The Human Authored Certification was created to do something simple and important: give writers a way to certify that their work was written by a human.

In a landscape where AI-generated content is everywhere and often indistinguishable from human writing, that certification is only as valuable as the trust behind it. The process of earning it has to be as credible as the certification itself.

The first version of the program was straightforward. Only Authors Guild members were eligible. Members went through a manual vetting process to join, which was more than sufficient to confirm they were human.

The second version changed the equation.

Non-members could now earn the certification, which meant the manual path was gone. The nonprofit needed a way to verify identity at scale, quickly and accurately, without compromising on the values that made the program worth having: privacy, fairness, speed, and mission alignment.

1. The Evaluation Came Before the Tool

A lot of technology decisions start with a demo. Someone finds a promising product, the demo goes well, and momentum builds before the hard questions get asked.

That is not how this one started.

Before evaluating any specific vendor, we established what the right tool would need to do, and what it would need to avoid. The evaluation criteria came from the mission, not from a feature comparison chart.

The questions that shaped the process:

• Does this tool align with the goal of streamlining onboarding while maintaining trust and compliance?
• Can staff and recipients clearly understand how verification decisions are made?
• How does the vendor collect, store, and retain data, and does that meet strict privacy requirements including US and international standards?
• How does the vendor approach bias in their identity verification models, and is that methodology transparent and documented?
• Can staff review, override, or appeal verification decisions when needed?
• What is the vendor's public ethics commitment and security track record?

Only after those criteria were defined did the vendor evaluation begin.

The first step was not demos. It was direct outreach. We contacted vendors with a set of questions before any meeting was scheduled. That process did much of the filtering on its own. Several vendors focused on enterprise-scale deployments or required certification thresholds well above what the Authors Guild needed. They self-selected out. That narrowed the field to vendors actually sized for the use case before a single demo was scheduled.

2. What Made Veriff the Right Fit

Right-sized verification. One of the most important fit questions was not about features. It was about level. Identity verification exists on a spectrum. At the higher end are Level 3 supervised in-person verification programs, specialized AML certifications, and tools built for finance and high-regulatory environments where the stakes and compliance requirements are significantly more demanding.

The nonprofit did not need that. They needed KYC, Know Your Customer level verification: reliable, privacy-respecting identity confirmation appropriate for a certification program. Using a higher-level solution would have added cost and friction without adding anything the use case actually required. Matching the verification level to the actual need, rather than defaulting to the most comprehensive option available, was part of the evaluation from the start.

Verification level and the price are connected. Higher-level programs carry pricing that reflects the compliance infrastructure behind them. The Authors Guild did not need that infrastructure. Choosing KYC-level verification was the right decision for the use case and the right decision for the budget. That tradeoff was presented transparently to the client alongside the recommendation. They understood what they were getting, what they were not getting, and why, before the decision was made.

Transparency. Veriff's portal provides clear documentation of the data entered, the data reviewed, and the final verification decision. There are no black boxes in what the client or recipient sees. Every decision has a record.

Veriff has also proactively notified us when their sub-processors changed. We may not always have the bandwidth to conduct a full review of each change, but the notification itself is a meaningful transparency signal. We know who is in the data-handling chain. If something warrants a closer look, we have what we need to ask the right questions.

Privacy. Veriff's data collection, storage, and retention policies met strict privacy requirements and minimized exposure of sensitive personal information. Compliance with data privacy laws in the US and internationally was confirmed and documented before the integration moved forward.

Bias mitigation. Veriff publishes a detailed, performance-based approach to addressing bias in identity verification. Their public conversation about why bias in machine learning is more consequential than bias in human decision-making was specifically aligned with the nonprofit's mission. Coat Rack is not in a position to train data models at that level. What we look for is whether a vendor is approaching this problem the way we would if we had their resources. Veriff was.

Human oversight. The integration included clear protocols for staff to review, override, or appeal any verification decision. Client administrators can override decisions entirely. Every case is logged for historical review.

That oversight has turned out to be one of the most practically valuable parts of the integration.

Every verification decision is fully visible in the portal: the document uploaded, the assessment made, the outcome reached. If a user made a simple human error, an administrator can ask Veriff to allow them to try again. Nothing is final without a path to review.

That capability has mattered in practice. There have been individuals who contested their verification outcome with some force. In more than one case, a review of the portal revealed they had submitted a photograph of an ID displayed on a computer screen rather than the actual document. The portal made that visible. The administrator could see exactly what was submitted, understand what happened, and handle the situation appropriately, in a human way, inside an AI-powered process.

That is the design working as intended. The AI makes the initial assessment. The human has full visibility and full authority to understand, investigate, and intervene. The system does not remove human judgment. It gives human judgment something real to work with.

3. What the Integration Made Possible

The Human Authored Certification expanded to non-members with a verification process that matches the integrity the certification is designed to represent.

Writers who earn it know that the process behind it was evaluated with the same care that goes into their own work. The organization knows that if a decision needs to be reviewed or overridden, the tools and protocols to do that exist, are documented, and are in the hands of their staff.

There is something worth naming directly here.

The nonprofit used an AI-powered tool, evaluated carefully and integrated thoughtfully, to protect and expand a program that exists specifically to distinguish human creative work in the age of AI. That is not a contradiction. That is what responsible AI adoption actually looks like: a clear use case, criteria that come from the mission, a vendor who can answer the hard questions, and human oversight built into the design from the start.