I've been asking one question about every AI app I use: who does this thing actually work for?

Think about lawyers. Yours and the other side's are both smart, both polite, both helpful in the moment. You only find out which one is yours when your interests and theirs split. AI is the same. Most of the time it just answers the question. The tell is everything around the answer.

Some of what's below is uncomfortable. None of it is a conspiracy. It's incentives. Follow who pays and the design makes sense.

Who pays for it

Somebody pays for all those servers. If it isn't you, it's someone who wants something from you.

Some AI apps now show ads right under the answer. One goes further. Its help page says that when ad personalization and memory are on, it "may save and use memories and reference recent chats when selecting an ad."

Since last December, another major company has used people's chats with its AI to help personalize the ads they see across its apps. In the US there's no setting to keep your AI chats out of it. Both companies say sensitive topics like health are kept away from ads.

Still, what you tell an AI can end up helping someone sell you something. The app now has two customers. Only one of them is paying.

What it writes down about you

Researchers recently looked at what one popular AI assistant had saved about 80 real people who volunteered their data. 96% of those memories were written by the AI on its own. Only 4% came from someone saying "remember this."

A lot of it wasn't facts either. About half the memories recorded something psychological (fears, wants, plans). One noted that someone leaned on it for advice because getting to a doctor was hard and money was tight. Another read like this:

A real memory line, quoted in a 2026 study of what one AI assistant saved about its users.

Nobody asked it to keep those. It decided.

It also got people wrong. Roughly 1 in 11 memories weren't well supported by what the person actually said. Ask for bands like Nirvana and it may decide you love Nirvana. Funny with music. Less funny when it's your health.

What you can't see

Most of these apps have a memory screen. It feels like the whole picture. One leading assistant's help page asks the question directly, and answers it:

“Does the memory summary include everything [it] remembers?”

“No.”

From one leading AI assistant's own help page, checked October 22, 2026

The same page says there's no separate storage limit on what it can pull from your past chats. Last year a security researcher dug into that layer and found notes like "User's name is Johann. Confidence=high," plus summaries of about 40 of his recent chats. He couldn't see any of it in the app.

Imagine a bank that shows you the balance but not the transactions. That's roughly the deal with AI memory today.

When delete isn't delete

You'd think deleting a chat deletes it. Mostly, eventually. With exceptions.

At one major assistant, if a human reviewer has read your chat, deleting it doesn't delete their copy. It can be kept for up to three years. At another, deleting a chat doesn't necessarily delete a memory the AI already made from it. And last year a court ordered one AI company to hold on to chats people had deleted, for about four months, as part of a lawsuit.

None of this is hidden exactly. It's in the help pages. It's just not what anyone pictures when they tap delete.

Defaults decide for you

At many of the big AI apps, your chats help train the model unless you find the setting and turn it off. Even on some paid plans.

Turning it off often only works going forward. What the model already learned from you stays learned.

Most people never open settings, so the default is the real policy. If a setting is three layers deep, it's probably the one that matters most.

It agrees with you a little too easily

A Stanford-led study published in Science this March found AI chatbots affirmed people about 49% more often than other humans did, even when the person was in the wrong.

Agreement feels good. It keeps you talking. It doesn't always help.

Put that next to an AI that remembers what you want and also sells ads.. you can see where the worry comes from.

What AI on your side looks like

Flip each one.

AI on their sideAI on your side
Ads under the answerNo ads
Your chats train the model unless you find the switchNever used to train AI
It decides what to remember about youIt keeps only what you choose
A summary of what it knowsRead it, edit it or remove any line anytime
Chats kept until you delete themChats clear on your timer
Agrees with you a little too easilyIt can still be wrong. Check it.

No ads. So there's no one else in the conversation.

We don't sell your data.

Never used to train AI. There's no switch to go find.

Chats clear on your timer: 7 days by default, or 8 hours or 30 days. You pick.

Memory is your call. It's off until you choose how it should remember, and it keeps only what you choose. You can read it, edit it or remove any line anytime, and export it free on every plan. The chat clears. The Knowledgebook doesn't.

One more, because it matters most. AI on your side still makes mistakes. It can still be wrong. Check it. An assistant that tells you that up front is easier to trust than one that never does.

Check your own AI in five minutes

Open settings and look for anything about training, data controls or "improving the model." Decide if you want it on.

Ask it what it remembers about you. Then open the memory screen and see if the two match.

Find out how long chats are kept, and what delete actually removes. Our step-by-step delete guide walks through one popular app.

Give it a plan you know is shaky and see if it pushes back.

Look under the answer. If there's an ad, you know who else is in the room.

Why we built Ask Safely this way

We started Ask Safely for the questions people actually carry around (ie the bill that makes no sense, the letter from the insurance company, the health worry, the money math etc.). Those are the questions you'd only ask something that works for you.

So the business is simple. Free to start. People who want more pay for it. No advertiser, no training deal, no reason to keep you talking longer than you need.

I think the next few years of AI get decided by one question people start asking out loud: whose side is this on?

Most people already suspect the answer. Now there's a choice.

FAQ

What does "AI on your side" mean?

AI built to work for the person typing, not for an advertiser, a model in training or an app that wants more of your time. At Ask Safely that means no ads, chats that are never used to train AI, and chats that clear on your timer.

Does AI remember everything I tell it?

It depends on the app. Some AI apps save memories on their own and also draw on past chats in ways you can't fully see. Check the memory settings, and ask the app what it remembers about you.

How do I know if an AI app trains on my chats?

Look in its settings for a data or training control. Many AI chat apps train on conversations by default unless you turn that setting off, and turning it off often only applies going forward.

Does Ask Safely show ads?

No. Ask Safely has no ads, and we don't sell your data.

Does Ask Safely remember things about me?

Only what you choose. Knowledgebooks are off until you choose how Ask Safely should remember. You can read, edit or remove any line anytime, and export it free on every plan.

How long does Ask Safely keep my chats?

Chats clear on your timer: 7 days by default, or 8 hours or 30 days. Anything you choose to keep goes in your Knowledgebook.

Can AI on your side still be wrong?

Yes. It can still be wrong. Check it, especially for health, money and legal questions.

Sources