Find your controls

1. Choose your model

   Open the AI model menu at the top of the chat. Choose a model that supports the task and inputs you need, such as text, files, or images.

2. Open the Quick menu

   Select the + control (Add files & more) beside the message box to find modes, search settings, reasoning or creativity, answer detail, and output language. More tasks opens Saved settings so you can start from a ready-made setup.

3. Add context, then send

   Use the add control to attach relevant files or select reusable context. Describe your goal, audience, and desired format in the message, review your settings, and send.

Available controls depend on the selected model, mode, and your account. Some options require signing in or an eligible plan. If a control is missing, check the model and mode instead of assuming it is enabled.

Start with your use case

1. Understand something quickly

   Start with Explain simply. For a current fact, keep web search on and choose Fast with Brief answer detail. For a question about text you already supplied, web search may not be needed. Try: “Explain this term in three bullets and give one everyday example.”

2. Write or rewrite for your audience

   Start with Reply to a message, Rewrite or translate, or Create a social post. Add a Brand Voice when style matters and choose the output language. Try: “Rewrite this email for a new customer. Keep a warm, direct tone, stay under 150 words, and do not add new claims.”

3. Research a decision

   Start with Help me choose. Use Deep when you need broader research, set the goal to Compare, and narrow sources to the relevant region and dates. Try: “Compare these three tools for a five-person team. Cite current pricing, explain the tradeoffs, and flag anything you cannot verify.”

4. Debug or review code

   Choose a model suited to coding, provide a small reproducible example, and use Medium or High reasoning when available for a multi-step issue. Search official documentation if version-specific behavior matters. Try: “Explain this failure, propose the smallest fix, and include a regression test. State any assumptions.”

These are starting points, not guaranteed outcomes. Review the selected settings and edit the suggested prompt before sending. Remove secrets and private data from examples you share.

Choose how much work the answer needs

• Mode: use Search for everyday questions with web evidence, Research for more rigorous source-driven work, or Creative for drafting and ideation. Creative starts from prior knowledge and may still search when web search is enabled.
• Search and research depth: Fast and Deep control the breadth of the search or research work. Start with Fast for a focused question; choose Deep when the task needs more investigation. Other modes may also offer Balanced. Deeper work can take longer and use more credits.
• Reasoning Level: Low, Medium, and High adjust reasoning effort on supported models. Use a lower level for straightforward tasks and a higher level for difficult analysis, planning, or debugging. More reasoning does not guarantee a correct answer.
• Creativity Level: when this slider is available, move toward Deterministic for more constrained wording or toward Creative for more varied ideas. Creativity is not a fact-checking control. Reasoning models may show Reasoning Level instead of this slider.
• Answer detail: Automatic, Brief, and Detailed control how much explanation you receive where supported. A brief answer can still need deep research; a detailed answer does not necessarily need it. In writing modes, request the length and structure directly in your prompt.

Do not turn every setting up by default. Match research depth to the evidence you need, reasoning to the difficulty, and answer detail to how much you want to read.

Focus the evidence, not just the wording

Use web search when the answer depends on current information or external evidence. For rewriting, brainstorming from supplied material, or summarizing your own text, you can turn web search off when available. This does not make the model infallible or turn the chat into an offline workspace.

• Search engine selects the retrieval provider, such as Perplexity, Google, or Bing, where available. It is separate from the AI model that writes the answer. Start with the default and compare another engine if the results miss useful sources.
• Choose a research goal such as Answer, Explore, Verify, Compare, Challenge, or Timeline to explain the kind of result you need.
• Set a date range for time-sensitive questions. Broaden it if useful background or older evidence is being excluded.
• Use Primary sources only or include specific domains when you want official documentation, original research, or first-party announcements. Exclude domains that are not useful for the task.
• Set the search region and search language to focus retrieval. Output language separately controls the language of the answer.
• Add Academic or Social sources when relevant and available. Treat social posts as individual reports, not representative market research.

Filters guide retrieval; they do not prove that a claim is true. Open important citations and check the source, date, and supporting passage. If results are sparse, relax one filter at a time.

Give the answer your context and style

• Attach a relevant file for a one-off task. Tell the AI which part to use and what it should extract, compare, or summarize.
• Select an Infobase for reusable source material. Keep the selection focused on the current task.
• Use a Skill for reusable instructions about how to approach the work, a saved prompt for a repeatable request, and Brand Voice for consistent writing style.
• Choose Output language for the response, then specify audience, tone, length, and format in your prompt. Include an example when the format must be precise.

Try: “Using the attached brief, write a launch email for existing customers. Use our Brand Voice, keep it under 180 words, include the three benefits in the brief, and end with one call to action. Flag missing details instead of inventing them.”

Save a setup you want to reuse

1. Test the setup first

   Run a representative task and check whether the model, research settings, detail, and language fit your needs.

2. Save settings

   Open Saved settings in the Quick menu. Save settings updates the selected set; if Default is selected, it asks for a name. Choose a name that describes the job, such as “Weekly competitor brief”.

3. Review it before reuse

   Choose your saved settings for the next task and check the active model and options. Add the current prompt, files, Skill, or Brand Voice separately; saved settings are not a saved conversation or a bundle of attachments.

Choose Default in Saved settings when you want a starting point again. Switching saved settings does not erase earlier conversation context. Start a new chat for an unrelated task.

Refine one thing at a time

• Too long? Choose Brief where available, or ask for a word limit and a specific structure.
• Too generic? Add the audience, constraints, source material, and an example of the result you want.
• Outdated or weakly supported? Enable search, adjust dates and sources, and ask the AI to separate evidence from assumptions.
• Missing important tradeoffs? Try Compare or Challenge, supply your decision criteria, and increase depth or reasoning only if needed.
• Taking too long? Try Fast and lower reasoning for a simpler task, or split a broad request into focused questions.

Change one control, send a follow-up, and compare the result. Review important facts, calculations, citations, and code before using an answer in real work.