|
|
Was this forwarded to you? Get The Fourth Estate AI Brief in your inbox →
|
|
|
|
View this issue in your browser
Friday Deep-Dive
Nine AI privacy settings you may have never opened
You may be pasting things into AI tools every day: meeting notes, a client's ad copy, drafts of stories that have not run yet. The chat box feels like a private notepad. It is closer to a form that sends everything to a company's servers, where it is stored and, on most personal accounts, used to train the next version of the product. The settings that control that exist on every major platform. They are free, and almost nobody has opened them. This week: the nine that matter, each one checked against the company's own help pages, plus the list of things a publication should never paste no matter what the settings say. Credit to @jackcoder0 on X, whose thread got me started on this one.
|
Your AI chat box is not a private notepad
ChatGPT, Claude, Gemini, Copilot, and Grok all have settings that decide whether your conversations train the model, what the tool remembers about you, and how long your chats stick around after you delete them. On most personal accounts, training is switched on until somebody turns it off. The fix takes about five minutes. In the full post: the exact menu path to turn off training on all five platforms, how to see what memory has stored about you, where the private chat button lives on each tool, why delete does not mean gone, the seven things a publication should never paste into any chatbot, what that February court ruling about AI and legal privilege actually said, and a five-step setup you can hand to your whole staff.
A publication handles more confidential material than almost any small business in town: sources, unpublished stories, advertisers' numbers, subscriber lists. Every one of those is one tired paste away from sitting on a tech company's server. Nobody needs to stop using AI to fix that. Somebody at your publication just needs to spend five minutes in the settings, and everybody needs to know the short list of what never goes in the box.
Source →
The nine settings, platform by platform →
|
|
What is?
Training Data
What it is: Training data is the enormous pile of text, images, and conversations an AI company feeds a model to teach it how to write and answer. The model does not file your chat away like a folder in a cabinet. It learns patterns from millions of examples, and on most personal accounts your conversations can be some of those examples. Once something has been used to train a model, it cannot be pulled back out. That is why every training switch only affects what happens from the moment you flip it.
Why publishers care: Two reasons. The practical one: on most personal accounts, training is on until someone turns it off, and your staff are the ones typing meeting notes and client details into those boxes. The one closer to home: training data is exactly what publishers have been fighting AI companies over in court, the question of whether they can learn from our work without asking. That same question applies, in miniature, to every chat your staff has. The difference is that for your own conversations, the opt-out already exists and takes one click.
|
|
|
|
Like what you see here, please forward to someone who you think would enjoy reading about AI in publishing!
|
|
Trevor
Trevor Slette
Co-founder, Quadd.ai · 28-year community publisher
trevors@quadd.ai
·
Book a 15-min call
·
LinkedIn
P.S. Forward this to whoever at your publication uses AI the most, then hit reply and tell me which setting surprised you. I will share the best answers in a future issue.
|
|
|
|
|
Run a newspaper. Use Quadd.
AI tools built for small newsrooms. Document intelligence that turns court reports and box scores into copy you can paste into your layout. Audio transcription that finds the quote without you scrubbing for it. AP-style proofreading that tightens copy without flattening voice. Built by someone who knows what a Tuesday night feels like. Free for seven days.
Request a 7-day trial →
|
|
|
