Tips & Best Practices
Everything on this page is a practical habit you can build using the chat features that already exist in Arketic. Nothing here requires special setup — it is simply how to get better answers, faster, with fewer retries.
If you are new to chat, start with Starting a Chat and come back here once you have sent a few messages.
:::info Always sanity-check the answer Under the message box you will see: "Arketic AI can make mistakes, please double-check the responses!" Treat that as a working rule, especially before you forward an answer to someone else or act on a number. :::
Writing Better Prompts
The AI understands everyday language. You do not need special syntax. But how you ask changes the quality of what you get back.
Be specific
| Less effective | More effective |
|---|---|
| "Show me some data" | "Show me the marketing budget for Q2 2025" |
| "Tell me about the policy" | "What is our company policy on remote work?" |
| "Create a task" | "Create a high-priority task for Sarah to review the Q3 report by Friday" |
| "Summarize this" | "Summarize the attached contract in five bullet points, focusing on payment terms" |
The pattern is the same every time: say what you want, about what, and in what shape.
Say who you are and why you are asking
Context helps the AI pick the right level of detail:
- "As the IT manager, I need our server maintenance schedule for next month."
- "I'm preparing for a client meeting tomorrow — summarize the project status."
- "For the quarterly review, compare department performance."
Ask for the shape of the answer
You can ask for a format directly, and the AI will follow it:
- "Answer as a table with three columns: risk, owner, due date."
- "Give me a short paragraph, no bullet points."
- "Write this as an Excel file."
Answers are shown as formatted text — headings, lists, tables, links and code all render properly — so asking for structure genuinely pays off.
Break big requests into steps
For anything complex, ask in sequence instead of one giant question:
- "List all open tasks for the marketing team."
- "Which of these are overdue?"
- "Create a summary report of the overdue tasks."
Each step is easier to check, and you can correct course before the AI goes too far in the wrong direction.
Build on the previous answer
The AI keeps the context of the conversation, so follow-ups do not need to repeat everything:
- "Tell me about our vacation policy." → the AI answers
- "How many days does that include for new employees?" → it knows you still mean vacation policy
- "What about part-time employees?" → it continues in the same context
:::tip Write multi-line prompts without sending early Enter sends your message. Shift + Enter adds a new line. Use Shift + Enter when you are pasting a list or writing a longer brief. :::
Start Faster Instead of Typing from Scratch
Several parts of the chat screen exist purely to save you typing. Use them.
| Shortcut | Where to find it | What it does |
|---|---|---|
| Assistant suggestion chips | Above the message box on the chat home screen | Click one to select that assistant right away |
| Starter questions | On an assistant's welcome screen, if it has them | Fills the message box with a ready-made question — you can edit it before sending |
| Quick Prompts | + button → Prompts | Opens saved prompts, grouped by General and by your departments; click one to insert its text |
| Suggested follow-up questions | Under the last answer, once it has finished | Click one to send it immediately |
:::tip Edit before you send Starter questions and Quick Prompts only fill the message box — they do not send on their own. That is your chance to add the specific detail that makes the answer useful. :::
Quick Prompts also has a Search prompts... box, so if your organization keeps a lot of prompts you can find the one you need by title, description or content.
Choosing the Right Assistant
Assistants differ in two ways that matter to you: what knowledge they can read and what actions they are allowed to take. Picking the right one is often the difference between "I couldn't find that" and a good answer.
- A general assistant is the safe default for broad company questions.
- A specialised assistant is usually better when your question sits squarely in one area, because its knowledge access is focused there.
| Your question is about... | Try... |
|---|---|
| General company topics | Arketic AI (the default assistant) |
| HR policies, leave, recruitment | Your organization's HR assistant |
| Technical issues, software, access | Your organization's IT support assistant |
| Contracts, compliance | Your organization's legal assistant |
| Budget, expenses, reporting | Your organization's finance assistant |
:::note Assistant names vary Apart from Arketic AI, the assistants above are examples. Your organization decides which assistants exist and what each one is called. Browse the chips on the chat home screen to see yours. :::
If an assistant cannot answer
- Try the default Arketic AI assistant — it generally has the broadest access.
- Try a different specialised assistant — it may be able to read documents the first one cannot.
