Of the 48 AI tools Yuzoki tracks, 32 say they use what you type or upload to train or improve their models unless you do something about it. Only 15 of them give you a switch in settings, 6 make you send an email or fill in a form, and 11 describe no way to opt out at all on the free plan. For anyone who pastes client notes, contracts or unreleased plans into an AI tool, that difference decides whether privacy takes ten seconds or is not on offer.
The answer in one table
Every tool falls into one of five groups. The first three use your content; the difference is how much effort it takes to stop them.
| Group | Tools | What it means |
|---|---|---|
| Used unless you switch it off in settings | 15 | One setting, usually under Privacy or Data controls |
| Used unless you email or fill in a form | 6 | Possible, but you have to ask |
| Used, and no way out is described | 11 | No training opt-out on the free plan, or you must pay to stop it |
| Says it does not train on your content | 7 | No setting needed, by the maker's own statement |
| Not clear from the official pages | 9 | Default not stated, or the pages conflict |
That makes 32 of 48 tools where your content can end up in training unless you act, or with no way to stop it. Only 7 say plainly that they do not train on it.
The 15 with a switch, and where it is
These are the tools where privacy takes one setting. The names below are copied from each maker's page, because the switch is rarely called "training".
| Tool | Setting to turn off |
|---|---|
| ChatGPT | Settings, Data controls, Improve the model for everyone |
| Claude | Settings, Privacy, Help Improve our AI models |
| Gemini | Gemini Apps Activity, Keep Activity |
| Perplexity | Account settings, Preferences, AI data retention |
| Grok | Settings, Data (Data Controls in the app), Improve the Model |
| DeepSeek | Improve the model for everyone |
| Vibe (Mistral) | Settings, Data & Account Controls, Enable data sharing |
| Doubao | Settings, Privacy and permissions, help improve the model |
| GitHub Copilot | Copilot settings, Allow GitHub to use my data for AI model training |
| Bolt | Settings, General, Model training |
| Lovable | Account settings, Preferences, AI model training |
| Figma | Team settings, AI, Content training (admins only) |
| Grammarly | Account privacy page, Product Improvement and Training |
| Gamma | Settings, Data Controls |
| ElevenLabs | Terms and privacy, Data use, Improve the models for everyone |
Several makers state the default in plain words. OpenAI says data sharing "is enabled by default" on Free, Plus and Pro. Google says that if you are 18 or over, Keep Activity "is on by default". Perplexity, Grammarly and Figma say the same in plain words for their individual or entry plans. Anthropic's pricing page lists model training as "Opt-out" on Claude's Free, Pro, Max and Ultra plans. Bolt is the one tool in this group that changes the default by country: it is off for accounts based in the EU, the UK and Switzerland.
The 17 that make it hard or impossible
Six tools accept an opt-out only on request. FLUX asks for an email with the subject "Training Opt Out", HeyGen and Kling take requests by email, Kimi asks you to contact customer service and pass an identity check, Zapier links a form, and Meta AI takes an objection through its Privacy Center. Meta says the forms you see depend on your region, and we read its page from the EU, so readers elsewhere may not be offered the same option.
Eleven tools describe no training opt-out for a free user. Midjourney's own summary of its training data answers "Yes" to whether user prompts were used to train its models. Runway's terms say inputs and outputs "may be used by the Company to train and improve its AI models". Suno, Pika, Luma, Otter, Qwen and ERNIE list training or model improvement as a use with only general privacy rights on offer. The free plan of DeepL uses your texts to train its models, and the only way out is DeepL Pro. v0 shows "Data Opt-out: No" for its Free, Premium and Plus plans. Replit uses prompts and responses to improve its models and says the opt-out "is available to Pro users", so a free user has to pay to get the switch.
Generators are heavily represented here. Five of the eleven (Midjourney, Runway, Suno, Pika and Luma) make images, video or music, which matters for anyone uploading product shots, a face or a voice.
The fine print that catches people
Switching off training does not reach backwards. Claude, Perplexity, Grok, ElevenLabs, Meta and Replit all say on their pages that the opt-out covers new or future content only. Anything you shared before the change may already be in a training run.
Feedback is a back door. ChatGPT says that if you rate a reply with a thumbs up or down, "the entire conversation associated with that feedback may be used to train" its models, even after you opt out. Grok and Gemini Notebook say the same about feedback. If a chat is sensitive, do not rate it. ChatGPT's voice mode also has its own training control, separate from the main setting.
Business plans are the cleanest switch. Of the 15 tools with a setting, 11 state that their business, team or enterprise plans are excluded from training by default. Google's privacy page for Gemini says work and school accounts "may be subject to different data handling terms" without saying more, so check your admin console there.
Nine tools are unclear, including two that many offices use daily. Microsoft Copilot's current privacy controls page for the app updated on 18 August 2026 lists no training switch and states no training policy. Canva uses your content only "if this is consistent with your Privacy Settings" without saying what the setting starts as. Cursor, Cohere, Descript, Elicit, GLM, Hailuo and Ideogram complete the group. The seven that say they do not train on your content are Adobe Firefly, Fireflies, Notion, Synthesia, Make, n8n and Gemini Notebook, which excludes feedback you choose to give.
We think every AI maker should state its training default in one sentence on its pricing page, as OpenAI's help page and Anthropic's pricing table already do. A setting you cannot find is a setting most people never change.
What it means for your role
- If you run marketing or creative work: treat the image, video and music tools in the no-way-out group as public. Do not upload unreleased product shots, client logos or a colleague's voice to them without a contract that says otherwise.
- If you are a founder or run operations: list the tools your team uses and check them against the table above. Every tool in the first group takes one setting; put that setting in your onboarding checklist.
- If you work in finance, legal or HR: use a business or enterprise plan for anything with personal data. Eleven of the 15 tools with a switch state that those plans are excluded from training by default, and one contract is easier to defend than 30 personal settings.
- If you are a developer: GitHub Copilot Free, Pro, Pro+ and Max may use your interactions for training since 24 April 2026. Check the setting before your next commit.
Do this in the next ten minutes
- Open the three AI tools you use most and find the setting named in the table. Turn it off if you share work content.
- Stop rating sensitive chats with a thumbs up or down.
- For anything confidential, use a private or temporary chat where the tool offers one. ChatGPT says temporary chats are not used to improve its models.
- If your team relies on a tool in the request, no-way-out or unclear groups, ask the maker in writing what it does with your content.
Our step-by-step guide to keeping your data private in AI tools walks through the settings for the biggest assistants, and writing an AI policy for your company covers the rules to set for a team.