Can AI help you type faster?

Sometimes. AI can cut the time to finish a bounded task, such as a short piece of code or a memo with a clear brief. The evidence that it makes everyday typing faster is much thinner. The most direct test we found of an AI phrase keyboard found that overall typing speed was about the same as a conventional smart keyboard, even though the AI saved more presses.

"AI help" covers three different tools, and they behave differently. AI phrase suggestions appear while you write. AI speech recognition turns your voice into text. AI writing assistants draft or rewrite text you then accept. The first is closest to typing, and the studies on it are small.

Bounded tasks show large time savings

Time savings appear where the task has a clear end and the output can be checked. In a randomized experiment, developers who had GitHub Copilot finished a coding task 55.8 percent faster than the control group ¹. That is a coding task, not typing, and the developers still had to read and test the code.

A writing study points the same way. In a working paper, professionals who had ChatGPT took 37 percent less time on a second writing task, a drop of 10 minutes ². The result is for completion time. It is a working paper, and its published version may report different figures, so check that before you quote it.

These results describe finishing a task, not the speed of your fingers. A person who types a long email from scratch still spends the same time on each key unless the AI writes part of it for them.

AI phrase keyboards save presses but not much time

SpeakFaster is a research system that uses large language models to predict whole phrases from abbreviated input. The paper reports that it saved 57 percent more motor actions than traditional predictive keyboards, but only in an offline simulation ³.

When people typed with it on a phone, the picture was smaller. In a pilot with 19 participants who are not augmentative communication users, the overall speed of text entry was similar to conventional smart keyboards ⁴. The speed results for people who rely on augmentative communication are a different group from most readers of this page, so they do not carry over directly.

Gmail already uses a related idea. Google's Smart Compose, built on a large neural language model, gives real-time suggestions in Gmail to reduce repetitive typing ⁵. The paper's abstract describes the system's design and does not report a typing speed or keystroke count, so check each suggestion before you accept it.

Speech is the AI tool with the clearest speed gain

For raw text, speaking is much faster than typing. In a controlled study on an iPhone, English speech entry reached 153 words per minute, against 52 words per minute on the built-in keyboard ^{6 7}. The recognizer in that study was Baidu's Deep Speech 2, a neural network model. It was a 2016 system on short phrases, so treat the gap as a rough guide. Our question on whether dictation is faster than typing covers the trade-off with correction, which affects the speed you get in practice.

AI writing help changes what you write

Speed is not the only effect. In an online experiment, participants writing a post about social media were shown suggested text from a large language model. The model had been configured to argue one side. Using it affected the opinions expressed in their writing ⁸.

So an AI suggestion is a choice of wording, and it can carry a view. For a routine reply, a checked suggestion costs little. For writing with a point of view, the view should be yours.

It depends on the task, the tool and the accuracy you need

AI helps most when the output is long, the structure is clear, and checking is quick. It helps least when the text is short and personal, or when you must be sure every claim is right. The studies above cover coding, professional writing, phrase prediction and speech, and none measures a typical person's everyday typing across apps over a month.

For people with motor impairments, the picture is different. A 2008 simulation of word prediction for communication aids found a 58.6 percent gain in communication rate ⁹, and the SpeakFaster paper is written for users with severe motor impairments. Our overview of accessibility and dictation covers this in more detail.

What to do

1. For short, routine replies, accept an AI suggestion after you have read it.
2. For writing with a point of view, draft the argument yourself, then use the assistant to tighten the wording. Read every change it makes.
3. For long text that changes each time, speak it and then edit. Compare the time to finish with the time to type it, including corrections.
4. Check any number, name or claim the AI adds before you send it. Speed means little if the text is wrong.
5. Test your own work for a week with and without AI, timing the finished text, not the first draft. The method is in our guide to how typing speed is measured.

For the tools that predict single words, read the question on whether word suggestions save typing time. For repeated phrases, the question on what a text expander is covers the simpler, free option. The overview of text expanders and autocomplete puts these tools side by side, and the wider picture of typing speed starts with the baseline numbers. For the model behind AI phrase prediction, see large language model.