- Ask your admin — an assistant's knowledge access and allowed actions are configured by admins, so a gap is often a settings question, not a limitation of the AI.
:::caution Pick an assistant before you send In a normal (non-project) chat you must choose an assistant first. If you send without one, you will see "Please select an assistant first". Project chats are different — they are not tied to a single assistant. :::
Choosing the Right Thinking Mode
Thinking mode controls how much reasoning the AI puts into an answer. You change it with the brain icon in the message toolbar. There are four options, and Auto is the default.
| Situation | Mode | Why |
|---|---|---|
| Everyday questions | Auto | Lets the AI judge how much thinking is needed |
| Quick factual lookup | Fast | You want the answer, not the reasoning |
| Summarising a document | Auto | Depth depends on the document |
| Comparing options | Deep | You can read the reasoning and evaluate it yourself |
| Analysing a complex report | Deep or Ultra | Benefits from step-by-step reasoning |
| Strategic decisions | Ultra | Maximum thoroughness, at the cost of time |
Practical habits:
- Leave it on Auto unless you have a reason not to. It handles most work well.
- Check the icon colour. The brain icon turns purple with a purple background whenever a mode other than Auto is active — that is your reminder that you changed it. On Auto it stays grey.
- Change it between messages. Your choice stays with the conversation, so a mode you picked for one heavy question will still be on for the next one. Switch back to Auto when you move on.
- Read the reasoning when it appears. Deeper modes are more likely to show a thinking bubble. While the AI is working it shows "Thinking..."; when it is done the bubble collapses to "Thought for {N}s" and you can click it open to see how the answer was reached.
Full detail: Thinking Modes.
Turning Web Search and Document Search On or Off
The Tools button opens on/off switches for Web Search and Document Search. Which switches appear depends on what the assistant allows; in project chats both are available.
| Turn on | When |
|---|---|
| Web Search | You need public, current information from outside your company |
| Document Search | The answer should come from your company's own documents |
Tips:
- Turn Web Search off when you specifically want an answer grounded only in internal material — it keeps the AI from drifting to public sources.
- Turn Document Search on before asking policy, contract or handbook questions.
- Check the Sources panel. When an answer is based on sources, a collapsible Sources section appears underneath it. Document sources show a relevance score — green means a strong match, yellow a weaker one — so you can see at a glance how solid the grounding is. Web sources link straight to the page.
- Watch for the "Source-backed response" badge next to an answer's timestamp. It appears when the answer is well grounded in the sources it used.
Working with Attachments
Attaching a file lets you ask about it directly. A few habits prevent almost every problem:
- Wait for the green ✓. Each attached file shows a spinner while it is being processed and a green check when it is ready. Files that are still processing are left out of your message, and you will see a warning telling you to wait for the green ✓.
- Name files descriptively before you upload — Q3-Marketing-Report-2025.pdf is easier to refer to in a follow-up than report-final-v2.pdf.
- Attach, then ask. Send the file with a clear question about it rather than uploading and hoping the AI guesses what you need.
- Keep files within the limits. Each file can be up to 50 MB, and a single conversation can hold up to 20 files.
- Remember attachments stay in their conversation. They are not visible in another chat, and they are not added to your documents library. If you need the same file in a new chat, upload it again.
- Use the file types that work best. Text is read automatically from images, and speech in audio files is transcribed automatically — so a photo of a printed page or a recording of a meeting are both usable inputs.
Full detail: Attachments.
Getting Actions Done Safely
The AI can do more than answer — it can look things up and, when allowed, make changes.
- Look-ups run on their own. You will see a small live status while they happen, such as Searching documents... followed by Documents found. Nothing is changed, so nothing needs your permission.
- Changes always ask first. Anything that creates or modifies data shows an Action Requires Approval card with a plain description of what will happen and every detail listed out, then Approve and Cancel buttons.
Habits that keep this smooth:
- Read the card, not just the button. The card lists the exact values — title, description, priority, who it is assigned to, the deadline and so on. This is your last checkpoint.
- Cancel and rephrase if any detail is wrong. It is faster than approving and fixing afterwards.
- Give the details up front so the card comes out right the first time. "Create a high-priority task for Sarah to review the Q3 report by Friday" produces a card you can approve immediately; "create a task" produces one you will have to correct.
