How to use your prompts

Claude

Fits

Bottom lineFits comfortably — no action needed.

1. Open a Project, go to Project Instructions (or Project Knowledge), and paste the full prompt.
2. Claude Projects share a 200,000-token context window across instructions, uploaded files, and chat history — a single prompt uses only a small fraction of that.
3. No Project? Paste the prompt as your first message in a regular chat — it works identically for that conversation.

Ollama

Fits

Bottom lineFits comfortably — no action needed.

1. Add the prompt as the SYSTEM instruction in a Modelfile, or paste it as the system prompt in your preferred Ollama front-end.
2. Since Ollama runs entirely on your own machine, there's no publisher-imposed field limit — the only ceiling is the context window of whichever model you've pulled (commonly 8K–128K+ tokens, i.e. plenty of room).

Grok

Usually fits

Bottom lineFits for most files; a few of the longest may need the first-message method.

1. Go to Settings → Personalization → Custom Instructions and paste the prompt.
2. The current limit is 12,000 characters. If your specific prompt is longer, paste it as your first message in a new conversation instead — the full text still works, it just won't persist automatically across new chats.

If the 12,000-character limit ever changes, the first-message method works as a reliable fallback.

Perplexity

First message

Bottom lineWill likely truncate — use the first-message method.

1. Perplexity Projects cap Instructions at 8,000 characters, which is shorter than nearly every prompt in the library.
2. Paste the prompt as your first message in a new Perplexity thread instead of the Project Instructions field.

ChatGPT

First message

Bottom lineWill truncate — use the first-message method.

1. None of ChatGPT's persistent fields reliably fit these prompts: the Custom Instructions box (~5,000 characters), a Custom GPT's Instructions field (~8,000 characters), or a Project's Instructions field (~8,000 characters).
2. Paste the full prompt as your first message in a regular ChatGPT conversation. It will use the whole prompt for that conversation; start a new chat with the same first message whenever you want a fresh session.

Gemini

Paste & check

Bottom lineWorks well — paste it and check.

1. Create a Gem and paste the prompt into its instructions.
2. Google has not published a hard character limit for Gem instructions; their own guidance recommends 500–2,000 characters for best results, but doesn't state where a hard cutoff (if any) actually is.
3. If the Gem seems to ignore parts of the prompt, fall back to pasting it as your first message in a regular Gemini chat instead.

Microsoft Copilot

Paste & check

Bottom linePaste it directly — use the fallback if anything's skipped.

1. Go to Copilot (copilot.microsoft.com or the Copilot app) → Settings → Personalization → Custom Instructions, and paste the prompt there.
2. Microsoft hasn't published a character limit for this field.
3. If the specialist's behavior seems incomplete, or parts of the prompt seem to be ignored, fall back to pasting the full prompt as your first message in a new chat instead — this works regardless of the Custom Instructions field's actual limit.

This is the everyday Copilot chat, not Copilot Studio's custom "agents" feature (a separate Microsoft 365 tool for building your own bots), which has an official 8,000-character instructions limit — if you're building an agent rather than just chatting, use that number instead.

DeepSeek

First message only

Bottom lineNo persistent field exists — first-message method only.

1. DeepSeek's chat app currently has no custom-instructions or persistent system-prompt feature at all (this is a known, requested-but-unbuilt feature, not a length issue).
2. Paste the full prompt as your first message at the start of every new conversation.

Kimi

First message only

Bottom lineNo persistent field exists — first-message method only.

1. Kimi's consumer chat interface does not currently offer a persona or custom-instructions setting for everyday users.
2. Paste the full prompt as your first message at the start of every new conversation.

Claude — Projects

Team ready

Bottom lineCreate one Project per company; every chat inside it shares the same files and instructions.

