When someone on your staff types, pastes, uploads or records something in an AI tool, that content can travel further than the answer on the screen. This lesson shows where it can go and gives you 5 questions that tell you which of those places a given tool sends it. Your staff are probably using these tools already this week, so it helps to know before the next donor list or case note goes in.

The idea: your content can go to 5 places

Think of anything you put into an AI tool as a letter you hand to a stranger. It gets read so they can reply. After that, it might be filed, shown to someone else, used to teach new staff, or passed along. AI tools work much the same way. Here are the places your content can end up.

1. Processed to produce the answer

This part is unavoidable. The tool has to read what you gave it to write a reply. Your text, file or recording goes to the vendor’s systems, gets worked on, and the answer comes back.

2. Stored on the vendor’s side

Most tools keep your chats and uploaded files as history, often so you can scroll back and pick up where you left off. That storage is on the vendor’s systems, not only on your computer. How long they keep it is called data retention.

Deleting a chat does not always mean it is gone right away. Some vendors keep deleted content for a further period before removing it for good.

3. Used to train or improve the vendor’s models

With many consumer tools, what you type may be used to train or improve the vendor’s AI models unless you opt out. Training means your content becomes part of what the vendor uses to make future versions of the tool better. The setting that stops this is usually called a training opt-out.

4. Reviewed by people at the vendor

Some vendors let their own staff read a sample of conversations, for example to check for safety problems or abuse. That means a person you have never met may see what your staff typed.

5. Passed on through connected apps and plug-ins

Many AI tools can connect to other services: your email, a calendar, a file drive, or add-ons made by other companies. When a connection is active, your content can be passed on to those services, and each one has its own rules.

Your plan changes the answer

Whether your content is used for training often depends on which plan you are on. Many vendors treat free or personal accounts differently from business, team or enterprise accounts. They usually say so in their privacy notice or on a page about how they use data. A business plan may have training off by default, offer shorter retention, or even offer zero data retention, where the vendor does not keep your content after producing the answer.

So the same tool, from the same vendor, can behave quite differently depending on whether someone signed up with a personal email or your organization bought a team plan.

This is an illustration, not a real organization.

A small food bank runs on volunteers. The volunteer coordinator has a messy sign-up sheet: names, phone numbers, preferred shifts, some typed in all caps, some with notes in the margins. She opens a free chat assistant on her phone, pastes the whole sheet in, and asks it to sort everyone by shift and fix the formatting. Thirty seconds later she has a clean list. It feels like a small win.

Here is where that sheet can go:

Processed for the answer. The names and phone numbers go to the vendor’s systems so the tool can sort them. Expected.
Stored as history. The conversation, including every phone number, is likely saved in her chat history on the vendor’s side. If she deletes it, the vendor may keep it for a further period.
Used for training. On a free account, the vendor may use the conversation to improve its models unless she has opted out. She has never opened the settings, so she has not.
Seen by a reviewer. If the vendor samples conversations for safety checks, a person at the vendor could read the sheet.
Shared through connections. If she had linked the assistant to another app, the content could also flow there.

None of the volunteers agreed to any of this. They gave their phone numbers to the food bank, not to a tech vendor.

Now picture the same task on a business plan the food bank set up properly:

Training on the food bank’s content is turned off, and an admin has locked that setting for everyone.
Retention is short, so chats and files are removed sooner.
Connected apps are limited to ones the food bank has approved.

The sheet still gets processed to produce the answer. That part never goes away. But the number of places it can end up drops sharply, and the food bank knows the rules instead of guessing.

A third option costs nothing: remove the phone numbers before pasting, or use made-up placeholder names. The tool can sort shifts without knowing who anyone is.

The steps to take: 5 questions for any tool

You do not need to read every word of a privacy policy. These 5 questions answer most of what matters for any AI tool:

Does it train on our content by default? If the answer is yes, someone has to act to stop it.
How do we turn training off, and who can do it? On some plans each user has to change the setting themselves. On others an admin can do it for everyone.
How long is our content kept? Ask about chats, uploaded files, and what happens after someone deletes them.
Can an admin control settings for everyone? If not, your protection depends on every staff member remembering to change their own settings.
Which plan are we on? This is often the most important question, because the answers to the other 4 can change with the plan.

Where do you find the answers? You have 2 options:

Use this site’s tool profiles. Our tool safety profiles answer these questions for everyday AI tools, using the vendors’ own published pages. When a vendor does not publish something, we mark it “not documented” so you can see the gap.
Read the vendor’s pages yourself. Look for the privacy notice, a page about data use or data controls, and the details of the plan you pay for (or do not pay for).

Treat “not documented” as a reason to be careful, not as a reason to assume the best. If a vendor will not say how long it keeps your content, plan as though it keeps it for a long time.

Common mistakes

Assuming “free” and “paid” behave the same. The same tool often handles data differently depending on the plan. Check which one each staff member is actually using.
Thinking delete means gone. Deleting a chat removes it from your screen. The vendor may keep it for a while longer.
Relying on each person to change their own settings. People forget, new staff join, settings reset on new devices. Where you can, have an admin set training and retention for everyone.
Forgetting about connected apps. Linking an AI tool to email or a shared drive can quietly widen where your content goes. Review connections the same way you review the tool itself.
Checking once and never again. Vendors change their terms and settings. Look again when you renew a plan or when a tool adds new features.
Pasting personal details when they are not needed. Many tasks work fine with names and contact details removed. The safest data is the data you never put in.

Try it this week

Set aside 15 minutes.

Pick the AI tool your staff use most. If you are not sure, ask 2 or 3 people.
Open its profile in our tool directory. If it is not listed, open the vendor’s privacy page instead.
Make a simple table with 2 columns: the question, and the answer.
Fill in the 5 questions: training by default, how to turn it off and who can, how long content is kept, whether an admin controls settings for everyone, and which plan you are on.
Write “not documented” anywhere the answer is missing, and circle anything that surprises you.

Keep the table. It is the first page of your organization’s AI safety notes, and you will build on it.

Key takeaways

Anything staff put into an AI tool can be processed, stored, used for training, reviewed by people at the vendor, or passed on through connected apps.
Free and personal accounts often handle your data differently from business, team or enterprise plans, so knowing your plan matters.
Deleting a chat does not always remove it right away, and some vendors offer business customers shorter or zero retention.
The 5 questions about training, opt-out control, retention, admin settings and your plan answer most of what you need to know about any tool.
When a vendor does not document something, treat that gap as a risk until you learn more.

What AI tools do with your data