Kirra Pendergast · 01/08/2025 · 3 min read
We’re living through a moment where digital instincts haven't caught up with digital reality. A moment where a feature can feel like a private diary, but act like a press release. And most people have no idea it's happening.
ChatGPT’s “Share” button is one of those design sleights-of-hand that seems harmless. Elegant. Convenient. You’ve just had a conversation with the AI, maybe it helped you draft a policy, unpack a trauma, or name the thing you couldn't say out loud. It wraps the entire thread in a link and offers it up: clean, clickable, ready to pass on. And you do, send it to a colleague, a client, a classroom. Click. Copied. Shared.
But here’s the part the interface doesn’t tell you: those links are public. Not semi-public. Not “only if you have the URL.” Public. Search-indexable. Discoverable. Crawled by Google. And increasingly, they’re turning up in live search results.
Paste the right search string and you’ll surface complete ChatGPT conversations, no account names, sure, but full context. Real queries. Real answers. Real risks.
Now, think carefully about what people are putting into these exchanges.
These aren’t sanitised brainstorming sessions or harmless hypothetical debates. These are unfiltered disclosures: abuse histories, panic spirals, identity crises. Private medical queries. Legal what-ifs. Draft resignation letters. Notes to dead parents. Teachers navigating disclosures. Kids navigating identity. It’s not content, it’s people, mid-process.
They’re typing into this box because it feels safe. Not performative like social media. Not surveilled like a search. A liminal space, quiet, responsive, seemingly contained.
But that feeling? It’s a design illusion.
Because when that link gets indexed, by accident or indifference, that illusion breaks. The damage isn't theoretical. It’s immediate. And while you might be “anonymous,” the breadcrumbs are real: phrasing, context, location cues, references. It doesn’t take much to connect the dots if someone is motivated. Or careless. Or cruel.
This isn’t an indictment of OpenAI or a rejection of the tool. It’s a call to close the gap between what people think the system is doing, and what it’s actually doing under the hood.
So, What Do We Do?
If you’re someone people turn to for guidance on tech, safety, digital wellbeing this is your cue to step in. Not with fear, but with clarity.
Don’t just warn people. Show them how the system works. Teach them where the guardrails fail. Give them the language to talk about digital trust in terms that matter.
Here’s where to start:
**Never assume the Share button is private.**If you wouldn’t publish it on a website, don’t send it via a shared ChatGPT link. If it’s raw, intimate, legal, or painful it stays out.
**Stop entering identifiable information.**Not just names. Think: diagnoses, workplace incidents, school names, birthdays, subtle phrasing. These aren’t just details, they’re identifiers in the wrong hands.
**Turn off memory.**This doesn’t erase what’s already there, but it stops the model from pulling your past into your present. It narrows the trail.
**Opt out of training.**In your settings, you can choose not to have your chats used to train future models. It’s not perfect, but it’s a start.
**Audit your privacy settings regularly.**OpenAI is moving toward greater transparency, but default settings don’t protect everyone. Make checking those settings part of your digital hygiene.
The Core Truth
“Anonymous” isn’t the same as “protected.”
The real risk lives in the gap between what users expect and what the system does. That’s where harm blooms. That’s where people get blindsided.
People trust what feels private. But the interface doesn’t show you the index bot. It doesn’t flash a warning when your trauma turns into a URL.
So treat every interaction with AI like it could one day be read aloud in a courtroom, or in front of your board, or by your 13-year-old.
Because for too many people already, it has.
Don’t let a private moment become a public fallout.
Build the literacy. Spread the warning. And keep pushing for infrastructure that matches the stakes. Contact us for up to date education and support - hello@ctrlshft.global
More from the journal
9 Oct 2026 · 5 min read
Who is checking that the child is a child?
If device makers become the authority that labels people as children, what stops an adult who wants contact with children from being labelled a child? On current evidence, this is a real gap, and almost no public guidance addresses it.A few years ago, over lunch in London, I sat across from a member of Pornhub’s senior management team. I had come to talk about one thing: the sheer volume of fetishised Roblox content on their platform. Roblox avatars, recorded in-game performing sexual acts, then
8 Oct 2026 · 4 min read
How to Talk to Your Children About Suicide
By George Peterkin - Managing Director of Mind Your Health and Founder SoS BEACON Sadly, conversations about the loss of three very young people has entered a great many Australian homes this past week. The subject has been all over our media, and a news story like this never stays where it started. It travels through social media feeds and school group chats, gathering rumour and half-truths as it goes, until it reaches a child alone with a phone or surfaces as a question from the back seat on