8 min read · 2026-04-29
How to Make Prompts Work Across Any LLM
Write prompts that are portable across ChatGPT, Claude, Gemini, local models, and future tools.
Portability starts with plain language
A portable prompt avoids product-specific tricks. It uses clear markdown, explicit roles, useful context, and concrete output formats. That makes it understandable to any capable language model.
Model-specific features can be useful, but they should not be required for the prompt to make sense. If the core instruction is clear, the prompt can travel.
Use markdown as the interface
Headings, bullet lists, numbered steps, tables, and bracketed variables are the closest thing to a universal prompt interface. They are readable by humans and models. They also make prompts easier to save, edit, and share.
A model-agnostic prompt should look like a small document. It should not depend on hidden settings or a particular chat product.
Separate task from tool
Do not write prompts that say use ChatGPT to do this or answer like Claude unless the tool itself is the topic. Write the job instead: analyze, rewrite, compare, summarize, draft, classify, generate, critique, or plan.
When the task is clear, any model can attempt it. You can then compare outputs based on quality rather than whether the prompt was tied to a brand.
Expect different strengths
Different LLMs may handle the same prompt differently. One may be better at structured writing, another at concise summaries, another at code, another at creative exploration. Portable prompts let you test those strengths without rewriting the task every time.
If consistency matters, tighten the format. If creativity matters, loosen the constraints and ask for multiple directions.
Build prompts for the future
The model landscape changes quickly. A durable prompt library should not be locked to today's tool names. It should preserve the thinking pattern: context, task, constraints, output, review.
That is the philosophy behind SixtySevenSites: prompts as reusable work instructions, not platform-specific incantations.
Field notes
Comparing LLMs fairly starts with a stable task. Use the same prompt, same source material, same constraints, and same scoring rubric. Otherwise you are comparing prompt drift instead of model behavior.
Different models have different strengths. Some are better at long-form synthesis, some at coding, some at concise rewriting, and some at following strict formatting. A comparison should measure the job you actually need done, not an abstract idea of intelligence.
Use rubrics instead of vibes. Score outputs for accuracy, completeness, formatting, source fidelity, practical usefulness, risk handling, and whether the answer follows the requested constraints. This makes the comparison repeatable.
For business workflows, test prompts with real source material. Synthetic examples hide the edge cases that matter: messy notes, incomplete requirements, conflicting constraints, and ambiguous audience information.
The most useful result of a model comparison is often a prompt adjustment. If one model fails because an instruction is ambiguous, rewrite the prompt so all models receive clearer context and output requirements.
How this connects to the library
This guide is supported by related prompt categories such as Research, Data Analysis, Coding & Debugging, Technical Writing. Those categories turn the article ideas into reusable prompts, so readers can move from explanation to execution without opening a blank chat.
The strongest workflow is to read the guide once, choose the closest prompt card, paste real context into the bracketed variables, and then ask the model for a critique pass before using the output. That pattern keeps the answer grounded, editable, and easier to trust.
Use the article for judgment and the prompt cards for repetition. The article explains what good looks like; the prompts make that standard easy to apply across new projects, teams, audiences, and tools.
For best results, save the prompt that matches your recurring workflow and improve it after each real use. Add the context that produced the strongest answer, remove instructions that created noise, and keep a short note about when the prompt should not be used.
Useful prompts from the library
These examples connect the article to copy-paste prompts you can use immediately. Each card opens the full prompt page with more context, customization notes, and related prompts.
You are an expert research strategist. Help me create a strategy map for [project / audience / offer]. Context: [describe the goal, audience, constraints, examples, and what has already been tried]. Output format: give me a concise recommendation, then a structured draft I can copy, then 3 improvement ideas. Keep it specific, practical, and avoid generic advice.
You are an expert data analysis strategist. Help me create a strategy map for [project / audience / offer]. Context: [describe the goal, audience, constraints, examples, and what has already been tried]. Output format: give me a concise recommendation, then a structured draft I can copy, then 3 improvement ideas. Keep it specific, practical, and avoid generic advice.
You are an expert coding & debugging strategist. Help me create a strategy map for [project / audience / offer]. Context: [describe the goal, audience, constraints, examples, and what has already been tried]. Output format: give me a concise recommendation, then a structured draft I can copy, then 3 improvement ideas. Keep it specific, practical, and avoid generic advice.
You are an expert technical writing strategist. Help me create a strategy map for [project / audience / offer]. Context: [describe the goal, audience, constraints, examples, and what has already been tried]. Output format: give me a concise recommendation, then a structured draft I can copy, then 3 improvement ideas. Keep it specific, practical, and avoid generic advice.
Implementation checklist
- Use one stable prompt.
- Use the same source material.
- Score with a rubric.
- Test realistic messy inputs.
- Record model-specific failure modes.
- Turn findings into prompt improvements.