# Prompt-format mirroring and structural priming in output > Source-anchored research for Prompt-format mirroring and structural priming in output. Run status: COMPLETED. Parent: [Instruction-file hygiene for LLM lexical tics: linting, mirroring, hook filtering](https://llms-explorer.com/tree/ai-isms-instruction-hygiene/) · 3 facets · 6 facts · page: https://llms-explorer.com/tree/prompt-format-mirroring-and-structural-priming-in-output/ ## How it works - Language models show structural priming: they reuse recently seen syntactic structures, including when prime and target share no lexical overlap, indicating abstract structural representations. (high confidence) — [source](https://aclanthology.org/2022.tacl-1.60.pdf) ## Comparisons and alternatives - Practical mitigation for instruction files: write the instruction file in the plain style wanted in the output (prose, little markdown) and state the desired format positively, rather than only listing prohibitions. (low confidence) — [source](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices) ## Facts and statements - Anthropic's prompting guide says the formatting style of a prompt may influence Claude's response style, and that removing markdown from the prompt can reduce the volume of markdown in the output. (medium confidence) (verified as of 2026-09-23) — [source](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices) - Prompt format alone can change measured performance a lot: up to 76 accuracy points on LLaMA-2-13B in few-shot settings, persisting with larger models, more shots, or instruction tuning; format performance correlates only weakly across models. (high confidence) — [source](https://arxiv.org/abs/2310.11324) - Format sensitivity falls with model strength: GPT-4 variants varied less across plain text, markdown, JSON and YAML than GPT-3.5-turbo, and no single format is best across models. (medium confidence) — [source](https://arxiv.org/html/2411.10541v1) - A 2026 study (9,649 experiments, 11 models, YAML/markdown/JSON/TOON context) found no significant aggregate accuracy effect of context format (p=0.484), though individual models, mostly open-source, showed format-specific differences. (low confidence) (verified as of 2026-02-01) — [source](https://arxiv.org/abs/2602.05447)