9 AI Mistakes to Avoid in Copilot, ChatGPT and Gemini
Copilot is now built into Word, Excel and Outlook, and ChatGPT, Gemini and Claude are open on millions of desktops every day. The tools are genuinely useful. The way most people use them is not. Invented facts sent to clients, confidential data pasted into a public chatbot, and workflows that collapse the day a subscription lapses are all happening right now. Here are nine AI mistakes that cause real damage and exactly what to do instead.
AI tools are genuinely useful for drafting, summarizing, brainstorming, and speeding up repetitive work. The mistakes are not really about AI being bad. They are about people treating a text prediction tool as if it were an accurate, unbiased, all-knowing assistant. It is not. Use it for what it is good at, check what it gives you, don’t trust the AI hype, and keep your judgment in the loop.
Trusting AI Output Without Checking It
This is the big one. AI tools generate confident sounding text even when they are wrong. This is called hallucination, which is a polite word for making things up, and it happens more than the vendors admit.
If you ask an AI for a statistic, a legal citation, a court case, or a specific fact, you cannot assume it is correct. AI models have produced fake case law that got lawyers sanctioned in court, and invented statistics that looked completely plausible.
The fix is simple: treat AI output as a first draft from an enthusiastic but unreliable intern.
Verify anything that matters before you send it, publish it, or act on it.
Using AI as a Search Engine for Latest Facts
AI chatbots are language prediction tools, not fact databases. They predict what a good answer looks like, not what is true.
AI services are only as current as their last web search, and many answer from training data without checking anything.
For anything time sensitive (prices, current events, current staff at a company, today’s exchange rate) or anything where being wrong is costly (medical, legal, financial specifics), use a real search engine or authoritative source instead.
Save AI for drafting, summarizing, and brainstorming, not for looking up the latest info.
Pasting Confidential Data Into Public AI Tools
People paste customer records, financial figures, and internal strategy documents into free AI chat tools without thinking about where that data goes.
Depending on the tool and your account type, your input may be used to train future models or stored on servers outside your control. Samsung engineers leaked internal source code this way in April 2023 after pasting it into ChatGPT to debug it.
There are options in AI services to stop your prompts from being used to train the models. Whether you trust that the option is honored is another matter.
Even if data is not added to training, it’s still saved on the AI servers. The files and prompts could be accessed in a data breach or if your account is hacked.
Before pasting anything into an AI tool, ask yourself: would I be comfortable if this text ended up in a training dataset or a data breach? If not, do not paste it. Use your organization’s approved AI tool with proper data protection, not a personal account.
Not Locking Down Your AI Account
Too often people forget to secure the login to any AI service. An AI account can accumulate a lot of information about you or your business that you don’t want hackers to see.
Regular Office Watch readers will know we strongly urge people to secure important logins with two-factor authentication or passkeys. Either method makes it very difficult for someone to hack into an account, even if they discover the password.
Skipping the Edit
A lot of AI generated writing has a distinct flavor: it overuses words and phrases like “delve,” “moreover,” and “in today’s fast paced world,” and it tends toward bland, generic sentences.
If you copy and paste AI output straight into an email, report, or article without editing, readers can often tell. It reads as impersonal and sometimes just wrong for your specific situation. Always read it, cut the filler, and add the specific details only you would know.
It is much better to create your own writing style, or more likely styles. Our book Write better and faster with any AI has a chapter all about making and using a custom AI prompt to write the way you do.
Assuming AI Understands Context It Was Never Given
AI tools only know what is in the conversation, plus whatever they were trained on. They do not know your company’s internal policies, your boss’s preferences, or the email thread from three weeks ago unless you tell them.
Vague prompts get vague, generic answers. “Write a proposal” gets you boilerplate text. “Write a two page proposal for our client Acme Corp, focused on cutting their shipping costs by 15 percent, in a formal tone” gets you something usable. Be specific, or expect generic output.
Letting AI Make Decisions It Should Not Make
Using AI to draft a performance review, screen job candidates, or make judgment calls that affect real people, without human oversight, is a growing problem.
AI models can and do reflect biases present in their training data. Amazon famously scrapped an AI recruiting tool in 2018 because it had taught itself to downgrade resumes that included the word “women’s”. AI images tend toward white, male faces because that is what most photos on the Internet show and the AI was trained on those images.
If AI is helping with a decision that affects someone’s job, health, or finances, a human needs to review the reasoning, not just approve the output.
For images with people, be at least a little specific about gender, age and perhaps nationality.
Over Relying on AI and Losing the Skill Yourself
Students who let AI write every essay, and workers who let AI write every email, risk losing the underlying skill of writing and thinking clearly.
Use AI to speed up work you already know how to do, not as a total replacement for learning.
If you cannot explain or defend what the AI produced, you have outsourced too much.
Building a AI Workflow With No Backup Plan
Every paid AI tool, like Copilot, is a subscription dependency. If your license lapses or your account is blocked, the AI features stop. If the company changes pricing or availability, your workflow changes with it, whether you like it or not.
Do not build critical business processes entirely around a specific AI feature without a backup plan. Know what you would do if that feature disappeared tomorrow.
Ideally, keep a separate library of prompts you can use in any AI service.
Write better and faster with any AI
Copilot Notebooks: The Safer Way to Use Microsoft’s AI
Copilot’s Legal Trap or what you’re giving to Microsoft
Essential ways to strengthen your Microsoft account
Busting the myth about Two-Factor Authentication
Check your Microsoft account recovery options … now before it’s too late
Which Copilot Model Should You Use in Excel? Auto, GPT or Claude
PowerPoint AI Images Tested: Which Copilot Model Actually Works
Turn Copilot Off in Microsoft 365: How to Guide for Word, Excel, PowerPoint & Outlook
AI Credits in Microsoft 365: What They Are and How They Impact Your Copilot Usage
Publisher 365 Kill Switch: What Microsoft’s New Email Gets Wrong