as “to rank higher” / “to stay #1”.

A product ranked r by a model earns 6 − rpoints (1st = 5 … 5th = 1), summed across all models that answered. Ties break by how many models mention the product at all. Name variants (“LangSmith” vs “LangChain LangSmith”) are merged before scoring; template echoes and non-products are discarded.

Developer-tool categories rank challengers— established incumbents (AWS, GitHub, Datadog…) are listed separately rather than scored, because “the models recommend AWS” is not news.

Every poll is stored with its date. When a question is re-polled, the leaderboard shows movement against the previous poll — who rose, who fell, who entered. Ranks shift as models update; that movement is the point.

• Model answers vary run to run. A leaderboard is a dated sample of each model's answer, not a physical measurement.
• This measures what AI models tell users — their perception, shaped by training data and search — not independent product testing.
• Current rankings were collected June–July 2026; each page shows its own poll date.

If your product is ranked here, the full per-model breakdown — every reason, every fix, and movement over time — exists for your category. DM @modelsagree on X for the complete report.