Home / Knowledge base / Disadvantages of AI tools
What are the disadvantages of AI tools?
AI tools like Claude, ChatGPT and Gemini write texts and even build websites. That saves a lot of time. But they also bring risks for your data, your customers and your website. With a few fixed rules you prevent most problems.
The first risk is called prompt injection. Someone with bad intentions puts invisible text on a website. For example an instruction to forward data. If an AI tool reads that page it may mistake that instruction for a real one. So never let an AI simply do things on websites you do not know. And do not give it access to your bank or your admin accounts.
The second risk is in sharing. A shared chat or a shared AI page is a public web page. In the summer of 2026 shared conversations from Claude ended up in Google for a few days. The same thing happened with ChatGPT in 2025. So never share a chat with customer data, passwords or secret plans. Check in the settings which links are still open and revoke the ones you no longer need.
The third risk is training. With the free and the individual paid versions of Claude your conversations are used by default to train the model, unless you turn that off. Other providers have similar rules. Turn that setting off if you work with business data. With business plans and connections through the API this training does not happen.
The fourth risk is that AI makes mistakes that look good. A text can contain a fact that is wrong. Code can work and still have a security hole. Or your site stops working and nobody understands why, because nobody wrote the code themselves. So always have someone check the work. Make sure you have a backup and an explanation of how the site is built.
The fifth risk is fake downloads. Scammers make sites and ads that look like those of well-known AI brands. People who download something there sometimes get software that steals passwords. Only download an AI app from the official site of the maker and never through an ad.
We work with AI every day ourselves. That is why we have fixed rules. Texts we read on other websites only count as data for the AI and never as an instruction. We check what the AI writes for links, code and strange instructions before it goes live. Small things that are easy to undo we carry out ourselves. Bigger changes only go live after approval. For customer data we use business connections where your data is not used to train AI.
Real-life example
A business owner asks ChatGPT to check a quote and shares the conversation with a colleague through a public link. The quote contains the name and address of a customer. Anyone with that link can read the quote, even if the link ends up somewhere else. It is better to make the quote anonymous first or to simply send the colleague the file.
How we help with this
At Marketing Friends we take care of this for businesses across Europe. Read more about Website maintenance and hosting or start with the free scan.
Also good to know
Curious how your website scores? Within a minute you see your score. Then your free growth plan, with your progress every Monday.
Start your free growth plan