Draftsmith Review: An AI-Driven Aide for Editors

As generative AI developments continue to dominate headlines and permeate every aspect of the knowledge economy, their use poses unique concerns for editors. I’ve given two well-attended talks now for Editors Canada on what AI means for us, and both share a key takeaway that I still stand by: great human editors aren’t going anywhere. 

Generative AI doesn’t stand a chance of replacing us, for myriad reasons—namely, its capabilities and appeal appear to be hugely overstated, its limitations and risks generally downplayed, and its environmental impact disastrous. But (and this is the annoying part) it’s not entirely without its merits, or it wouldn’t be so darn popular.

For me, these merits still aren’t worth the formidable risk that AI poses to our planet and to my clients’ intellectual property. But what if you could tap into some of these benefits without the risk?

That was the intriguing proposition presented to me by the team behind Draftsmith, an AI-powered editing assistant that positions itself as a more conscientious and purpose-built alternative to traditional large language model (LLM) tools.

After playing around with the product for a few weeks, I’m delighted to share my unbiased opinion of it because (spoiler) I think Draftsmith represents the most productive step toward effective AI-assisted editing yet.

This review looks at how Draftsmith stacks up against its larger, flashier counterparts—and whether its precision-focused approach actually delivers better value for editors and anyone else who works on professional documents. I found that Draftsmith distinguishes itself across five key dimensions:

• Environmental impact

• Security/privacy

• Usability with Word

• Reading comprehension

• User authority

I’ve organized this review to tackle each of these metrics in detail, and summarized my takeaways in a table at the end of this post.

1. Environmental Impact: Lightweight by Design

The AI arms race is racking up staggering environmental costs. Traditional LLMs consume massive computational resources, which in turn require vast supplies of energy and water. One study estimated that a single string of prompts uses about 500 millilitres of water due to cooling requirements. (Please consider this next time you’re tempted to ask ChatGPT what your cat might look like as a Marvel superhero.)

As an eco-conscious consumer, I was relieved to discover that Draftsmith takes a different path. Unlike the popular platforms powered by expansive general-purpose LLMs, Draftsmith is powered by an SLM—a small, specialized language model optimized for the narrower task of editing sentence-by-sentence within Word documents. This design choice leads to significantly lower resource consumption. It’s a meaningful distinction, especially for universities, publishers, and individuals looking to align their tech stack with their sustainability goals. I think it also sets a valuable precedent for how industry-specific tools can compete with less versus more (streamlined efficiency > bloated excess).

2. Security and Privacy: Protecting IP at Every Step

One of the most frequently cited concerns for AI use among editors—for good reason—is data privacy. LLM platforms harvest user input for continued training or retain it in ways that create severe risks, even when companies claim they don’t.

For writers, editors, and researchers dealing with sensitive or proprietary work, this is a dealbreaker. Checking a box in the terms and conditions absolutely will not protect your (or your clients’) valuable IP. As an example of the risk this creates, one of my own clients recently secured a 6-figure consulting contract with a company in the finance industry looking to apply his research. If I were to enter his papers into ChatGPT for a quick (and bad) grammar check, there is every chance that that information would eventually be spit out by the platform for another user who asks it to write a graduate essay or industry report on the same topic—in which case, there goes my client’s 6-figure contract, and quite possibly his entire career since it’s largely based on that ground-breaking research.

Draftsmith directly addresses this risk, in several ways.

• First, the tool is working with no more than a paragraph at a time. Some may see this “zoomed in” view of the document as a weakness; I think it’s brilliant. Even the biggest and best-funded AI companies haven’t produced a tool with true contextual understanding, so Draftsmith’s creators were wise not to try and tackle that challenge. Instead, they’ve opted to fill the huge gaps that bigger AI platforms leave in data protection and privacy with this narrower focus on smaller bites of text.

• Second, Draftsmith’s servers don’t actually store your text—they pass it on to Microsoft Azure’s OpenAI servers. And, again, these servers only ever see one small piece of the “puzzle” that represents the full IP at a time. In addition, Microsoft (for all its flaws) has a strong legacy of data security and protection, as evidenced by its compliance with EU data boundary and data residency laws.

• Finally, the “puzzle pieces” aren’t just isolated from one another, but also encrypted, making it virtually impossible for even the platform itself to access the full information.

Make no mistake, you should still get your clients’ explicit permission if you’re going to use Draftsmith to edit their text. But I believe the protections in place are adequate even for unpublished research, documents under nondisclosure agreements, and internal corporate documents.

3. Usability with Microsoft Word: Seamless (for Some)

Admirable principles aside, let’s get into the details of actually using Draftsmith, because this is where Word users will finally feel seen.

For editors, Microsoft Word remains (somewhat astonishingly) the industry standard. It’s the only tool that allows for comprehensive, detailed edits and author review and approval of those edits. Yet most LLM-based platforms treat the editing process as an afterthought—offering clunky copy+paste workarounds or third-party plug-ins that steamroll existing formatting.

Again, Draftsmith flips this script to target editors in particular. It’s built specifically as a Microsoft Word add-in and operates natively within the Word interface. Even better, it supports and uses Track Changes, something no major LLM-powered editor currently does well. This small detail is monumental for editors, given that the Track Changes function is really what allows us to offer not just corrections but far more valuable collaboration with our clients. It also tells you when formatting is at risk of being overwritten by its edits.

An important caveat: Draftsmith will not work with pre-2016 versions of Word or pre-2022 versions of Word for Mac. Since I'm on an older MacBook, I was able to try out the product using Office Online, which posed some tech issues for me that won't affect those able to install the add-in on their computer.

