IA

What you should never paste into an AI chat

Por Bryan Taylor Reverón 6 min de lectura

Your team already uses AI chats. You may not know it, but they do. Someone pastes an email so it can be summarised. Another pastes a contract so it can be explained. It’s quick and it works. The problem isn’t the tool. The problem is what leaves your company every time someone hits «send».

This post answers two questions: what you should never paste into a free chat, and which AI you can use in your company without risking your data.

What happens when you paste something into a chat

Everything you type into an AI chat travels to the servers of the company that runs it. Those servers are often outside Europe, and you have no contract with that company.

On free versions, what you paste is often used to train the AI. In other words, your text may stay there, and it may resurface, mixed in, in the answer given to someone else.

Business paid versions let you ask that your texts not be used to train the AI, and the provider signs a contract with you stating what it does with your data.

What you should never paste into a free chat

A simple rule: only material you could already publish on your website goes into a free chat. Everything else stays out.

The following does not go in:

  • Contracts, offers and prices that are not public.
  • Lists of customers, suppliers or employees.
  • Names, phone numbers, emails or health data of any person.
  • Internal reports, accounts and plans.
  • Access keys, even «just for testing».
  • Anything a client has given you in confidence.

If you need help with a document like this, leave only the text, without names or figures. Or use a company AI, which we explain below.

Don’t trust what it says without checking it

An AI chat has no way of knowing whether what it writes is true. It writes what sounds good. That is why it invents figures, cites studies that don’t exist and puts words in the mouths of people who never said them. And it does so with the same confidence it has when it gets things right.

If you publish an AI text without reviewing it, what it says is what you are saying. Your company’s name goes under it. You are the one who answers for it.

What to check before publishing:

  • Check every fact, figure and quote against the source it comes from. If you can’t verify it, remove it.
  • Anything it says about your product is reviewed by someone who knows it.
  • Anything that could lead to a complaint is reviewed by the person responsible: warranties, terms, health or money.
  • The tone. AI texts all sound the same.

And a named person signs every publication. AI proposes. Your company decides.

Which AI you can use in your company

There are three levels, from least to most secure.

1. A free chat. It’s fine for anything already public: a draft of a note, ideas for a headline, understanding a word. Nothing more.

2. A paid business chat. The best-known chats have a business version. It costs money, but in return you can ask that your texts not be used for learning, they sign a contract with you and they tell you where they store the data. Check their terms before signing up, and choose one that stores data in Europe. With this you can work with internal documents. It is still not the place for health data or access keys.

3. Your own assistant, working with your documents. This is software built for your company. It reads only your documents and answers from them, saying where each piece of information comes from. Your data stays in Europe and nobody uses it to train another AI. Each employee only gets answers from documents they already had permission to read. If it cannot find the answer, it says so. This is one of the things we build: an assistant that answers using your company’s documents.

Whichever level you use, write a short policy: one page. Which tools your team can use, what for, and with what limits. Without a policy, each person decides for themselves, and that is how a report ends up in a free chat.

Record what was done with AI

Note in the post itself who used AI, for what, and who reviewed it. If someone complains a year later, you’ll have the answer in a minute.

In WordPress this fits in an internal note on the post. You don’t need a new system. You need the habit.

First: an AI text can copy passages from books or websites the AI was trained on. If you publish it, the one copying is you. Reviewing it before publication is what protects you.

Second, and less well known: a text written entirely by AI may be copied by anyone. The more you contribute, the more it is yours. Below is what European law says.

And Google

Google does not penalise a text for having been made with AI. It penalises texts made to deceive the search engine. Since 2024 it has targeted exactly what AI makes easy: publishing many similar, empty pages. A blog like that gains no visits. It loses them, and it drags the good pages down with it.

What works is what has always worked: answering a real question better than others, with what only you know. AI helps you organise and get started. What only you know, you have to put in yourself.

In six steps

  1. Decide what you want to say and to whom, before opening the chat.
  2. Give it only material you could already publish. Nothing confidential.
  3. Ask it for a draft, not the final text.
  4. Check every fact against its source. Rewrite in your own voice.
  5. Record what was done with AI, who reviewed it and who approved it.
  6. Sign with a person’s name. And state that you have used AI when the law or your policy requires it.

Used this way, AI genuinely saves time. Used to publish without review, what you save today you pay for later: your company looks bad, you lose visits or you end up with a legal problem.

Would you like us to help you write your company’s AI use policy? Or to set up an assistant that only reads your documents? Get in touch.

If you want to know what the law says

Europe has had an artificial intelligence regulation since 2024. For those who use chats, the key provisions apply from 2 August 2026.

  • If you publish a text made with AI to inform the public on matters of general interest, you must say so, unless a person has reviewed it and takes responsibility for it.
  • Images, audio or video that are fake but look real must always be labelled.
  • A chat on your website must state that it is an automated system.
  • In Europe, a text that no person has written may not be protected by copyright.

This is a summary: consult your accountant or adviser.