AI Watermark Detector Access: Who Qualifies and How to Request It

If you've been searching for an "AI watermark detector" and wondering whether you can get access to one for your content compliance workflow, the answer is more specific than most guides let on. Anthropic does offer watermark detection for content generated by Claude - but there are two separate paths, and which one applies to you depends almost entirely on why you need it.

This guide walks through both paths accurately, so you don't waste time applying for gated access you're unlikely to get, or miss the free public tool that may already do what you need.

Step 1: Determine Which Access Path Applies to You

Anthropic has published a support article explaining how Claude marks AI-generated content. The short version: Claude embeds an invisible watermark into text it generates. That watermark is designed to be detectable - but Anthropic has made a deliberate choice about who gets to detect it reliably, and why.

There are two tiers. Understand which one fits your situation before you do anything else.

The free public checker is available to anyone. It's called the Claude Content Checker, and it lives at claude.com/check-content. Paste in a piece of text, and it tells you whether that content carries Claude's watermark. No account required, no application, no approval process. If your use case is "I want to verify whether a specific piece of content was generated by Claude," this tool is probably what you need.

The gated private-preview capability is a separate, higher-confidence or bulk detection system. Access is restricted and requested via a dedicated Google Form. This tier exists primarily for organizations with a legal basis under EU law - regulators, auditors, and entities operating under EU AI Act-relevant compliance obligations. If you're a marketing team at a SaaS company checking your own AI-assisted drafts, this is almost certainly not the path for you.

What this step will NOT tell you: whether your specific request will be approved. Anthropic hasn't published a detailed public rubric for gated approval. The eligibility criteria above are the clearest public signal available.

Step 2: Gather Your Request Documentation

If the free public checker covers your needs, you don't need to gather anything. Go use it. But if you're an organization with a genuine legal or regulatory compliance basis - say, you're operating under EU AI Act obligations and need bulk detection capability - here's what to prepare before you open the request form.

The Google Form will ask for organizational context. Based on what Anthropic has disclosed publicly, you should be ready to provide:

• Your organization name and type (regulator, publisher, compliance body, etc.)

• The legal basis for your request - specifically, your EU law basis if applicable

• Your primary use case: what you're detecting, in what volume, and why high-confidence detection matters for your situation

• Contact information for the right person at your organization

A prepared use case statement will help. Here's a template to adapt:

[ORGANIZATION_NAME] is a [TYPE: regulator / publisher / compliance body] operating under [LEGAL_BASIS, e.g., EU AI Act Article X]. We are requesting access to Claude's watermark detection capability to [PRIMARY_USE_CASE]. Our estimated detection volume is [MONTHLY_VOLUME] pieces of content per month. The primary contact for this request is [NAME, TITLE, EMAIL].

What you get back:

• A clear use case statement you can paste directly into the form

• The organizational context Anthropic will need to evaluate the request

• A record of what you submitted, for internal compliance documentation

What to do with it:

• Verify that your legal basis is accurately stated - vague claims here are likely to slow or block approval

• Have a legal or compliance lead sign off on the use case description before submitting

• Keep a copy of everything you submit, including the date

• If you're unsure whether your organization qualifies, read Anthropic's support article again before submitting - the form is not a general inquiry channel

What this step will NOT tell you: whether your legal basis is sufficient. That determination sits with Anthropic's team, and they haven't published the approval criteria publicly.

Step 3: Submit Your Access Request

The submission channel for gated access is the Google Form Anthropic has provided for this purpose. There is no email alias or platform dashboard for this - the form is the official path.

Before you open it, have your use case statement and organizational details from Step 2 ready. The form is not long, but submitting a vague or incomplete response is worse than waiting until you're prepared.

Here's a template for your use case field:

Organization: [ORGANIZATION_NAME]
Type: [REGULATOR / PUBLISHER / COMPLIANCE BODY / OTHER]
Legal basis: [EU AI ACT / OTHER EU LAW BASIS - BE SPECIFIC]
Use case: We need bulk or high-confidence watermark detection to [DESCRIBE WHAT YOU ARE AUDITING AND WHY].
Estimated monthly volume: [NUMBER] pieces of content
Contact: [NAME], [TITLE] - [EMAIL]

What you get back:

• A form submission confirmation from Google (save this)

• Follow-up from Anthropic's team if your request moves forward - timeline is not published publicly, so build in reasonable buffer if you have a compliance deadline

• Either approval with activation instructions or a response explaining why the request doesn't qualify

If you're a marketing team using the public checker, this step doesn't apply. Open claude.com/check-content, paste your content, and read the result.

