protologue

SelfCheckGPT

Also called sampling-based hallucination detection, consistency-based hallucination detection.

SelfCheckGPT detects hallucinations by sampling several responses to the same prompt and checking whether each claim in the main response is supported by the others, on the principle that facts the model knows are reproduced consistently while fabrications vary.

Description

It needs no external knowledge base and works with black-box models, which makes it a practical first check. Its limitation is that consistent errors, which the model repeats confidently every time, are not caught, so it complements rather than replaces retrieval-based fact-checking.

Sources

  1. Manakul et al. (2023). SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models.

Cite this entry

Protologue. (2026). SelfCheckGPT. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0122). https://protologue.com/t/selfcheckgpt/

BibTeX
@misc{protologue_selfcheckgpt,
  title = {SelfCheckGPT},
  author = {{Protologue}},
  year = {2026},
  howpublished = {Protologue: A Taxonomy of Prompting and LLM Techniques, v1.4.0},
  note = {Entry PTL-0122},
  url = {https://protologue.com/t/selfcheckgpt/}
}

Markdown JSON