Does Marcus train AI models on my data?

Contractual guarantee, not just policy

Marcus commits in writing that your data stays out of training pipelines. This appears in the Terms of Service you accept when you create an account. We do not treat training exclusion as a courtesy or a temporary policy—it is a binding obligation.

The scope covers everything you type into Marcus: project prompts, page instructions, design tweaks, and any content your end users submit through forms or inputs on published sites. None of it feeds back into model training, whether ours or a third party's.

What counts as your data

We define your data as any content you or your site visitors create inside Marcus. This includes prompt text you write when building pages, the HTML/CSS/JavaScript Marcus generates for you, uploaded images and files, form submissions from published sites, and analytics metadata like page views.

General anonymized usage telemetry—such as which features get clicked most often or average session length—does not include your content and may inform product improvements. But no prompts, no code, no user-submitted text ever appears in aggregated datasets used for model fine-tuning or research.

Anthropic Claude and commercial API terms

Marcus uses Anthropic's Claude models via their commercial API. Under Anthropic's current policy, API inputs and outputs on commercial plans are excluded from training. This means the requests Marcus sends to Claude—your prompts and the code Claude returns—do not contribute to future model versions.

Anthropic publishes these terms publicly, and we rely on the commercial-tier safeguards. If their policy were to change, we would notify customers and offer migration options. As of today, your content does not train Anthropic's models.

No human review by default

Your prompts and generated output are not reviewed by Marcus staff unless you explicitly grant permission for a support case. If you contact support and share a project link, we will ask before looking at the content. Automated error logs capture stack traces and system events, not your prose or code.

We do not run sentiment analysis, keyword mining, or quality scoring on your projects to train classifiers or ranking algorithms. Marcus improves through direct feature development and user feedback, not silent data mining.

Data retention and deletion

Projects you delete are purged from live databases within 30 days. Encrypted backups may retain deleted data for up to 90 days for disaster recovery, after which it is overwritten. You can request a full data export or accelerated deletion by contacting support.

If you cancel your subscription, all project data is deleted according to the same 30-day schedule unless you explicitly export it first. We do not keep an archive of churned customer projects for research or competitive analysis.

Third-party integrations and exports

When you export a project as a ZIP file, that code leaves Marcus's infrastructure and becomes subject to wherever you host it. If you connect external analytics tools, email services, or payment gateways to a published site, those providers operate under their own data policies.

Marcus does not share your project data with third-party analytics or advertising networks. The only external AI service we use is Anthropic Claude, covered above. Any future integrations requiring data sharing will be opt-in and disclosed in advance.

Questions and transparency

If you want to verify our data handling practices, email hello@aimarcus.love and ask specific questions. We will answer directly, not with boilerplate. If a future product feature required any change to training exclusions, we would notify all customers by email at least 60 days before the change took effect and allow you to opt out or export your data.

Privacy is not a marketing angle for us—it is the default. Your work belongs to you, and we have no interest in using it to train models, sell insights, or improve anything except the direct service you are paying for.