Step 4: Activate and Test Your Detector Access

This section covers both paths, because the activation experience is completely different depending on which tier you're using.

If you're using the public Claude Content Checker: There is no activation step. Navigate to the tool, paste the text you want to check, and run the detection. The output will tell you whether the content carries Claude's watermark. That's the workflow.

Here's the test flow:

1. Go to https://claude.com/check-content
2. Paste [YOUR_CONTENT_SAMPLE] into the input field
3. Run the check
4. Note the output: detected / not detected, plus any confidence signal provided
5. Document the result if you need an audit trail

What you get back:

• Detection result: whether the content carries Claude's watermark

• A fast, no-friction check you can run on any piece of content you're evaluating

If you've been approved for gated access: Anthropic will send activation instructions as part of the approval communication. Follow those instructions as provided - the setup process for the private preview capability is not publicly documented, so work from what you receive directly.

What to do with it:

• Run a test on a known Claude-generated sample first to confirm the tool is working as expected

• Document your test methodology for compliance purposes

• Set up a repeatable workflow for your team so detection checks are consistent, not ad hoc

• Keep records of what you checked, when, and what the result was - especially if this feeds into regulatory reporting

What this step will NOT tell you: the technical mechanism behind the watermark. Anthropic hasn't published the implementation details publicly, and for good reason - see the next section.

The Whole Loop on One Page

Here's the full process at a glance, because most people land on this page with a specific question and need to move fast.

Step	Input Required	Output	Typical Time
1. Determine your path	Your use case and legal basis	Know whether you need the public tool or the gated form	10 minutes
2. Gather documentation	Org details, legal basis, use case statement	Completed request package	30-60 minutes (gated path only)
3. Submit request	Completed documentation or text to paste	Form confirmation; follow-up from Anthropic	15 minutes to submit; review timeline unknown
4. Activate and test	Approved access or public tool URL	Detection results, working workflow	Under 5 minutes for public tool; varies for gated

Cadence note: If you're using the public checker, check Anthropic's support documentation quarterly - the tool's capabilities and any policy changes will be reflected there first. If you have gated access, review your use case documentation any time your compliance scope changes, and re-contact Anthropic if your organizational basis shifts. There's no public renewal schedule documented for the private preview.

This loop assumes you're dealing with text content that may have been generated by Claude specifically. The watermark system is Claude-specific - it doesn't detect content from other AI models.

Where This Breaks

A few honest failure modes worth knowing before you start.

You apply for gated access as a marketing team and get rejected. The private-preview capability is oriented toward organizations with EU law compliance obligations - regulators, auditors, entities operating under the EU AI Act. A marketing team wanting to check whether competitor content is AI-generated, or to validate their own drafts before publishing, is almost certainly not the intended use case. The recovery path: use the public Claude Content Checker instead. It's free, it works, and it's the right tool for most marketing use cases.

The public checker returns an unexpected result. The public tool checks for Claude's watermark specifically. If content was generated by a different model - GPT-4, Gemini, Mistral - the checker won't detect it, because there's no Claude watermark to find. That's not a bug. If you need to detect AI-generated content broadly, not just Claude-generated content, this tool is the wrong solution. No single publicly available tool reliably detects all AI-generated text with high confidence.

You need bulk detection and the free tool doesn't scale. The public checker is a single-text interface. If you're running compliance checks on thousands of documents, the manual workflow breaks fast. The gated path exists for exactly this reason - but as noted above, it's not open to general marketing teams. If you're in this position and don't have an EU law basis, the honest answer is that there's no publicly available path to bulk Claude watermark detection right now.

The watermark is absent from older content. Anthropic's watermarking system applies to Claude-generated content going forward from when it was implemented. Content generated before the watermark was introduced won't carry it, so a "not detected" result on older content doesn't confirm it wasn't AI-generated - it just means there's no watermark present.

Why is gated access structured this way at all? Anthropic is explicit about the tradeoff: if a highly reliable detector were freely available to everyone, bad actors could use it to reverse-engineer how to strip the watermark - testing content until detection fails, then using that version. Restricting the high-confidence tool to organizations with a verified need keeps the watermark harder to game. The public tool is useful; the gated tool is for higher-stakes verification where the integrity of the detection system itself matters.

If you've submitted a gated request, completed all steps, and haven't heard back after a reasonable window, reach out to Anthropic's support team with your form submission date and any confirmation you received.