A spell checker is the cheapest quality control you will ever run on a piece of writing. It takes seconds, it costs nothing, and it catches the class of error you are least able to see in your own work. It is also routinely misused — treated either as a rubber stamp that means the piece is finished, or as an authority whose every suggestion must be obeyed.
It is neither. This piece covers what a Spell Checker can actually do, the three categories of error it will never catch, and the short manual pass that closes the gap.
What a spell checker is actually doing
At its simplest, a checker tokenises your text and asks whether each token appears in a dictionary. If it does not, the word is flagged and the tool suggests near neighbours — words a small number of edits away from what you typed. That is the whole mechanism behind classical spell checking, and it has not changed much in forty years.
Modern checkers add a second layer: rules and statistical models that look at the words around a token. This is what lets a good checker flag their in “their are three reasons” even though their is a perfectly real word. It is also why checker quality varies so much. The dictionary layer is close to a solved problem; the context layer is where the differences live.
Our checker uses the LanguageTool engine, which combines a dictionary with several thousand hand-written grammar and style rules. Your browser talks to it directly, which means two things worth knowing: your text does not pass through our servers, and the free service applies its own rate limit if you check very long passages repeatedly.
The three things it will always miss
1. Real words in the wrong place
The dominant category. Form for from, manger for manager, public missing a letter. Context rules catch the common pairs and miss the rest, and confidence collapses in technical writing where the surrounding words give the model little to work with. We catalogued the worst offenders in 12 Spelling Mistakes That Slip Past Every Spell Checker.
2. Anything that is factually wrong
A checker has no idea whether the price, the date, the name or the statistic is right. These are the errors that cost the most and the ones no automated pass will ever find. Verify them separately, against the source, every time.
3. Your own vocabulary
Product names, internal jargon, industry abbreviations and regional spellings all get flagged. The danger is not the false positive itself; it is what happens after the tenth one, when you stop reading the list and start clicking past it. That is precisely when a real error goes through.
Use the right language variant
This is the single highest-value setting and the one most people leave on automatic. English (US) and English (UK) disagree about color and colour, organize and organise, traveled and travelled, and about where punctuation sits relative to quotation marks.
Automatic detection works well on long, unambiguous passages and badly on short ones. A 200-word product description with three proper nouns in it can easily be detected as the wrong variant, at which point the checker produces a long list of confident, useless suggestions — the exact condition that trains you to ignore it.
Set it explicitly. It takes one click and it removes most of the noise.
A five-minute pass that works
- Run the checker first, not last. Clearing the genuine misspellings early means your later read-through is not distracted by them.
- Read every suggestion, apply the ones you agree with. Spelling and agreement suggestions are nearly always right. Style suggestions about wordiness are a matter of taste, and a checker cannot tell that your fragment was deliberate.
- Read the piece aloud. Non-negotiable, and by far the highest-yield step. Homophones stop making sense when spoken, which is exactly what the checker could not tell you.
- Search for your own repeat offenders. Everyone has three or four. Use find-and-replace to check every instance before publishing.
- Check the numbers against the source. Prices, dates, names, statistics. Nothing automated will do this for you.
Five minutes, and it catches the overwhelming majority of what would otherwise have been published. The step people skip is reading aloud, and it is the one that does the most work.
Where it fits with the other tools
Checking is the last step, not the first. If the draft still needs cutting, the Word Counter tells you how far over you are and how long the piece takes to read — Word Counter: How Long Should a Blog Post Really Be? covers what target to aim for.
If a passage has been reworked heavily, check the seams. Errors cluster where an original sentence meets an edited one, and that is doubly true of anything an Article Rewriter has touched — see Article Rewriter: What Rewriting Tools Can and Cannot Do for why.
And once the words are right, the technical side is a separate job: crawl waste and page weight decide whether anyone gets to read the thing at all.