4. Reading “Comprehension”: Precision Over Pseudo-Understanding

One of the defining features of LLMs is their ability to "understand" context across vast swaths of text. This capability is great when you need a hasty overview of a textbook—not so great when you’re making high-level changes to content that must be polished and accurate.

Draftsmith deliberately limits its scope, allowing you to review no more than one paragraph at a time. I appreciate this approach because it mirrors how readers read and human editors edit: one line at a time with the immediate surrounding text as the primary context.

While restricting its “comprehension” to the immediate context might seem like a limitation, it’s actually a strength—especially for writers looking for line-level clarity and grammar support, not conceptual rewrites (which AI still doesn’t do very well). This constrained reading comprehension leads to more granular, controllable edits, and avoids the kinds of sweeping alterations that LLMs tend to introduce. You’re less likely to get restructured arguments or inaccurate summarizing, and more likely to get tight, targeted edits that clean up sentence structure or alter the tone without guessing at your intent.

For instance, check out this sentence, accurately simplified with Draftsmith’s Plain English Converter. The suggested text is too casual for some contexts, but for a task like turning research into public-facing, readable reports or articles, this function could be very useful:

I did get a few suggestions that misconstrued meaning, but they weren’t nearly enough to put me off Draftsmith because of the final and most important metric...  

5. User Authority: Every Edit Is Still Your Edit

Most AI tools overwrite or paraphrase text with a haphazard “black box” approach, leaving users who care about quality backtracking and trying to figure out exactly what’s been changed and why. Even careful prompt engineering provides only a semblance of real “control” over the output. Examining these processes—let alone undoing them—is time-consuming work that makes me wonder who could possibly find these platforms useful for editing.

Draftsmith, by contrast, marks every change clearly in advance of actually making it, making it easy for the user to accept, reject, or revise its suggestions.

This ability to preview and peruse AI edits without losing control (and without leaving the Word document) is hands down Draftsmith’s most valuable feature. It makes the tool feel like a natural extension of professional writing and editing workflows, not an experiment requiring side-channel workarounds. It also invites a more collaborative, creative relationship with AI, where the user decides which tool to use and how, and remains the final decision-maker over even the smallest decisions.

The fact that Draftsmith’s edits are initially suggested (not actually made) means it’s easy and fun to play around with its various functions. “Punchier” (an option under the “Engagement Tuner” in the add-in ribbon) was one of my favorites and made many edits I would actually make (e.g., changing “The study has 3 contributions” to “This study makes three key contributions”)—but watch out it doesn’t cut important sentences that it deems irrelevant or not “punchy” enough.

Certain functions will have outright negative effects in certain contexts (all of the “Readability Tuner” options oversimplified the sample academic text that I fed it, and even removed vital citations entirely) so editors will need to experiment to discover which ones they like for their specific type of editing. I suspect the “Polish” and “Remove Typos” functions (under “Editing Helper”) will probably be the ones in heaviest rotation with professional editors.  

It’s no surprise to me that these two functions made the fewest mistakes of any I tried, because Draftsmith comes from the same company (Intelligent Editing) that offers PerfectIt, which many editors already rely on for polishing and removing typos. Both were so light-handed that they really only changed outright errors: ideal, in my opinion, for a final pass just to catch anything tired eyes missed. 

Conclusion: A Useful Tool for Overworked Editors Who Have the Right Tech Stack

Anyone hoping to have Draftsmith take over their edits and complete an excellent line edit of 10,000 words in a few minutes will be sorely disappointed. No such AI tool exists and, like the team behind Draftsmith, I doubt it will ever exist. But for writers and editors who want a helping hand with their work and value precise control over convenience, Draftsmith is the best AI option I’ve seen yet. It’s not the most broadly capable tool—and that’s the point. This add-in avoids the pitfalls of overambition by focusing on a very specific use case: helping users make clean, context-sensitive edits directly within Word, with full transparency and minimal risk. Its light environmental footprint, strong privacy protections, integration with Word, focused editing scope, and commitment to user authority make it a welcome alternative to the big-name LLMs.

The product wasn’t without bugs (what tech is?). At one point in my tests, it inexplicably turned an entire sentence into two random words from that sentence. However, the fact that it only ever proposes—and doesn’t actually make—glitchy changes means there’s nothing to undo, making it a lot easier to forgive or tolerate bugs in this format.

I deeply appreciate that Draftsmith doesn’t even attempt to emulate a skilled editor. Rather, it’s clearly designed to support human editorial agency. In this way, Draftsmith is not unlike an AI version of the tools we’ve always used to edit—a thesaurus, dictionary, or style guide—something you pick up when you’re stuck or burned out, so you can do more of this extremely cognitively demanding work than you could without it. If you have the right setup, don’t hesitate to give it a try.

Metric	Other AI Tools	Draftsmith
Environmental impact	Powered by large language models, using a staggering amount of resources (especially water)	Powered by a small language model, using lower levels of resources
Security/privacy	User data is used for model training, often even when users are told it won’t be; IP should be considered at risk if entered into any popular AI platform	User data is not stored or used for model training; it is also isolated and encrypted so that IP is well protected
Usability with Word	Can be used within Word via third-party add-ins (few of which are free), but can’t use/involve Track Changes at present and often obliterates formatting	Operates right in Word as an add-in and applies Track Changes; warns when there’s a risk of overwriting formatting
Reading “comprehension”	Takes the entirety of what’s presented and makes broad changes that reflect its “understanding” of the overall content	Takes no more than 1 paragraph at a time and makes specific, “zoomed in” changes to each sentence based on that limited context
User authority over each individual change	Possible with careful prompt engineering and significant time/effort toward double-checking all outputs	Guaranteed