What Data Do AI Apps Actually Collect? A Factual Comparison

When you type a prompt into an AI app, where does that text go? Who can read it? Is it used to train the next version of the model? These are reasonable questions, and the answers vary significantly across services.

We looked at the publicly available privacy policies and documentation of the major AI services as of April 2026. Here's what we found.

ChatGPT (OpenAI)

Account required: Yes. Email, phone number, or third-party sign-in.

What's collected: All conversations are stored on OpenAI's servers. According to OpenAI's privacy policy, they collect prompts, responses, uploaded files, and usage data.

Training: By default, conversations in the free and Plus tiers can be used to train future models. Users can opt out via Settings > Data Controls > "Improve the model for everyone." Team and Enterprise tiers are opted out by default.

Retention: Even with training opt-out, conversations are retained for up to 30 days for abuse monitoring, then deleted. With the "Temporary Chat" feature, conversations are not used for training but are still retained for the monitoring window.

Third parties: OpenAI may share data with service providers, and in certain cases, with Microsoft as outlined in their enterprise agreements.

Claude (Anthropic)

Account required: Yes. Email or third-party sign-in.

What's collected: Conversations are stored on Anthropic's servers. Their usage policy and privacy documentation state that inputs and outputs are collected.

Training: Anthropic states that free-tier conversations may be used for training, safety research, and product improvement. Pro and API users' data is not used for training by default. Users can opt out of training for the free tier.

Retention: Conversations are retained for safety monitoring purposes. Specific retention periods vary by plan.

Third parties: Anthropic uses cloud infrastructure providers (AWS, GCP) to host their services.

Gemini (Google)

Account required: Yes. Google account.

What's collected: Conversations are stored and associated with your Google account. Google's Gemini Apps Privacy Notice states that prompts, responses, location data, and feedback are collected.

Training: By default, conversations with Gemini may be reviewed by human annotators and used to improve Google's products and machine learning technologies. Users can turn off Gemini Apps Activity, but this means conversations won't be saved in your activity history (Google may still retain them for a limited period).

Retention: When Gemini Apps Activity is on, conversations are retained for up to 18 months by default (adjustable to 3 or 36 months). Human reviewers may see conversation content.

Third parties: Google's broader data ecosystem means Gemini data interacts with Google's advertising and product infrastructure, though Google states Gemini conversations are not used for ad targeting.

Google AI Edge Gallery (On-Device)

Credit where it's due: Google's AI Edge Gallery runs models locally on Android devices, similar to Cortis's approach. When running in on-device mode, prompts are not sent to Google's servers.

Limitations: As of April 2026, it supports a limited set of models with smaller context windows, doesn't integrate with system-level automation (no Shortcuts equivalent on Android that matches iOS App Intents depth), and is still in an experimental phase.

Data collection: The app itself may collect usage analytics (app opens, model downloads, crash reports), but inference happens locally.

Cortis (NERON)

Account required: No. No email, no phone number, no sign-in of any kind.

What's collected: Nothing is transmitted. Zero network calls during inference. All conversations are stored in a local SQLite database on your device. The app does not contain analytics SDKs, tracking pixels, or telemetry.

Training: Not possible. We never see your prompts or responses. There is no mechanism for us to access your conversations.

Retention: Your data stays on your device until you delete it. Uninstall the app, and everything is gone.

Third parties: None. No cloud providers process your data. The one exception: if you choose to use the optional Pro remote inference feature, those specific queries go through our server — but this is opt-in, clearly labeled, and local inference remains the default.

Verification: You can verify this yourself. Run Cortis in airplane mode. Every feature works (except remote inference and model downloads). Monitor network traffic with a tool like Charles Proxy — you'll see zero outbound connections during local inference.

The Comparison Table

| | Account | Conversations stored | Used for training | Human review | Works offline | |---|---|---|---|---|---| | ChatGPT | Required | Server-side | Default on (opt out) | Possible | No | | Claude | Required | Server-side | Free tier default on | Possible | No | | Gemini | Required | Server-side | Default on | Yes, by default | No | | AI Edge Gallery | Not required | Local | No | No | Yes | | Cortis | Not required | Local only | No | No | Yes |

What This Means

We're not saying cloud AI services are bad. They offer more powerful models, longer context windows, and capabilities that on-device models can't match today. Many people reasonably decide that tradeoff is worth it.

But the privacy differences are real and significant. If you're discussing sensitive business information, personal health details, legal matters, or anything you wouldn't want stored on someone else's server — that's worth considering.

The choice should be informed. Now you have the facts. For a deeper comparison of on-device AI options, see Cortis vs Google AI Edge Gallery or our roundup of the best free AI apps for iPhone.

Try Cortis

Cortis is available for free on the App Store and Google Play. No account required, no data collected. Download a model over Wi-Fi once, and everything runs on your device from that point forward.

---

NERON LLC builds AI tools that respect your privacy. Cortis is our on-device AI assistant for iOS and Android.