A month-long proofreading experiment by a veteran Windows reporter has delivered a startling verdict: free versions of Microsoft Copilot and Google Gemini consistently outperformed Microsoft Editor at catching the kinds of mistakes that actually matter—the ones where a perfectly spelled word is the wrong word entirely. In over 30 days of testing on real articles and documents, writer Sean Endicott found that the previous-generation proofing tool missed context-sensitive errors that the AI assistants flagged routinely, while sometimes suggesting incorrect “corrections” that a human editor would never make.

The Test and the Verdict

The trial, documented at Windows Central, was deliberately narrow: Endicott wrote all his copy himself, then used Copilot and Gemini solely as proofreaders—not as co-writers. He set out to answer a simple question: can modern AI provide a more reliable editorial safety net than the familiar squiggly underlines in Microsoft Word?

The answer, after reviewing thousands of words across breaking news, editorials, and personal documents, was an emphatic yes. The AI assistants caught a significant class of errors that Microsoft Editor missed entirely—errors where the wrong valid English word appeared in a sentence, like “or” instead of “of,” or a tense that didn’t match the surrounding paragraph. Editor, by contrast, remained a solid first-pass spellchecker but repeatedly stumbled on contextual judgment.

Copilot generally impressed at editorial flow—suggesting rephrasings, reorganizing arguments, and pointing out weak transitions—while Gemini proved more methodical and direct with mechanical corrections, often returning cleaner lists of definite mistakes without the conversational fluff. Both, however, required careful prompting to avoid introducing their own quirks, such as overusing “actually,” inserting em dashes, or erroneously “correcting” product names based on outdated training data.

The test was not a laboratory benchmark with controlled error rates. It was one professional writer’s real-world experience over weeks of daily documentation. That makes its conclusions practical rather than statistical—and for Windows users who rely on Editor as their final line of defense, they deserve a close look.

Why Context Changes Everything

A traditional spelling and grammar checker works by pattern matching. It has a dictionary of correct words and a set of rules about what is likely an error. It does not understand the sentence.

When Endicott typed “Footballs wear down over time,” Microsoft Editor flagged “over time” and suggested changing it to “overtime.” A human reader—or an AI language model—recognizes immediately that “overtime” would be nonsense in that sentence. Editor saw two words that often join into a compound noun and blindly applied that fix. Copilot and Gemini left the phrase untouched because they processed the intended meaning.

The reverse error is even more telling. In multiple instances, the writer accidentally typed “or” when he meant “of.” Both are correct spellings; a conventional checker will not highlight either. But an AI that reads the full phrase can detect that “the cost or living” is semantically broken. Gemini and Copilot did so routinely; Editor did not.

This gap is not about better algorithms for known mistakes. It is about a fundamental shift from rule-based proofing to language-model-based reasoning. Generative AI does not just match words against a database; it predicts what word should come next given the whole text. That predictive ability makes it sensitive to context in a way that the older tool class cannot replicate.

What This Means for Your Writing Workflow

The practical implications split cleanly across the typical Windows user base.

Home Users and Students

If your writing is occasional—emails, essays, short reports—Microsoft Editor remains the fastest, most frictionless option. It works inside Word and Outlook without extra steps. For catching blatant typos, repeated words, and outright grammar errors, it is sufficient. The new insight is that it should not be your only check. Even free AI tools can add a second, far more perceptive review for important documents. The extra step takes only a minute: paste your text into Copilot (via Bing Chat or the Copilot app) or Gemini, ask for a list of corrections, and verify.

Power Users and Professional Writers

For anyone producing polished prose regularly—journalists, marketers, technical writers, analysts—the experiment makes a compelling case for a multi-layered approach. Start with Editor to clean up obvious debris while drafting, but never publish without an AI-based review pass. The reason is not just catching more errors; it’s that AI can explain awkward phrasing, identify logical gaps, and suggest structural improvements that go beyond grammar. Copilot excels at this conversational editing, though you must manage its tendency to rewrite everything into a generic AI voice. A tightly worded prompt—explicitly instructing it to preserve your style and only list issues—is essential.

IT Professionals and Developers

Technical documentation, error messages, and UI strings pose a unique challenge: they often contain code snippets, version numbers, file paths, and product names that a grammar checker loves to flag. Running such text through an AI proofreader without clear boundaries can introduce more noise than value. The key is to tell the assistant that technical terms, placeholders, and quoted code are off-limits for correction. Gemini’s ability to return concise, mechanical lists without stylistic commentary may feel more appropriate here, but both tools require vigilance. Also, privacy is paramount: internal docs or unreleased product info must never be pasted into a free consumer chat service unless your organization explicitly permits it. Enterprise versions of Copilot exist for this reason.

