Skip to main content

Our Thinking/AI Explained

Your AI isn't a diary. It's a postcard.

What actually happens to what you type into AI

We tell AI things we'd think twice about saying out loud. Here's where those words really go, and why a little "postcard thinking" lets your organisation use AI with confidence.

By Digital Panda

· 7 minute read

There's something about a blank chat box. It's quiet, it's patient and it never judges. So we tell it things.

The awkward reply to a volunteer who's struggling. The donor spreadsheet that won't add up. The board paper that needs to sound more confident than you feel. The health worry you're not sure how to raise with your manager. Most of us would think twice before saying any of that across a busy office. In a chat box, we barely think at all.

That isn't carelessness. It's how these tools are designed to feel: personal, private, a bit like a diary. But a diary stays in your drawer. What you type into AI goes on a journey.

What it feels like

A diary

Private. Just you. Stays where you put it. Nobody else will ever read it.

What it's really like

A postcard

It almost always arrives safely, and usually nobody pays it much attention. But it passes through other hands, it can be kept, and you'd never write your bank details on one.

That's the right level of caution for AI. Not fear, and certainly not "stop using it". Just the same quiet awareness you'd have writing a postcard. So where does it actually go?

The journey

Where your words actually go

The details vary between tools and plans, but most messages make some version of these five stops.

  1. 01

    It leaves your device

    Your words are sent to the AI company's servers to be processed. The thinking doesn't happen on your laptop.

  2. 02

    It's saved to your history

    So you can pick the conversation up later. On most tools it stays there until you delete it, or until an auto-delete period runs out.

  3. 03

    A person might read it

    Some providers have trained reviewers read a sample of chats. Google says reviewed Gemini chats are separated from your account and kept for up to three years.

  4. 04

    It might teach the next AI

    Personal accounts can use chats to improve future models unless you opt out. Anthropic, for example, keeps Claude chats for five years if you allow training, or 30 days if you don't.

  5. 05

    "Delete" isn't always the end

    Copies can be kept for safety checks, and occasionally for legal reasons, for a while after you hit delete.

Business accounts are different. Business plans from the major providers generally don't train on your chats by default and come with admin controls. More on that below.

When postcards go astray

Two stories worth knowing

Neither of these involved hackers or anyone doing anything wrong. That's exactly why they're useful.

Summer 2025

Shared chats turned up on Google

ChatGPT let people share a conversation as a link, with an optional tick-box to make it findable by search engines. Plenty of people ticked it without realising what it meant, and private-sounding conversations started appearing in Google results. OpenAI switched the feature off, saying it had created "too many opportunities for folks to accidentally share things they didn't intend to".

May to September 2025

Deleted chats kept for a court case

During a copyright case brought by the New York Times, a US court ordered OpenAI to preserve ChatGPT conversations, including ones users had deleted. Enterprise and education accounts were exempt. The obligation ended in September 2025, but for a few months "delete" didn't mean what people assumed.

Both were resolved, and both providers responded sensibly. The lesson isn't that AI companies are careless. It's that once words leave your hands, what happens next isn't entirely up to you.

Why it matters more at work

At home, the postcard carries your secrets. At work, it carries other people's.

Clients, donors, staff, service users. They trusted your organisation with their details, not the AI tool someone opened on a busy afternoon. And under UK GDPR, personal data you put into AI is still covered by data protection law, with all the same responsibilities attached.

8.5%

of the prompts employees sent to AI tools contained potentially sensitive information. Nearly half of that was customer data, and over a quarter was staff data.

Source: Harmonic Security, analysis of business AI use, Q4 2024

The question isn't "is this AI safe?" It's "would I write this on a postcard?"

The reassuring bit

This is one of the easiest risks in tech to manage

You don't need to stop using AI, and you don't need a security team. You need three habits, and ideally a short agreement across your team about how you'll use it.

Habit 1

Put it in an envelope

Use business accounts for work. They generally aren't used for training by default and give you admin controls. If you're on Microsoft 365, sign in to Copilot with your work account and look for the green shield.

Habit 2

Write less on it

AI rarely needs the real names. Swap them for "Client A" or "a volunteer", describe the situation rather than pasting the whole file, and leave out anything that identifies a person.

Habit 3

Check the settings once

On personal accounts, switch off the training setting. For one-off questions, use a temporary chat: ChatGPT's aren't used for training and are kept for up to 30 days for safety checks.

Agreeing it as a team? Our piece on Shadow AI covers what goes into a simple AI acceptable use policy.

Screenshot this

The postcard test

Before you press send, ask:

  1. 1Would I be comfortable if a stranger read this?
  2. 2Does it include someone else's personal details?
  3. 3Is it confidential to my organisation?
  4. 4Am I signed in to a work account?

If you'd hesitate on 1, or answer yes to 2 or 3 while on a personal account: take the details out, or switch to a business account.

Everything else? Send away. That's what it's there for.

Our take

AI is one of the most useful tools to arrive in working life in a long time, and we use it every day.

Thinking in postcards doesn't make it any less useful. It just means your team uses it the way they'd handle any other information that matters: with a little care about what goes where.

That care is what lets clients, trustees and staff trust how you use AI. And it's far easier to build those habits now, while AI is still new in your organisation, than to unpick them later.

Working alongside you

Want your team using AI with confidence?

We'll help you find out what's already in use, move it onto proper business accounts and agree simple rules that everyone can follow. ISO 42001 certified, no jargon, no hard sell.

Keep reading

More from Our Thinking

See all articles

Written by

Digital Panda

Website, IT, security and AI for SMEs and charities. ISO 42001 certified.

Sources