Avoid AI privacy mistakes with general precautions before you share data. This guide makes no claim about how often users make privacy errors.
We checked the linked vendor sources on October 14, 2026. This post reports no comparative product test.
1. Check ChatGPT model-training controls
Check the service settings before you submit text. In ChatGPT, disable "Improve the model for everyone" to exclude new chats from model training. OpenAI states that it can still use the conversation for model training if you submit feedback on a response. See the OpenAI data controls guide.
2. Get approval for sensitive material
Check your organization rules before you submit client documents, source names, credentials, or internal code. Use only an approved app and data configuration. Local execution does not replace an access review.
PocketLLM is pre-launch. Private, on-device AI, starting on iPhone, with more platforms planned. No account required, no in-app tracking, and chat processing stays on your device. Join the waitlist.
3. Check ChatGPT account requirements
Do not assume that every service needs a phone number. OpenAI states that phone verification is not generally required for ChatGPT. See its phone verification guide. Check the account and workspace before each sensitive task.
4. Check offline access
Download the required model files first. Enable airplane mode. Disable Wi-Fi separately if it remains active. Test the exact task with non-sensitive text. A successful response does not prove that the app sends no data later.
5. Review old chats
Check saved chats, exports, sync, backups, and the service retention policy. Delete records that you no longer need under your retention rules. Do not assume that deletion from the screen removes all copies.
6. Inspect each image
Check screenshots for names, email addresses, private URLs, and visible documents. Remove sensitive details before upload. Inspect the final image after the edit.
7. Review audio separately
Check the audio policy for transcripts, audio retention, and model-training choices. Choose text or audio according to the task and the applicable policy. This guide assigns no universal privacy rank to either format.
8. Read the app policy
Check the app privacy policy, permissions, and optional network features. Do not treat an offline response or a store label as a complete network audit.
9. Set rules for shared devices
Use separate device access where appropriate. Check account access before a child uses an AI service. Follow the school policy and the service age rules. A local app still needs supervision.
10. Separate Chrome Incognito from ChatGPT retention
Chrome Incognito limits local browser history, but websites can still observe activity. See Google Chrome Help.
OpenAI states that Temporary Chat excludes chats from history and model training while they remain temporary. OpenAI may retain a copy for up to 30 days for safety. A saved chat follows regular account settings. Data sent through GPT actions follows the recipient policy. See the Temporary Chat guide.
Set the local conditions
Local inference keeps prompts on the device during inference when the selected model runs there. For a local-only workflow, select no cloud model. Disable chat sync and cloud backup. Check the app telemetry policy. Disable optional network features. Review extensions, exports, and local storage access.
Frequently asked questions
Can OpenAI use feedback after I disable model training?
OpenAI states that feedback on a response can permit model training on the associated conversation. See the OpenAI data controls guide.
Does ChatGPT always require phone verification?
OpenAI states that ChatGPT does not generally require phone verification. See its phone verification guide.
Can websites observe activity in Chrome Incognito?
Chrome Incognito limits local browser history, but websites can still observe activity. See Google Chrome Help.
PocketLLM is pre-launch, with a waitlist for the iPhone launch. More platforms are planned. See the PocketLLM model catalog for its separate model list.