This page covers everyday chat skills: writing good requests, attaching and dictating content, grounding answers with web search, and steering responses.
Writing requests that work
You don't need special syntax — but specific, contextual asks produce markedly better answers.
| Technique | Example |
|---|---|
| State the role or audience | "Write as an IT manager to non-technical executives." |
| Specify format | "Return a table with columns: risk, impact, owner." |
| Give constraints | "Under 200 words. Australian English." |
| Provide context | Paste or attach the source material rather than describing it. |
| Iterate | Follow up: "Shorten. Add a cost estimate row." |
Set a default style once with System Prompt in My Settings → General — for example "Prefer concise bullet points; ask when unsure."
Selecting a model
Open the model selector above the input to see available models. Your administrator decides which appear. Models differ in strengths (reasoning, coding, multilingual, vision) and speed. Switch models mid-conversation to compare styles; see Model Comparison for side-by-side testing.
Attach files
- Drag files into the message box, or use the attach control.
- Common document, text, image, audio and video formats are supported.
- Attached content is understood for the conversation: summarise it, query it, extract from it, or compare several attachments in one request.
- Images can be captured from a camera where supported (input menu → Capture).
- Very large images are optionally compressed before sending; the compression limit is configurable in Interface settings.
Note:
For bulk document processing (entire folders), use the Document Parser or build a Knowledge Base instead of attaching one-by-one.
Voice input and dictation
Select the microphone control to dictate. Speech is transcribed into the message box; depending on your settings the message can send automatically after transcription. Voice options — transcription language, auto-send, playback — are in My Settings → Audio. Voice mode (full conversation by voice) is available where enabled by your administrator.
Web search and citations
When Web Search is enabled for the conversation (chat controls), the assistant can retrieve current information from the internet and will show sources under the answer. Use it for anything time-sensitive: prices, releases, news. In regulated, offline environments administrators disable web search entirely.
When answers use your attached files or knowledge base, source references show where available — hover or select them to see the underlying material.
Response actions
Hover a response to reveal:
- Copy — clipboard copy (formatted or plain, per your settings).
- Regenerate — produce a new answer for the same message.
- Branch / edit — change a previous message to explore a different direction without losing the original thread.
- Rate — thumbs up/down with optional comment; useful feedback for administrators' evaluations.
- Read aloud — spoken playback where enabled.
- Follow-up suggestions — some answers include suggested next questions; configure whether they appear or are inserted into the input in Interface settings.
Chat controls and Valves
The chat controls panel (top-right of a conversation) exposes per-conversation options, depending on administrator permissions:
- Capabilities — Vision, Web Search, Code Interpreter, Image Generation, Video Generation, Citations, File Upload.
- Model parameters (Valves) — Temperature, Top-P, Max tokens and similar "fine-tune the behaviour" dials; the reference page Model Parameters explains each.
- Tools / Knowledge — choose which tools or knowledge bases this conversation may use.
Thinking and progress
Models that support reasoning show their Thinking process, and tool-using answers show the steps taken (searches performed, code executed). Expand them to audit how an answer was reached.
Artifacts
When the model produces a standalone deliverable (a document, code file, diagram, HTML page), it can open as an Artifact beside the chat for preview and editing. Automatic artifact detection can be toggled in Interface settings.
Keyboard behaviour
By default Enter sends and Shift+Enter adds a line break. Switch to Ctrl+Enter to Send in Interface settings if Enter-as-newline suits your workflow better.