Published October 8, 2026 · 7-minute read
It's a fair question, and more people are asking it. Voice dictation gets better every year. AI autocomplete finishes your sentences. Coding assistants write whole functions. If the machine does the typing, why should you care about your WPM?
Here's the honest answer: typing speed does still matter — but the reasons have changed. The research from the last two years tells a more interesting story than "AI killed typing." It says typing was never quite the bottleneck we thought it was, and the skills that replace it aren't the ones the hype suggests.
Let's start with the uncomfortable part, because the data is real. The landmark study is from Stanford HCI (Ruan et al., published at CHI 2016), where researchers pitted state-of-the-art speech recognition against smartphone keyboards. Speech entered text about 3.0x faster than the keyboard in English (161.20 vs. 53.46 WPM) and about 2.8x faster in Mandarin (108.43 vs. 31.31 WPM) — with 20.4% lower error rates in English. That study is a decade old, and speech models have only improved since.
A more personal data point: a developer who tracked two full weeks of real work in September 2026 found raw voice dictation averaged about 140 WPM (vs. 65 WPM typing), but after correcting the 12% error rate, effective speed settled around 90 WPM — roughly 40% faster than typing, with 20–40% of total time spent fixing mistakes. That's one person's self-tracking, not a lab result, but the pattern rings true: dictation is faster, and correction eats a big share of the gain.
So yes: for producing a first draft of long-form text, speaking beats typing. This isn't controversial anymore.
Now the interesting part. If AI and autocomplete eliminate typing, productivity should be soaring wherever they're used. Independent researchers decided to check.
In 2025, METR — an independent AI evaluator, not a vendor — ran a randomized controlled trial with 16 experienced open-source developers on 246 real issues in their own repositories, using Cursor Pro with Claude 3.5/3.7 Sonnet. Before the study, developers predicted AI would make them 24% faster. Afterward, they believed it had made them 20% faster. The timestamps and screen recordings said otherwise: they were 19% slower with AI allowed. The gap between felt speed and measured speed was 39 percentage points.
A larger 2026 follow-up (57 developers, 800+ tasks, agentic tools rather than plain autocomplete) moved the point estimate toward neutral — about −4%, with confidence intervals crossing zero. METR called it "only very weak evidence" of any speedup at all. The broader literature splits by task type: a Microsoft/GitHub/MIT study found 55.8% faster on a single self-contained greenfield task, while a Google enterprise RCT found roughly 21% faster unadjusted — not statistically significant after adjustment. The consistent pattern: AI helps on bounded, well-scoped tasks and disappears in complex, familiar work.
IBM's framing at the April 2026 general-availability launch of its Bob coding platform says the quiet part out loud: the bottleneck was never the typing. In real software work, the slow parts are understanding legacy systems, coordinating across teams, satisfying compliance, and reviewing AI-generated work before it ships. Meanwhile, GitClear's 2025 analysis of 211 million lines of changed code found repos with heavy AI assistance had 41% more code churn — code written and rewritten within two weeks. Somebody has to review all of that. Guess how that review gets done.
At the keyboard. With your fingers.
First, the blunt practical case: employers still test for it. 2026 job guides list minimums that will disqualify you on the spot — office admin around 40 WPM, data entry 50–75 WPM at 98–100% accuracy, customer-service chat 45–65 WPM, medical transcription 70–90 WPM at 99% accuracy, legal secretary 65–85 WPM. India's SSC CGL data-entry test demands 2,000 key depressions in 15 minutes with a 5% error tolerance. Nobody waives these because AI exists. Note what these tests measure: net WPM, where every error drags your score down. Accuracy has always been the currency, not raw speed.
Second, the strategic case. Everyone now has the same AI. The model doesn't tire, doesn't get bored, and never has a bad day — so what separates the fast from the slow is not the engine, it's the driver. That's the premise of our Super IQ campaign: human bandwidth is the differentiator. In practice that means: steering (clear prompts travel from brain to screen through your hands), reviewing (the 41%-churn world above), and correcting (the 20–40% of dictation time that flows through your fingers anyway). A fast, accurate typist tests three ideas in the time a slow typist tests one — more attempts, more feedback, faster learning.
Third, the environment case: much of life is still a keyboard. Offices, exams, coding interviews, live chat support, lecture halls — you can't dictate in any of them. The quiet open-plan floor is not getting quieter.
The honest conclusion isn't "typing is dead" or "typing is everything." It's a workflow:
Typing speed isn't the skill that builds careers anymore. It's the skill that keeps you from being the bottleneck in front of the tools that are supposed to make you fast.
This article is the starting point. Copy any of these prompts into your favorite AI assistant to learn more about whether typing speed still matters in the AI age:
🔎 Design your personal dictate-then-edit workflow
I write [emails / reports / code comments — your type] about [N] hours a week on a [Windows/Mac] computer. Walk me through setting up a dictate-first-draft, type-the-edits workflow: which free tools to try first, how to handle punctuation by voice, and how to measure whether it's actually faster than how I work now.
🧭 Test yourself on the research
Explain the METR AI-productivity study in plain language: what they tested, why developers felt 20% faster while measuring 19% slower, and what the 2026 follow-up found. Then quiz me with 5 multiple-choice questions to check I understood the difference between perceived and measured speedup.
💡 Build a 4-week typing plan
I currently type [your WPM] at about [your accuracy]% accuracy and want to reach [target WPM] with 95%+ accuracy for [data entry / office work / programming — your reason]. Give me a 4-week plan with 15 minutes of daily practice, specific drill types for each week, and how to tell if I've hit a plateau versus just having a bad week.
Tip: replace the bracketed parts with your own situation — the more specific your prompt, the more useful the answer.
One sharper AI prompt + one new typing topic every week. Free forever, unsubscribe anytime.
We never share your email. One email a week, that's the deal.