- If the AI says it cannot do something, it is usually because the action is limited to managers or admins, or it is switched off for that assistant. Ask your admin rather than rephrasing endlessly.
:::caution Connected apps confirm before sending If your admin has enabled it, the assistant can also work with services your organization has connected. Anything that sends or changes data there shows you the service, the action and every value first, and waits for your approval. :::
Full detail: Using Tools.
Asking for Files, Tables and Diagrams
When you ask for rich output, it opens in a panel beside the chat and you watch it build live.
- Say the format you want. "...as an Excel file", "...as a presentation", "...as a diagram". Word, Excel, PDF and PowerPoint files arrive as a download card in the chat.
- Iterate instead of restarting. Ask for a change — "make the second column a percentage" — and the AI regenerates it.
- Preview before you download. The download card's Preview button opens the file in the side panel so you can check it first.
- Download what you want to keep. Generated files live inside the conversation.
- Press ESC to close the panel when you want the full chat width back. If you generated several, tabs appear so you can switch between them.
Full detail: Artifacts & Generated Files.
Steering a Response While It Is Running
You are not stuck once you press send.
| You want to... | Do this |
|---|---|
| Stop a long answer you no longer need | Press the Stop button — the send button turns into it while the AI is answering. What was written so far is kept. |
| Get a different attempt at the same question | Hover the last answer and click Regenerate |
| Continue an answer that got cut off | Hover the answer and click Continue |
| Reuse an answer elsewhere | Hover any message and click Copy |
Rate answers. The thumbs-up and thumbs-down on assistant messages are worth using — they are how good and bad answers get flagged. Click again to remove a rating.
Keeping Conversations Useful
- One topic per conversation. Context is what makes follow-ups work, and mixing unrelated topics in one chat makes it weaker. Start a new chat when you change subject.
- Expect long chats to be compacted. In a very long conversation you may see a note that older messages were summarised. If an early detail matters, restate it rather than assuming it is still in view.
- Find old chats in All Chats. The Recent list in the sidebar covers your latest conversations; All Chats covers everything, with a filter box and Today / Yesterday / N days ago date badges to help you locate one.
- Your personal conversations are private to you. Project conversations are the exception — they are shared with the project's members.
Full detail: Chat History.
Tips for Managers
- Ask about a team member directly. From the + button choose Select Employee, search by name or email, and pick the person. A chip with their name appears above the message box and your question is answered in their context.
- Clear the chip when you are done using the × on it, so later questions are not still about that person.
- Some actions need your role. Certain changes are limited to managers or admins. If an action is refused, that is usually why.
Tips for Project Chats
- Use a project chat when the work is shared. Project conversations are visible to the project's members, so the thread doubles as a record for the team.
- They are not tied to a single assistant, and web and document search are available by default — so they are a good place for wide-ranging questions.
- Keep one project per initiative so the context stays focused.
More on projects: Projects Overview.
Getting Better Answers from Your Documents
- Upload before you ask. The AI can only use documents that are already in the system.
- Check the access level of documents you rely on — it decides who the AI can share that content with. See Access Levels.
- Remove outdated versions. Two versions of the same policy is the most common cause of a confidently wrong answer.
- Point at the part you mean. "What does section 3 of the IT policy say about passwords?" beats "tell me about the IT policy."
- Ask across documents. "Compare the Q1 and Q2 budget reports" and "find all documents that mention GDPR" both work.
A Short Checklist
Before you send:
- Is the request specific — what, about what, in what shape?
- Is the right assistant selected?
- Are Web Search / Document Search set the way this question needs?
- Have attachments finished processing (green ✓)?
After you get the answer:
- Did you check the Sources?
- If it triggered an action, did you read the approval card before approving?
- Was it good or bad — did you rate it?
Getting Help
- Search the docs — use the search bar at the top of this site.
- Check the FAQ — common questions are answered in the FAQ section, with chat-specific ones in Chat & Assistants FAQ.
- Check Troubleshooting — for specific problems, see Troubleshooting.
- Ask the AI — it can answer questions about how to use the platform.
- Contact your admin — for access issues, role changes, or assistant configuration.