The invisible typos: catching colour/color mixes, e-mail vs email, and other consistency slips

Run your document through a spellchecker and it comes back clean. Every word is a real word, correctly spelled. And yet something is off: page two organises a workshop that page five organizes, the intro talks about e-mail while the FAQ says email, one table reports revenue of 1,000,000 and another of 1.000.000, and "Product Manager" appears three ways in as many paragraphs. None of these are typos. All of them are errors — consistency errors, the kind spellcheckers are structurally blind to because every individual instance is valid.

Readers rarely consciously catch these slips, but they register them. A document that drifts between conventions reads as assembled rather than written — and often that's exactly what happened: three contributors, two style guides, one paste from an old deck. For anything with multiple authors or multiple drafting sessions, a consistency pass is a distinct editing step, separate from proofreading, and it's one a machine does far better than a tired human eye.

British and American spelling: the classic tell

The most recognizable inconsistency is dialect mixing: colour and color, centre and center, realise and realize, catalogue and catalog in the same document. It happens innocently — a UK writer edits US boilerplate, a paragraph arrives from a chatbot with different defaults, autocorrect fights the author's instincts — but it is the single fastest way to signal that nobody did a final read.

The fix requires a decision before it requires an edit: pick a target dialect. If you publish for a US audience, every -our and -ise is a deviation even when it's used consistently; for a UK audience, the reverse. A good checker therefore works against a declared target locale, not just internal agreement — flagging both true mixes (both spellings present) and words that are consistent but wrong for the market you chose.

Hyphenation drift: e-mail, email, and the compound problem

Compound terms migrate over their lifetime: e-mail became email, web-site became website, on-line became online. Style guides update at different speeds, and writers carry whichever era they learned. The result is documents that use start-up and startup interchangeably — sometimes in the same sentence — and unlike dialect, there is often no "correct" answer, only a consistent one. Detecting this mechanically means finding every hyphenated term, checking whether its closed-up form also appears, and flagging the pairs that occur both ways so a human can pick a winner.

Capitalization: is it the Board or the board?

Capitalization inconsistency is subtler because sentence position legitimately changes a word's case. The interesting signal is a word that appears capitalized and lowercase multiple times each, mid-sentence — Agile vs agile, Government vs government, a product name that sometimes gets its trademark treatment and sometimes doesn't. In legal and policy writing this can even change meaning, where "the Agreement" is a defined term and "the agreement" is just a noun. A checker that requires each variant to appear more than once avoids drowning you in sentence-initial false positives and surfaces the genuine coin-flips.

Numbers and dates: the internationalization trap

Number formatting is where multinational documents betray themselves. English convention writes one thousand as 1,000; much of continental Europe writes 1.000; ISO and scientific styles use a thin space, 1 000. Mix two of these and some readers will misread a value by three orders of magnitude — 1.500 is one and a half in one convention and fifteen hundred in another. Dates are worse: 04/07/2026 is April 7th to an American and 4 July to nearly everyone else. A document that mixes slash-dates with "July 4, 2026" and ISO 2026-07-04 forces the reader to re-derive the convention paragraph by paragraph. The safe rule for international audiences: spell the month out, always, and pick one thousands separator document-wide.

Running the sweep

The text consistency analyzer automates exactly this pass. Paste your text (at least 30 words, though it earns its keep on full documents), choose US or UK English as the target, and it checks five families at once: British/American spelling pairs with counts for each side, mixed thousands separators, words capitalized both ways mid-text, hyphenated terms that also appear closed up, and coexisting date formats. Everything rolls up into a consistency score out of 100 — each distinct issue costs ten points — with a clean sweep rated green, up to three issues yellow, and more than that red. The score is a summary; the tables under it are the actual to-do list, showing each flagged pair with occurrence counts so you know which variant is currently winning.

Deciding the winner is your job, and it's usually easy: keep whichever variant matches your style guide, or failing that, whichever the document already uses most. Then fix globally — a case-sensitive find-and-replace per issue — rather than correcting instances as you happen to read past them, which is how half-fixed documents are made.

Consistency is a habit, not a heroic edit

The cheapest time to run this check is right before anything ships: the blog post before it's published, the proposal before it's sent, the chapter before it goes to the editor. It takes seconds and it catches the class of error that survives every other tool in your pipeline, precisely because nothing is misspelled. Run your next draft through the text consistency analyzer and see what three drafting sessions left behind — then, with conventions settled, the cliché detector and filler word detector cover the neighbouring passes: not whether your text agrees with itself, but whether it says anything worth agreeing on.