Spelling checks made software correction unremarkable

The writer's problem

Long before any writer weighed whether to disclose AI assistance, word processors already made a quieter editorial decision on every page: which words to flag as probably wrong, and which grammatical patterns to question, without ever silently rewriting the sentence itself. That design choice, flag and let the writer decide, shaped what readers and writers alike came to expect from 'the software,' and it is the baseline against which every later, more capable writing tool now gets measured.

What the documents show

Microsoft's current developer reference for Word's CheckSpelling method, retrieved 16 September 2026, states that the feature 'begins a spelling check for the specified document or range,' checking 'all available stories,' meaning the main text plus headers, footers and text boxes, and that it displays the Spelling and Grammar dialog box rather than making changes on its own. Its companion, the CheckGrammar method, works the same way for grammar, opening the same dialog with the grammar option selected. Both methods take optional parameters, such as whether to ignore uppercase text or always suggest alternatives, that a user or a document's settings control rather than the program deciding on its own.

Read together, the two pages describe a mechanical, rule-based check, comparing text against a dictionary and a set of grammar patterns, that surfaces candidates for a human decision rather than resolving them outright. Neither page claims the check understands meaning, tone or factual accuracy; both describe it strictly as detection and suggestion, with acceptance left to the person at the keyboard.

The editorial choice

Treating a spelling or grammar flag as a suggestion requiring a decision, not as an automatic correction, is the inherited editorial norm this feature established decades before generative drafting tools arrived. It is an editorial choice to extend that same norm, flag and decide, to newer tools that go further than spelling, rather than assuming any software-generated suggestion should be accepted by default.

What stays with the author

Every accepted or rejected flag remains the writer's own decision under this design, and the documentation does not address, because it predates, questions about disclosing when a suggestion originated from a more generative or interpretive tool than a dictionary lookup.

• Does the tool flag an issue for a decision, or does it change the text before anyone reviews it?
• Would a reader assume a flagged suggestion was accepted automatically or reviewed first?
• Is the checking mechanism a lookup against fixed rules, or something that interprets meaning?

A dialog box asking a writer to accept or reject a single flagged word looks unremarkable now precisely because this feature spent decades normalizing the idea that software can point at a problem without deciding it, a norm worth naming before assuming a newer tool still honors it.