You typed the whole thing in one go. Dentist for Esha next Tuesday half four, remind me the day before.

The app kept Tuesday. It ignored the time, left Esha’s name in the title where it does nothing, and set no reminder at all. So you opened the item and filled in three fields by hand, which is what you were trying to avoid.

Typing naturally is two different features wearing the same name, and the gap between them is most of your evening admin.

Two features, one name

The older one is a date parser. It scans your line for phrasings its authors wrote down, highlights the match, and lifts it into the date field. It is fast, it works offline, and it is right or obviously wrong within a tenth of a second.

The newer one hands the whole sentence to a model and asks for fields back. Slower by a second, and it reads the parts a parser never looks at: who the thing is for, which calendar it belongs on, how long it will take, whether you asked to be warned.

Both are sold as “add tasks by typing naturally”. Only one of them reads the whole sentence you typed.

What a date parser actually reads

Credit where it is due, because this is a genuinely good piece of engineering and it has been in these apps for a decade.

Todoist reads dates in ordinary phrasing, including complicated recurring ones like every other Tuesday starting in March. Any.do’s Smart Type picks up tomorrow, next week and Friday at 9am with no symbols at all. Fantastical goes furthest: a date, a time, a length, a place, an alert, a repeat, and who is coming, all from one line.

The one that catches people out is Apple Reminders. Ask Siri and it understands you. Type the same sentence into the app and you get date pickers, because on Apple’s own help pages the typed route is fields and a calendar, and the natural language route is the microphone.

Everything that is not a date is punctuation

Here is the part the phrase “natural language” quietly covers up. The date is read from English. The other five fields are read from symbols, and you have to know them.

Fantastical does the same thing with a slash for the calendar and exclamation marks for priority. Any.do uses a hash for tags. None of this is a criticism of the design, which is compact and quick once it is in your fingers.

Learning five symbols is fine for one person. It is a non-starter for the second adult in the house, and they are the one you needed.

That is the household problem in a sentence. A syntax survives in an app one person uses every day for work. It does not survive contact with a partner who opens the app twice a week, which is the same reason the shared calendar you set up in March still has one user.

The four ways it goes wrong

Both kinds fail, and they fail in predictable places. Three of these are yours to avoid.

1. A word in the task looks like a date. Todoist documents this one itself: type Create monthly report and monthly can be read as a repeat. Anything with weekly, May, March or Sunday in its name is a candidate.
2. Two dates in one line. Book the MOT before the 14th, garage opens Saturday has two, and you will get whichever one comes first. Split it.
3. Ambiguous phrasing. Next Friday means this coming Friday to about half of people and the one after to the rest. Nobody can fix this for you. Write the number.
4. The day boundary. Type tomorrow at eleven at night and an app resolving that in the wrong timezone gives you the day after. This one is invisible until it bites.

The fourth is worth a note because it is the app’s fault rather than yours. Any quick add worth using resolves relative dates against the household’s own timezone and today’s actual date, not against a server clock somewhere else.

How to type so it lands

Five habits, and they work in every app in this piece. None of them take longer than typing badly and fixing it afterwards.

• Thing first, then when. Dentist Tuesday 4:30, not Tuesday 4:30 dentist. Parsers read left to right and so do you.
• Digits for the time. 4:30pm is understood everywhere. Half four depends entirely on whose list of formats the app was built from, and it takes ten seconds to find out which yours is.
• One date per line. Two dates is two tasks.
• A number beats a word whenever you catch yourself typing next.
• Watch the highlight. Every good quick add box marks the bit it understood, live, while you type. That highlight is the whole safety net.

The ten-second test for any app

Type one real sentence with a person’s name and a reminder in it, the messiest thing you would actually say. Then look at what came out the other side and count the fields you still had to fill in yourself. That number is the feature.

The screen between typing and saving

A model reading your sentence gets more of it, and it can also be confidently wrong in a way a parser cannot. A parser that fails to match half four leaves the words alone, visibly. A model that misreads it gives you 4:00 with no fuss and no clue.

So the thing to look for is what happens in the second between typing and saving. Good: it shows you the item it built, with every field it filled, and lets you tap one of them to fix it. Bad: a spinner, a green tick, and a thing you now have to go and find to check.

This matters more the more the app got right. If it filled six fields, you want six fields on screen, because checking one wrong field is cheaper than losing trust in the box and going back to forms.

What Chirpy does with the sentence

In the calendar there is a box that takes a line of English. It sends the whole sentence, and it gets back an appointment with the title, the day, the start time, the length, the place, the repeat and the people named in it, matched against the members of your actual household.

The date arithmetic is not left to the model to do in its head, which is the classic way these features file an appointment in the wrong year. Today’s date and the household’s timezone go in as facts, so Tuesday means the right Tuesday and eleven at night stays on the day you typed it.

Being straight about the limits: that box is the calendar’s. Lists in Chirpy take plain lines rather than parsed sentences, so a shopping list is typed as a shopping list. If what you want is a dated task for yourself and nothing else, Todoist’s parser is faster than any of this and it is free to try.

Where the household version pays for itself is the fields a to-do app has no place to put. Who it is for. Which of you is driving. Whether the whole house needs to see it, which is the difference between a swimming lesson on the grid and an evening that quietly eats two and a half hours.

It is the same principle running through every tool where the AI sits inside the thing you were already doing instead of in a chat box you have to remember to visit.

Common questions