KnitKnot

# Engines

The AI engines KnitKnot supports, how to choose those available on your plan, and what a captured response records.

KnitKnot supports four AI engines for buyer-question benchmarks. A run uses the engines you select from those available on your plan. Each returned answer is kept for review with its engine and available model context.

## The four engines

  • - **ChatGPT:** OpenAI's assistant.
  • - **Claude:** Anthropic's assistant.
  • - **Perplexity:** an answer engine with web search and citations.
  • - **Gemini:** Google's assistant.

Choose the engines when you [trigger a run](/docs/run-and-schedule-benchmarks/). A locked engine is marked **Paid** in the picker. Compare engine-level results only for the engines that actually ran and returned scored answers.

## Real responses, captured verbatim

Each question produces a captured benchmark response from the selected engine, with returned citations and model version stored beside it. Consumer chat sessions can differ because of personalization, geography, model version, and execution path. When a report says you lost a comparison on Gemini, you can open the captured Gemini response and inspect the supporting record.

## Model version on every response

AI models change often, and a model update can move answers even when your content stays the same. KnitKnot records available model-version context beside the captured response. Check it when comparing two runs. A version change is one possible explanation for movement; it does not prove the cause of a specific changed answer.

## Engine-by-engine in your report

Because scoring is per-engine, your report breaks results down by engine as well as in aggregate. Use it to spot divergence : a competitor that only beats you on one engine, a factual error that appears on Perplexity but not ChatGPT : and to prioritize. A gap on the engine your buyers use most is worth more than the same gap on one they don't.

Docs navigation
Docs › Reference

Engines

The AI engines KnitKnot supports, how to choose those available on your plan, and what a captured response records.

Updated

KnitKnot supports four AI engines for buyer-question benchmarks. A run uses the engines you select from those available on your plan. Each returned answer is kept for review with its engine and available model context.

The four engines

  • ChatGPT: OpenAI’s assistant.
  • Claude: Anthropic’s assistant.
  • Perplexity: an answer engine with web search and citations.
  • Gemini: Google’s assistant.

Choose the engines when you trigger a run. A locked engine is marked Paid in the picker. Compare engine-level results only for the engines that actually ran and returned scored answers.

Real responses, captured verbatim

Each question produces a captured benchmark response from the selected engine, with returned citations and model version stored beside it. Consumer chat sessions can differ because of personalization, geography, model version, and execution path. When a report says you lost a comparison on Gemini, you can open the captured Gemini response and inspect the supporting record.

Model version on every response

AI models change often, and a model update can move answers even when your content stays the same. KnitKnot records available model-version context beside the captured response. Check it when comparing two runs. A version change is one possible explanation for movement; it does not prove the cause of a specific changed answer.

Engine-by-engine in your report

Because scoring is per-engine, your report breaks results down by engine as well as in aggregate. Use it to spot divergence : a competitor that only beats you on one engine, a factual error that appears on Perplexity but not ChatGPT : and to prioritize. A gap on the engine your buyers use most is worth more than the same gap on one they don’t.