1. Create a Project and upload your company's background (brand info, product details, prior work) to its Knowledge Base.
2. Keep Project Instructions light and neutral — not a specific specialist's prompt, since that field applies to every chat in the Project, not just one.
3. Start a new chat per specialist, paste that specialist's full prompt as the first message, and rename the chat with the specialist's name (e.g. "Marketing Strategist").
4. Every chat in the Project automatically pulls from the same company knowledge, without you repeating it per specialist.

ChatGPT — Projects

Team ready

Bottom lineWorks the same way as Claude's Projects.

1. Create a Project, upload your company files, and set light Project-level Instructions — these override your global Custom Instructions and apply to every chat in the Project.
2. One chat per specialist, prompt pasted as the first message, all of them drawing on the same shared files.

Grok — Workspaces + Custom Agents

Team ready

Bottom lineTwo options — Workspaces for any number of specialists, or Custom Agents for a fixed roster of up to 4.

1. Workspaces: create one, upload your company's reference files, and set workspace-level instructions (these override your global Custom Instructions). Start a new chat per specialist inside it, pasting the prompt as the first message — the same shared-files pattern as Claude and ChatGPT, with no limit on how many specialists you add.
2. Custom Agents: Grok separately lets you save up to 4 persistent named agents account-wide, each with its own instructions — closer to "always-on personas" than Workspaces. Each agent's instructions are capped at roughly 4,000 characters, though, which is shorter than many specialist prompts — check a prompt's length before relying on this slot, and use Workspaces instead for any specialist that doesn't fit.

Perplexity — Projects

Team ready

Bottom lineSame shared-files-and-instructions pattern as Claude and ChatGPT.

1. Create a Project, upload your company documents, and set Project-level instructions.
2. Start a separate thread per specialist inside it, pasting the specialist's prompt as the first message.
3. One difference from Claude/ChatGPT: Perplexity doesn't carry conversation content between threads automatically — each thread still starts fresh beyond the shared files and instructions, so if one specialist's output needs to inform another, copy it across by hand.

Microsoft Copilot — Notebooks

Team ready

Bottom lineAvailable even on individual (non-business) Microsoft 365 plans, not just enterprise accounts.

1. Create a Notebook and add reference sources (Word, Excel, PowerPoint files, Copilot Pages) — Copilot answers only from what's curated in that Notebook.
2. Set notebook-level instructions for tone and format.
3. One conversation per specialist inside the Notebook, with the prompt pasted at the start.
4. Requires a SharePoint or OneDrive license, which is included with Personal and Family plans — so most individual subscribers already qualify without upgrading.

Gemini — no shared-workspace feature yet

Partial · workaround

Bottom lineEach Gem stands alone — there's no Claude/ChatGPT-style shared company folder across Gems.

1. Every Gem has its own separate instructions and its own separately uploaded files; nothing carries over between Gems automatically.
2. Workaround: put your company documents in one Google Drive folder, then link that same Drive folder to each specialist's Gem individually when you create it. It's a few extra clicks per Gem instead of one shared upload, but it gets every specialist reading from the same source material.

DeepSeek and Kimi — no team feature

No team feature

Bottom lineNeither platform has a persistent-instructions or shared-workspace feature at all right now — the same limitation that applies to a single specialist (see the Specialists guide) rules out grouping them, too.

1. The only option is to paste both your company's background and the specialist's prompt together at the top of every new chat, every time.

Ollama — Build it yourself

DIY

Bottom lineNo vendor "Projects" feature, because Ollama is self-hosted — but for the same reason, it's the most flexible of all nine platforms if you're comfortable with basic setup.

1. Run each specialist as its own named local model, with that specialist's prompt as its SYSTEM instruction.
2. A community front-end such as Open WebUI adds shared "workspace" folders and file-grounding across your local models, closer to the Projects/Notebooks pattern above.
3. For specialists that should actually hand work to each other automatically rather than you copying between them, a lightweight local orchestration setup (a short script, or a framework like AutoGen or CrewAI) can route tasks between them — this is a real technical project rather than a UI feature, worth it mainly if you or someone you work with is comfortable setting it up.