Artifacts
Generate structured documents directly from chat: three modes, versioning, inline editing, and export as PDF or Word.
Introduction
Artifacts on the meinGPT platform let you create structured documents like guides, templates, emails and more directly from the chat. These documents open in a separate area to the right of the chat where you can edit and export them.
With Artifacts, you can copy the generated content, download it as PDF or DOCX, and refine it across multiple versions. Especially useful for precise tasks where structured output matters.
The three Artifact modes
Per chat, you can control how aggressively the model should produce Artifacts. The mode selector lives in the chat input area (see Chat overview).
| Mode | Behavior |
|---|---|
| Auto | The model decides on its own whether an Artifact fits. Typical triggers: email drafts, longer texts, reports, structured lists. Short answers stay in chat. |
| Always | The model is instructed to produce an Artifact. Not a 100-percent guarantee, since the model sometimes ignores the instruction, but in most cases an Artifact will be created. |
| Off | No Artifact instruction. Responses come back as normal chat messages. |
You can set the default mode for new chats in your user settings.
Which models produce Artifacts reliably
The Artifact concept was originally introduced by Anthropic (Claude). That has a practical consequence:
| Model family | Artifact reliability | Reason |
|---|---|---|
| Claude (all variants) | Very high | Anthropic developed the concept and trained it into the models. |
| GPT (all variants) | Lower | The model is less familiar with the Artifact convention and ignores it more often, even in Always mode. |
If you want reliable Artifacts, pick a Claude model. With GPT, the same prompt may produce an Artifact one time and a normal chat reply the next. If you run into trouble, phrase the request more explicitly, e.g. "Create a document as an artifact with the following sections ...".
Step by step
The AI automatically detects when a structured document should be created and generates a matching Artifact. An Artifact can cover many use cases, e.g. a detailed guide, a custom email template, or a structured report.
To revise specific sections of an Artifact, select the text and either type your own improvement suggestion or pick one of the prebuilt suggestions. The AI then produces a new version with the chosen changes.
Particularly useful when adjusting writing style, clarity, or structure.
Every revision is saved as a new version. You can switch between versions, restore an earlier one, or compare them to track how the document evolved.
Once the Artifact is final, you can download it as PDF or DOCX, or simply copy the content to reuse it elsewhere.
Artifacts are stored in the chat automatically and remain available there permanently. Every version is archived in the chat history, so you can revisit earlier versions and their changes any time.
Inline editing regenerates the whole Artifact, not just the selected passage. When you select a sentence and ask the model to change it, the model often rebuilds the entire document. Use the version control to roll back quickly if the regenerated version is worse than the original.
Branding and export limitations
The branding template (logo, footer, header) that workspace admins set under Settings is currently only applied to document output in Workflows, not to Artifact exports in chat. If you need a customer-ready document with company branding, build a workflow with the document output toggle enabled.
For very long documents or complex layouts with images and tables, the PDF/DOCX export is useful. For structured data outputs (Excel, CSV, JSON) or PowerPoint slides, Workflows are the better choice.