How to really use AI in your business (without the hallucinations)

Quick answer: AI starts well and then hallucinates because it lacks context about your business. To really use AI, build a knowledge base (visual identity, products, customers, sales methodology), keep that documentation somewhere the AI can check, such as Linear, and treat the AI as a team you train, not a tool that gets it right on the first try. The method works with ChatGPT, Claude or Gemini.

I use artificial intelligence every day, and my clients use it more and more too. But what I hear the most is: "Daniel, you can do a lot of things I try to do, and for me the AI just doesn't answer properly".

What makes the difference isn't the tool. Everything here works with ChatGPT, Claude or Gemini.

Watch the full video on YouTube (in Portuguese)

Why AI starts well and then hallucinates

A client of mine was building a workflow to analyze documents and compare them. It couldn't be plain automation: it needed AI in the process. She opened Claude and started talking to it. The first few times, everything ran very well. Then it started hallucinating and not answering the way she wanted.

This happens because most people work only inside the chat. If the AI has no information about your company, your products, your customers and the methods you use to sell, a good result is unlikely.

"One shot" doesn't work anymore

"One shot" means sending everything at once and expecting a perfect result on the first try. For business tasks, the way to go is building it together with the AI: talk, correct, and keep a record of what worked.

Step 1: build a knowledge base about your business

Talk to the AI so it understands your business. Start, for example, with your visual identity. If you already have a file from your agency, great. If not, ask it to look at your website and document your identity. If your company is small and has none of this yet, no problem: ask the AI to create it with you.

Then do the same for the other areas: marketing, sales, finance and products.

The secret: several AIs, one project

I use Claude Code, by Anthropic, a lot, because it lets me create folders per project and keep each project's information there. And I use ChatGPT for other things. Think of each AI as an employee working on the same project.

Linear: the shared memory of your AIs

For them to share the same memory, I use Linear, a project management tool that can be connected to both Claude Code and ChatGPT. That's where each project's scope and tasks (issues) live. GitHub also works as a knowledge base. Linear has a free plan, with a limit on issues.

In practice: step by step in Linear

1. Create your Linear account and connect it to the AI you use.
2. Talk until you get a result you like.
3. Ask: "based on this, create a visual identity project in my Linear and record everything we did".
4. Do the same with your sales methodology, products and customers.

One project for the whole business

A mistake I made: at first, I had a separate project for each area. Today I have a Business Plan with everything inside: visual identity, marketing, sales, finance and products.

When I want to create something new, like promoting a mentorship program, I ask the AI to visit that project and create a campaign for each ICP (clinic owners, industry owners, industry executives). That same knowledge base is what lets an AI agent write personalized messages in outbound prospecting with AI.

The trick most people miss: why so many give up on AI

At first, the process is hard. The AI delivers a lot of things that need fixing, and most people give up halfway: "this AI doesn't do anything right". But this trial and error is part of the process, like a child learning to talk.

AI is a team you train

For your business, think of it as training a team. If you hired the best salesperson in the world, handing over a computer and a phone and saying "go sell" wouldn't be enough. You need to explain the business, the product and the customer, and follow up. With AI it's exactly the same.

Don't know how to guide it? Ask it

If you don't know how to guide it, ask the AI: "what do I need to tell you so you can help me create this?". And as it improves, always ask: "save the progress and update the project documentation". In the next conversation, ask it to revisit the project before starting.

Frequently asked questions

Why does AI sometimes answer well and sometimes hallucinate?

Because most people work only inside the chat: the AI has no information about the company, its products, its customers and its sales methodology. Without that context, it starts well and then begins to make mistakes or answer differently than expected.

What is "one shot" and why doesn't it work?

It means asking for everything at once and expecting a perfect result on the first try. For business tasks, it is better to build together with the AI: talk, correct and keep a record of what worked.

How do I build a knowledge base for AI?

Talk to the AI about each area of the business (visual identity, marketing, sales, finance, products), ask it to document what was defined and keep that somewhere it can check later, such as Linear, GitHub or a folder of files.

What is Linear used for with AI?

Linear is a project management tool that can be connected to Claude Code and ChatGPT. The AIs document each project's scope and tasks (issues) there and check that information in future conversations. It has a free plan, with a limit on issues.

Can I use ChatGPT and Claude at the same time?

Yes. Think of each AI as an employee working on the same project: connected to the same Linear workspace, they share the same documentation.

Why do so many people give up on AI?

Because at first the AI delivers a lot of things that need fixing. That trial and error is part of the training: whoever goes through it and records what they learned builds a knowledge base that makes the AI deliver much better results.

Want to go further?

I offer a one-on-one AI mentorship focused on your business, recorded so you can revisit it later. To learn more and book a conversation with me, talk to Aurora, my AI assistant, on WhatsApp.

If you want to automate processes and build AI agents, also check out n8n MasterFlow.