Writer launches customer-specific watermarking for AI text

Writer has introduced a text watermarking system that assigns invisible, customer-specific marks to AI-generated text, giving enterprise customers control over how generated content is identified and traced.

The system applies to text produced by Writer's Palmyra models when customers enable watermarking. In the European Union, it will be switched on by default to align with the EU AI Act. Customers outside the bloc can choose whether to use it.

The move comes as AI providers and their business customers face growing scrutiny over how they disclose AI-generated content. Watermarking has emerged as one way to label or trace such material, though approaches vary widely across the industry.

Writer said its approach differs from model-level watermarking because it ties the mark to the customer organisation rather than the underlying large language model. The company is positioning the watermark as a way for enterprises to show ownership of content generated using internal brand standards, customer data, business rules and compliance requirements.

Control options

Customers will not be limited to a simple on-or-off setting, Writer said. Where regulations permit, enterprises will be able to adjust watermark strength for different use cases and decide which types of content should carry a mark.

Text watermarking works by influencing a model's word choices to create a pattern that software can later detect. Writer said standardised methods can force customers to accept a fixed trade-off between how easily a mark can be detected and the quality of the final text.

Under Writer's model, a bank could choose stronger detection for customer disclosures, while a retailer might prefer weaker detection for marketing copy, where tone and wording are more sensitive. Different teams within the same organisation may also use different settings depending on the material they produce.

Writer said the system will show customers how watermarking affects output quality by content type. Subject to regulatory limits, users will also be able to choose where watermarking is applied, such as on blog posts but not advertising copy, or on email campaigns but not product descriptions.

The company is also introducing a multi-bit watermarking method that can embed custom information in marked text. Writer said this could include identifiers such as a marketing or advertising campaign ID, allowing customers to trace where material appears and how widely it circulates.

May Habib, Chief Executive Officer and Co-founder of Writer, said the design reflects demand from businesses for greater control over AI governance.

"Enterprises don't want their model providers making their AI decisions for them. They want ownership, control, and configurability, right down to how and when their AI-generated content is watermarked," said May Habib, Chief Executive Officer and Co-founder of Writer.

Habib criticised broader industry approaches that attach provenance primarily to the model provider rather than the end-user organisation.

"The labs' hatchet approach to watermarking is just the latest proof that they don't get the enterprise. They checked the compliance box without a second thought about how customers wanted their content marked. We've built hand-in-hand with the enterprise for six years and that partnership has shaped a very different roadmap," Habib said.

Regulatory pressure

The launch comes amid a tightening policy environment around AI-generated content, particularly in Europe. Companies using AI for regulated communications, customer-facing disclosures or marketing are under pressure to document how material is produced and reviewed.

Roger Beharry Lall, Research Director: Advertising Technologies and SMB Marketing Applications at IDC, said businesses are likely to seek tools that fit existing governance structures.

"As regulators extend disclosure and transparency expectations to AI-generated content, enterprises will want control over how they apply these requirements," said Roger Beharry Lall, Research Director: Advertising Technologies and SMB Marketing Applications at IDC.

"They will start to look for tools that fit the governance frameworks they already use to manage vendors and data," Lall said.

Writer said the watermarking feature will be available to customers from December 2. EU enterprises will have it enabled by default, while other enterprise customers will be able to turn it on at any time.