If you use artificial intelligence every day, it is obvious that you have already been through this situation: you are in the middle of an important work conversation and, suddenly, either ChatGPT or Claude starts to mix things up, providing information about a personal trip or another client who does not know I have nothing to do with that conversation.
This is one of the two most common problems for frequently using AI tools. The reason is tem nome and sobrenome: context and memory.
The more you speak with the IA, the more the armazena or history of your interactions. As time goes on, the chances of her mixing up matters and “contaminating” a chat with other information increase considerably. According to specialists in context engineering, the context memory of an AI agent works like the RAM of a computer: the more disorganized, the slower and more imprecise the tool becomes.
The news is that there is a present resource, not ChatGPT, Claude, Manus AI and Gemini that exactly solves that problem. Here, you will understand how to use AI Projects to isolate your conversations, organize your work, and stop receiving mixed responses. Let’s go la?
What is context contamination in AI?
Before talking about the solution, it is important to understand the problem. When you use a tool like ChatGPT, it tries to learn about you to deliver more relevant responses. The memory is working.
The problem is when this resource catches more than it helps. We are talking about this pollution phenomenon.
Contamination occurs when AI cannot separate what is relevant to your work from what belongs to your personal life. It mixes the contexts and delivers a response that unites commercial marketing concepts from roteiro to Disney.
| What do you want | Or that a contaminated AI delivers |
| An SEO analysis for a B2B client | An analysis mixed with the voice of your personal project |
| A travel route for the family | A list that includes technical terms of digital marketing |
| A clean programming code | A code that tries to solve problems from three different chats |
This is exactly to avoid this type of problem that we encounter with projects (or pasta, or notebooks, depending on the tool you use).
What about projects in AI tools?
Projects in AI are a way to encapsulate and organize everything within an independent space. You can create a personal project, a health project, a project for your company or a separate project for each client you serve.
The central point is that each project works in isolation.
Conversations recorded in a project do not interfere with others. You have built a specific memory for that context, without risk of contamination between subjects. If your account is professional, for example, it is ideal that each client has their own project, guaranteeing a management totally independent of each relationship.
Next, you will show how to activate and configure this resource in the main tools available on the market.
How to use non-ChatGPT projects
No ChatGPT (it is worth noting that this resource generally requires a paid account), the process of creating projects is quite direct, but there is a detail that many people end up ignoring.
To raise a project:
- In the left side menu, click on New project (or “Create a Project”).
- Give a clear name, such as “Travel for Disney” or “Client X”.
- Select an icon and a color to facilitate visual identification.
Memory security in ChatGPT
Here is the detail that is a little hidden: the configuration of the project. By clicking on it, you will find the Memory option.
You will see two options:
- The project can access memories from other chats and vice versa.
- Only the project: the project only accesses its own memories and is hidden from other chats.
If you do not want ChatGPT to mix the contexts, check the “Just projected” option. Attention: the tool warns that these settings cannot be altered afterwards. If you make a mistake, you will need to create the zero project.
Fonts and Instructions
Here the resource begins to show its real value. All chat created within the project appears automatically grouped. Also, you can add Fonts and Instructions.
It is possible to upload files (PDFs, sheets, logs) so that all new chats within the project can consult these documents automatically. In the instructions, you define how ChatGPT should behave: “I want you to act as a highly specialized tour guide at Disney for families with two small children.”
From the moment you save, all answers to this project must be given a guaranteed context.
How to use non-Claude projects
Claude also has the Logical Projects resource very similar to what we saw previously.
- In the main menu, click on Projects and then on New project.
- Give a name and add a description.
- In your configuration, you can insert the same custom instructions that you would use in ChatGPT.
Claude allows the upload of files as a knowledge base. The tool itself warns that “the project memory will appear here after some chats”, or it indicates that it is accumulating relevant information over time in an isolated manner per project.
To work with data analysis or review of long texts, Claude’s Projects tend to be quite efficient due to their large scope of context.
How to use Projects in Manus AI
Manus AI works with a different logic, focused on autonomous agents, but the organizational structure is quite similar to other tools.
In the Projects section, you create an isolated environment (like “Travel for Disney”). Inside it, you will find:
- Instructions: where you define the behavior in the context of the project.
- Files: to upload your knowledge base.
- Skills: a differential of Manus AI that allows you to connect specific tools and automation for that project.
- Scheduled tasks: to run automatic routines within the context.
Manus AI has a visually cleaner interface and does not display memory as explicitly as Claude, but all the messages and interactions are organized and isolated within the project.
How to use Non-Gemini Notebooks
The Google Gemini has a more recent resource that fulfills that same function, called Notebooks (or NotebookLM, depending on the integration).
You can create a Notebook, group conversations by topic and add to 300 different fonts. In the configuration of the Notebook, it is essential to ensure that the option “Consider all conversations in this notebook when responding” is activated.
Com isso, you work with more focus, knowing that Gemini considers only the universe of information that the specific paste, without mixing with your day-to-day searches on Google.
In short: organizing your chats on projects is not the quest to “make it beautiful”. It is a fundamental context engineering technique to ensure that artificial intelligence works for you, not against you.
To plan a trip, manage or market a client or study a complex topic, encapsulating information in independent contexts is the most direct path to precise and uncontaminated answers.
If you still haven’t applied that technique, start now. Create a project for each client, one for your personal life and another for studies. The difference in the quality of the responses appears before the first use! Organize your context and your productivity will respond.
Are you interested in issues related to digital marketing, AI and productivity? Please take advantage to also read our articles about Reports with AI to optimize your time in data analysis.