The Evolution of Proofing Tools

Microsoft Editor is far from a dumb dictionary. It offers grammar, clarity, conciseness, formality, and vocabulary suggestions, with configurability to turn off categories that don’t fit your style. In Word, it integrates seamlessly with the document, underlining issues in real time. For decades, that capability represented the cutting edge of writing assistance.

But the underlying architecture hasn’t fundamentally changed since the days of “Clippy.” It relies on a set of labeled rules, not on a model that reads the text. Microsoft has begun bridging this gap: Copilot is already available in some Microsoft 365 applications, including Word, where it can summarize documents or rewrite passages. However, the test specifically used the standalone consumer Copilot, not the integrated Word experience—and the standalone version proved dramatically more capable at proofreading than Editor alone.

The telling contrast is between two Microsoft tools: Editor, which lives in the pre-AI world of static rules, and Copilot, which understands full documents. The fact that a Google product (Gemini) also beats Editor underscores that the advantage comes from AI architecture, not from any one company’s secret sauce.

A Practical Multi-Step Editing Process

Drawing from the trial’s successes and caveats, here is a workflow that Windows users can adopt immediately without spending a dime:

  1. Write your first draft without interruption. Don’t chase red squiggles as you go; get the ideas down. Constantly accepting micro-suggestions during composition can fracture your train of thought and homogenize your voice.

  2. Run Microsoft Editor inside Word. Let it catch the low-hanging fruit: misspellings, doubled words, gross subject-verb mismatches. Use its category filters to focus on spelling and grammar first. This takes seconds and clears the obvious errors.

  3. Feed your text to Copilot for an editorial review. Use a prompt that sets firm boundaries, for example:

Act as a professional editor giving a concise, direct critique. Preserve my voice and tone exactly. Flag only: (a) clear logical gaps, (b) weak transitions, (c) buried ledes, (d) unintended tone shifts or repetition. For each issue, tell me the sentence location and why it’s a problem. Do not rewrite passages unless I ask.

Review the suggestions as you would from a human editor: accept those that sharpen your meaning, discard those that push toward a bland AI style.

  1. Run a second pass with Gemini for mechanical precision. Its more straightforward manner saves time here. Try:

List every definite error in grammar, punctuation, and word choice in this article. Correct any tense mismatches, wrong words, and missing articles. Do not comment on style, flow, or structure. Do not alter proper nouns, product names, or version numbers. Present only the original text fragment and the correction.

  1. Verify independently. Both AI tools can hallucinate facts, “correct” unfamiliar product names, and misattribute claims. Microsoft’s own guidance warns that Copilot output must be validated against trustworthy sources. Cross-check any suggested change that affects factual accuracy, proper names, numbers, or quotations. For time-sensitive or technical content, explicitly ask the AI to use web search if supported, but still confirm.

  2. Read the final version aloud—or use Word’s Read Aloud feature. AI catches patterns; your ears catch rhythm, accidental repetition, and those technically correct sentences that just sound wrong.

The Limits and Risks

No tool is infallible. The test revealed that both Copilot and Gemini have a tendency to introduce AI-isms: em dashes everywhere, filler words like “actually,” and a blandly positive tone. If you let them rewrite freely, they will sand away your distinctive voice. The cure is explicit prompting that insists on preservation of tone and minimal intervention.

Another hazard: Gemini frequently tried to “fix” product names because its training data was outdated. Copilot needed less nudging but still required vigilance. For Windows-related writing—with its constant stream of Insider build numbers, rebranded features, and niche hardware—this is a real landmine. Always instruct the tool to ignore proper nouns and version strings.

Privacy is the largest and most overlooked concern. Pasting unpublished content into a free consumer chatbot means entrusting it to a cloud service with data policies you may not have fully read. Microsoft states that consumer Copilot conversation history can be retained for up to 18 months, though users can delete entries. Google similarly uses activity data to improve services when settings permit. For sensitive writing—legal documents, internal business plans, personal data—stop and check whether your usage is compliant. Enterprise versions of Copilot offer more robust data handling, but the free tools are not sandboxes.

Looking Ahead: Microsoft’s Opportunity

The experiment is not a death knell for Microsoft Editor. It is a glaring signal that Microsoft has all the pieces necessary to deliver a contextual proofreading experience that feels native to Word, not a separate chat window. Copilot exists inside some Microsoft 365 apps already, but its current integration is more about generation and summarization than about a silent, Editor-like proofreading overlay. The path is clear: bring Copilot’s language understanding directly into the Word editing pane, with red underlines that actually understand “over time.”

Until that happens, the lesson for Windows users is unambiguous. Microsoft Editor remains a fine first spellcheck. But if you rely on it as your last check, you’re leaving real mistakes—the ones that erode credibility and confuse readers—sitting on the page. Free AI proofreading has already moved beyond that, and the gap will only widen.