Chatting and Conversations Using AmplifyAI

This guide explains how to get fast, accurate results from AmplifyAI by choosing the right model, setting the appropriate reasoning level, and using features like Web Search, file uploads, and prompt templates efficiently. It shows practical tips to avoid hitting chat limits, handle files and images correctly, and improve outputs by treating responses as drafts and refining prompts. Follow it to save time, reduce costs, and get more reliable AI-assisted work.

Kirstie Knighton
15 steps
2 minutes

Cr8

Choose an AI Model to Use

Choose an AI model to use. (See our guide to choosing models if you need help.) This list of models can change as Amplify and Mercer add, remove, or rename models. Do not assume the most expensive model is always the best choice. Match the model to the task and compare results for important work.

Using AmplifyAI Efficiently: Setting the Temperature, Response Length, and Reasoning Effort

Click the gear symbol above the model name.

Here you can adjust the Temperature and Response Length. **Temperature:** Lower values are more focused and repeatable; higher values are more varied and creative. Use lower settings for factual summaries, policies, grading criteria, or extraction. Use higher settings for brainstorming and alternatives. **Response Length:** Sets a preference from concise to verbose. It allows but does not guarantee longer answers. A precise word or item limit in the prompt is more reliable.

At the bottom right of your message window, there is a place to toggle the **Reasoning Level** (or thinking depth) of the AI model before it generates a response. See below for explanation of each level.

**Keep reasoning on Off or Low for most tasks** Off or Low should be used for almost all everyday work, including writing, summarizing, brainstorming, planning, document review, study support, and most routine analysis. Medium and High use additional computing resources and may cause users to reach their chat limits faster. Reserve them primarily for unusually complex multi-step problems, especially advanced mathematics or difficult coding.

- **Low (Lightbulb Icon):** **Fast & Direct** - *What it does:* Uses minimal internal processing or step-by-step reasoning. - *Best for:* Quick factual questions, simple text rewrites, short summaries, or casual brainstorming where speed is prioritized over deep analysis. - *Trade off:* Lower depth on complex topics - **Medium (Scale Icon):** **Balanced & Methodical** *(Default)* - *What it does:* Striking a balance between processing time and depth. The AI evaluates key details and structure before responding. - *Best for:* Everyday professional tasks, standard drafting, instructional guides, and general problem-solving. - *Trade off:* Moderate processing speed - **High (Brain Icon):** **Deep Thinking & Analytical** - *What it does:* The model engages in extensive internal step-by-step reasoning, logical validation, and thorough analysis before outputting an answer. - *Best for:* Complex coding tasks, advanced logic puzzles, intricate policy/data analysis, or nuanced multi-step problem solving where accuracy and depth are critical. - *Trade off:* Slightly longer wait time for response - **Off:** **Instant / Unfiltered Output** - *What it does:* Disables additional internal reasoning loops entirely, generating text directly token-by-token. - *Best for:* Real-time speed requirements, simple conversational banter, or basic formatting updates. - *Trade off:* No deliberate problem-solving step

**To use AmplifyAI efficiently and avoid reaching your chat rate limit:** - Keep reasoning on Off or Low for most tasks. - Leave Web Search off when current web information is unnecessary. - Use short conversations rather than very long unrelated threads. - Save uploaded files to the Amplify File Manager when you may need them again. Saved files can be analyzed later, so repeated uploads of the same files are unnecessary. - Upload only the files and material needed for the task. - Use assistants or prompt templates for repeated tasks. - Verify important facts, citations, calculations, quotations, policies, and dates before using the result.

Open the selected model's settings to see Custom Instructions, Temperature, and Response Length. The exact range and effect can vary by model.

**Important Distinction:** Reasoning effort is not temperature. Reasoning effort changes how much processing the model applies. Temperature changes how predictable or varied the wording and choices are.

**Turn on Web Search only when you need it** Select the globe icon beneath the prompt window to turn Web Search on. Leave it off for tasks that need only your prompt, uploaded material, or the model’s training data. If current information matters, enable the control and verify the returned sources rather than assuming the model searched automatically. Not using the Web is more cost efficient and allows the model to work from its built-in training knowledge. All current models can use the platform's Web Search tool when it is enabled.

Start a new Chat/Conversation

Click the "Type a message to chat with Amplify..." field at the bottom of the screen.

Tip! A useful prompt usually contains five parts: Task, Context, Source, Output, Limits. **Prompt Pattern:** Act as \[role\]. Using \[source/context\], do \[task\] for \[audience\]. Return \[format and length\]. Follow \[constraints\]. Identify uncertainty and do not invent missing information.

Tip--Prompt Optimizer! You can use the magic wand to have the AI help you write a better prompt! After writing your prompt, click on the wand and let AI improve your prompt.

The AI will generate an improved prompt for you, but you still have control over the prompt. You can edit it or delete it and start over.

Type or paste in a prompt into the text box. For a task based on a document, spreadsheet, PDF, presentation, or image, select Upload a File **(the + sign)** near the lower-left side of the message box. AmplifyAI opens your computer's file explorer. Choose a file from your computer. For a file already saved in Amplify's File Manager, use "Files" instead **(the multiple papers symbol to the left of the + sign)**. The interface does not display file-type or file-size guidance. Select the file and wait until its filename appears with a green check mark. When you're finished, click "Send" (paper plane icon) to submit the prompt.

Tip! If the model says it cannot see the file: Confirm the green check mark, mention the exact filename, reattach the file, try a new conversation, choose a model that supports the needed file or image type, remove password protection, or convert a scanned PDF to searchable text.

The symbols below the chat window are described below.

Watch the AI Generate an answer to your prompt.

Tip! Treat the first answer as a draft. Ask the model to explain a choice, show evidence, shorten or reorganize the output, identify uncertainty, or compare two alternatives. Begin a new conversation when the earlier conversation is no longer relevant.

Response Toolbar

The response toolbar will be to the right of the AI's response in your chat.

Tip! Here are what the icons mean in the response toolbar.

Sharing Convesations/Chats

Click your profile menu by clicking on your profile pic in the upper right.

Click "Sharing."

Click "Share with Others."

[[Enter or add the recipients. Click "ok."]]

Click the "Describe what you are sharing (required)." field. Write the required note describing what you are sharing.

Toggle on the conversation(s) you want to share. You can also share prompt assistants and folders. Review the selected items carefully because sharing makes the chosen material available to the recipients. Once finished, click "share."

Images and Files

AmplifyAI is not intended for image creation. It is primarily used for analyzing images and other uploaded content. Simple image creation is available only through GPT-5.6 Luna. All current models can analyze images that you upload. **Image analysis**\ All models can analyze images supplied in a conversation. For image analysis, ask what is visible before asking for interpretation. Check small text, numbers, and labels yourself. **Simple image creation**\ GPT-5.6 Luna is the only model that supports simple image creation. The other models do not create images but can analyze uploaded images. For files, upload only what the task actually needs. Avoid unnecessarily large files. If a file is not recognized, wait for the upload confirmation, mention the filename in your prompt, reattach it, remove password protection, or convert a scanned PDF to searchable text.