Are AI Drafts Really Faster?

ETEN Innovation Lab
Feb 26
6 min read

New Data Says Yes.

Before we talk about quantifiable acceleration for assisted translations and AI drafting, the Innovation Lab has been considering terminology to describe different types of AI translations. Our goal is to set clarity around purpose, quality, and readiness, and we welcome input as these terms continue to take shape.

We refer to AI-Assisted Translations as an umbrella term for initial drafts of text generated using artificial intelligence. They are not meant for use or publication and represent an early, provisional step in the translation process.
Exploratory Renderings are AI-assisted translations created specifically for experimentation and learning. They are provisional, unrefined, and unreviewed, generated using unknown or variable techniques, involve little to no human intervention, and exist to explore possibilities rather than function as translations.
Draft Zero is a historical term for very early AI-generated drafts. Unlike exploratory renderings, draft zeros involve intentional human engagement in the process and are typically further along in quality, ready to enter a formal review process, though they still require significant refinement by human translators.

The AI-assisted drafts discussed below, produced by Serval/Scripture Forge, can be best understood as draft zeros. To learn more about this terminology distinction, you can access the full document here.

With our current pace, the All-Access Goals are projected to be accomplished by 2041, creating an eight-year gap from the 2033 vision. AI drafting is widely believed to help close that gap—but does the data confirm it?

You’ve most likely heard encouraging stories about faster drafting, how Scripture Forge and Serval (the AI element driving the tool) is now in the utilizing stage of exploration and has been widely piloted across projects (see appendix). Translators on the ground in places like Africa and Asia have shared positive feedback like these quotes:

“The AI Scripture files seem to be working perfectly. May God bless this new beginning.”
“AI is making drafting easier and faster than manual drafting… People in the community are accepting the text and our team’s confidence level is also increasing.”
“The translation work is progressing so well that we must keep revising our plans to include more books. I’m now hopeful I’ll have the whole Bible in my language in just a few more years—in my lifetime!”

And through data collection efforts to inform AI, there are opportunities to expedite even low-resource languages to use Serval/Scripture Forge for their projects. This docuseries on translation work in Tanzania is a first-hand account of exactly that.

The anecdotal evidence points to AI drafting increasing the pace to achieve the ETEN All-Access Goals and see people everywhere gain access to quality Scripture in their language. Yet you may still be asking the question that the SIL Language Technology Team asked recently: Can we show this in the data?

The answer they found? Yes.

Early metrics looking at 75 projects using Serval/Scripture Forge indicate AI is significantly accelerating progress in translation efforts.

From Stories to Evidence: How We Measured Impact

As the Innovation Lab and ETEN partners experiment with new tools and methods, it’s important to move from the early stages of exploration to the later ones. At some point, we must decide: Is this useful for the Bible translation movement? Is this something we should scale?

The challenge with innovation isn’t just having new tools. It’s knowing when those tools are genuinely helping. Every Bible translation organization, and often every project, tracks progress differently. So, to evaluate the use of AI drafting in projects using Scripture Forge, the SIL Language team did something different: instead of relying on self-reported progress, they mined the actual Paratext project history.

Using Paratext’s version control data (with appropriate permissions and anonymization), they:

Built a baseline of about 3,000 standard translation projects, filtering out things like training projects, revisions, adaptations, and study Bibles.
Defined four common stages of a project (based on what’s happening in the text, not how a team chooses to label it):
- Drafted – every verse in a book has some content.
- Back translation drafted – if there’s a related back translation, all the verses in the book have content.
- Reviewed – the book has gone through the team’s chosen review processes.
- Completed – final edits and updates for that book are finished.

By watching how verses move through these stages over time, they created a shared yardstick for progress that any project can be measured against, regardless of organization, region, or internal terminology. Instead of “comparing apples to oranges” as Mike Cochran, SIL’s Director of Language Technology, described the initial challenge, they now had a usable data pool to conduct research in.

Zeroing In on AI Drafts

With a baseline in place and permissions from several organizations, the next step was to look specifically at projects using AI drafting through Serval in Scripture Forge. Since this tool has only been available since January 2024, the team narrowed the focus to projects that:

Have been actively drafting in the last 18 months.
Clearly brought AI-generated drafts into Paratext by sending them directly from Scripture Forge or exporting and importing them into their translation project.

This narrowed the focus to 75 projects where AI drafting was visibly integrated into real translation work. (Note: These 75 are a subset of the 600+ projects onboarded to Scripture Forge, because many of those are back translations, tests, or are ambiguous in the data.) With these projects to zero in on, the team could finally start answering the question everyone’s been asking:

Does Scripture Forge’s AI drafting actually move translation work through the pipeline faster?

What the Data Shows So Far

The short answer? Yes—AI drafting is accelerating translation work, and in some stages, the impact is dramatic.

While still early in the research phase and not having many projects with completed books or testaments to look at yet, the data reveals four main areas of insight:

1. Drafting Moves Nearly Twice as Fast

Because Scripture Forge instantly produces a draft with content in every verse, projects using it reached the “drafted” stage roughly twice as fast as comparable projects in the 3,000 baseline. This aligns with what translators have been saying on the ground: the time between “a blank page” and “a complete first draft” collapses from months of manual work to days.

But drafting pace is only the beginning of the story.

2. Review Progress Increased by Around 25%

The more meaningful metric is how quickly projects move into the reviewed stage—the stage where teams have worked through the AI draft, improved it, and prepared it for checks.

Over the past 12 months, Scripture Forge projects completed about 1,000 more reviewed verses per project compared to baseline projects. Put differently: AI-assisted projects progressed through review roughly 25% faster. Even after the initial drafting advantage, Scripture Forge projects continue to move through the real work of shaping and refining the translations more quickly.

3. Completion Rates Haven’t Shifted Yet (But We Aren’t Surprised)

At the final “completed” stage, Scripture Forge projects don’t yet show a strong difference from the baseline. But this is expected. After all, we’re only looking at 12–18 months of post-AI data. Many projects are working on large Old Testament books (like Psalms), where progress is still in the middle stages. And near-completion doesn’t show up in the metrics yet. As the pipeline fills out over time, this is likely the metric to change last.

4. AI Changed the Overall Trajectory of Projects

When comparing the same 75 Scripture Forge projects to themselves, they went from underperforming to above-average performance. From September 2022 to September 2023 the average verses completed per project was 1,197. Between September 2024 to September 2025 that number jumped to 4,684. This finding suggests that AI drafting may be especially helpful for projects that are struggling, helping them catch up to their goals and move forward more confidently.

Looking Ahead

This is only the beginning. As additional months of data accumulate and more projects consistently use Scripture Forge, SIL will continue refining these insights, tracking long-term trends, and working with partners to evaluate new tools and approaches. The Paratext metrics site and the algorithmic stage definitions are already available for organizations that want to assess their own projects against the same baseline.

For now, the evidence is clear: AI drafting is helping teams translate more Scripture in less time. We have heard this anecdotally from teams, but we’re thankful that the data reflects this same reality. It means we are one step closer to seeing every person have access to God’s Word in a language they understand best.

If your organization is exploring AI-assisted translation—or wondering whether now is the time—these early results offer a compelling case to begin experimenting, learning, and joining the growing community shaping the future of Bible translation. To learn more about AI drafting, connect with the Lab or check out these AI-powered drafting tools: Fluent, Codex, and Serval/Scripture Forge.