Have you wondered what happens to deleted chats in Gen AI models? I simply figured that they would hang around forever. The short answer? BoodleBox safeguards your privacy on deleted chats better than other chatbots…or so it seems to me. Am I wrong?

Let’s take a look and see how each chatbot handles your deletion request.

Did You Know?

Sign up for TCEA’s AI Amplified Educator Accelerator, a totally online learning opportunity for K-16 educators. You get six months of access via BoodleBox Unlimited to learn the skills you need to build bots (a.k.a. similar to Perplexity Spaces, OpenAI Custom GPTs, Google Gems, etc.). It’s only $149 and you get 22 hours of CPE credit, digital badge and certificate. Learn more about all of TCEA’s Gen AI professional learning offerings.

What Happens to Deleted AI Chats?

Updated May 19, 2026

I used to assume deleted AI chats disappeared. They mostly don’t. In most consumer tools, “delete” means the chat vanishes from your view right away. Backend deletion, training use, and legal retention vary by platform and license.

The table below describes what each vendor intends to do by default, but not what’s guaranteed. A court order can override a vendor’s deletion policy with no notice to you, of course. As of January 2026, there’s at least one example of that. Sigh.

The proof case: OpenAI’s deleted chats aren’t deleted

OpenAI returned to its standard 30-day deletion practice on September 26, 2025. But that’s not the whole story. In the New York Times copyright litigation, a federal judge ordered OpenAI to preserve and produce a sample of 20 million consumer ChatGPT conversations. The chats are randomly drawn from December 2022 through November 2024. District Judge Sidney Stein affirmed that order on January 5, 2026. OpenAI is also holding a separate set of consumer ChatGPT and API data from April–September 2025 under legal hold (OpenAI’s own explanation).

Deleted chats from those windows are included. Opting out of training did NOT exempt them. The data is de-identified and locked under court-supervised access. A user’s delete button does not control the outcome. The court does, which should give anyone pause.

What this does not touch: ChatGPT Enterprise, Edu, Business, and API customers with a Zero Data Retention agreement were excluded from the order. That exclusion is the strongest argument in this whole post for using school-approved Enterprise/Education licenses.

The table: vendor default policy, not a guarantee

Platform / License	Deleted from View	Backend Retention (Default)	Training Use	What I’d Watch
ChatGPT Free / Plus / Pro	Yes	Standard practice is deletion within 30 days, but consumer chats from Dec 2022–Nov 2024 and Apr–Sep 2025 are under court-ordered legal hold and cannot be deleted (OpenAI)	May be used for improvement unless training is turned off; Temporary Chats are not used (OpenAI)	The litigation hold is the real risk, not the delete button. Memory is separate — delete it separately (OpenAI)
ChatGPT Enterprise / Edu / Business	Yes	Governed by workspace/admin settings; explicitly excluded from the NYT preservation order	Business data is not used to train models by default (OpenAI)	Best fit for school-approved sensitive workflows
Google Gemini Free / Paid Consumer	Yes	Activity settings control retention; chats kept 72 hours even when activity is off (Google)	Human-reviewed chats are kept separately, disconnected from your account, up to 3 years — and are not deleted when you delete activity (Google)	As of early 2026, Gemini defaults to Gmail/Chat/Meet access for US consumer accounts (“Personal Intelligence”). Don’t enter anything you wouldn’t want reviewed
Gemini Workspace Education / Enterprise	Yes	Admin-set auto-delete (3 / 18 / 36 months; default 18) (Google)	Governed by Workspace protections / admin controls (Google)	Check your district admin configuration directly — defaults are not privacy-minimal…Check your district admin settings and TIGHTEN the the default 18-month retention.
Claude Free / Pro / Max	Yes	30 days by default; 5 years if you opt in to model improvement (Anthropic)	Not used for training unless you opt in; default is off (Anthropic)	Verify “Help improve Claude” is off in Privacy Settings; it’s the difference between 30 days and 5 years
Claude Enterprise / Education	Yes	Custom retention may apply (Anthropic)	Commercial plans are not used for consumer-style training (Anthropic)	Use admin-controlled retention

Bottom line for schools

The delete button is a UI convenience, not a privacy strategy. The OpenAI case proves a vendor’s stated policy can be frozen by litigation you’ll never be told about, with deleted and opted-out consumer data swept in.

Practical guidance that holds up:

• Use approved Enterprise/Education licenses — they were the ones excluded from the court order, and that’s not a coincidence.
• Turn off training/model-improvement where the consumer tier allows it (Claude, ChatGPT), and treat consumer Gemini as reviewable by humans by default.
• Keep student and staff PII out of consumer tools entirely. Assume anything typed in could be retained beyond the stated window.
• Document your local guidance and tell staff why: not because the vendors are careless, but because vendor policy is not the same as a guarantee.

Super Important Point

For schools, don’t rely on the delete button as a privacy strategy. Use approved Enterprise/Education licenses, disable training where needed, avoid student/staff PII in consumer tools, and document local guidance.

Action checklist

• Hard rule: No PII in consumer/free AI.
• Standardize on an approved assistant (Copilot EDU or Gemini EDU) and publish the allowlist/blocklist.
• Use local tools for recordings/IEP notes/transcripts when possible.
• Train staff: “Would you email it unencrypted? If not, don’t paste it into free AI.”