The Tab-to-Type era has arrived
If your typing test scores suddenly jumped, you’re not imagining it. Today’s operating systems and productivity suites surface inline edits and full‑sentence suggestions you can accept with a single key—often Tab or Enter. Gmail’s Smart Compose has long let you Tab to accept inline predictions, and Google Docs supports similar, in‑flow completions. That’s a boon for getting work done—less so for gauging your true composition speed. (support.google.com)
Microsoft is now rolling out text editing via Copilot on Windows, which can rewrite or refine text “in‑flow” during a Vision session and present previews you can accept on the spot. The December 19, 2025 Insider update began seeding this capability across channels, with ongoing feature evolution in 2026 across the Microsoft 365 Copilot stack. (blogs.windows.com)
Meanwhile, Grammarly has expanded its in‑line tooling across apps, advertising support that “works where you do” across desktop and web environments. The breadth—and the recent public debate over its AI features—illustrate just how ubiquitous inline assist has become. (grammarly.com)
Why traditional WPM is no longer enough
Classic typing tests assume your output equals your keystrokes. In the standard formula, one word equals five characters (including spaces), so WPM = (characters ÷ 5) ÷ minutes. But when AI inserts an entire clause you accept with a single key, traditional WPM inflates your skill by attributing machine‑generated text to human typing. (en.wikipedia.org)
It’s not just speed inflation. Research shows predictive suggestions can shape what we write, not only how fast we write it. Studies from Harvard SEAS and recent coverage of a Cornell Tech study indicate autocomplete can nudge phrasing and even attitudes—subtle “prediction bias” that can homogenize style or steer content. That’s another reason a fair test should separate what you typed from what you accepted. (seas.harvard.edu)
Consider scale: Gmail’s Smart Compose alone saves users over one billion characters of typing per week and serves a user base in the billions, with sub‑60ms latency targets for suggestions. With assistive predictions this pervasive, measuring pure composition skill requires new instrumentation. (ar5iv.org)
A fairer test design for 2026 and beyond
Let’s redesign typing assessments to reflect the modern reality.
1) Distinguish human input from accepted suggestions
- Event‑level separation: In a controlled test editor (browser or desktop), log two streams:
- Typed characters: physical keystrokes that produce characters (letters, numbers, punctuation, whitespace).
- Accepted suggestions: text inserted via acceptance actions (e.g., Tab/Enter for inline completions, click‑to‑apply rewrites, or dedicated accept shortcuts). Your instrumentation should classify insertions from suggestion providers distinctly from individual keypresses.
- Practical implementation notes:
- Track beforeinput/input events and tag insert origins (typed vs programmatic insertion). Where you control the editor, generate suggestions internally so you can mark acceptance events unambiguously.
- Disable paste and drop events or count them as “non‑typed” to prevent bypass.
- For third‑party tooling you can’t introspect, offer a locked, no‑extensions test mode (see below).
2) Report two complementary metrics
- Composition WPM (human‑only):
- Definition: WPM based solely on human‑typed characters.
- Formula: Composition WPM = (human‑typed characters ÷ 5) ÷ minutes. This aligns to the industry 5‑char word standard while excluding machine‑inserted text. (en.wikipedia.org)
- Suggestion Acceptance Rate (SAR):
- Definition: The proportion of final output characters that originated from accepted suggestions.
- Formula: SAR = accepted‑suggestion characters ÷ total output characters.
- Also report Accepts per Minute (APM) to reflect how often users lean on suggestions.
- Optional: Assisted WPM (total output):
- Definition: Traditional net WPM on the final text (typed + accepted).
- Use this to capture practical throughput alongside Composition WPM and SAR.
Together, these paint a fair picture: how fast you compose unaided, how much you rely on AI, and your overall throughput with assistance.
3) Provide a no‑suggestions test mode
- Built‑in toggle: Offer a native “No Suggestions” session where the app blocks all inline completions and rewrites.
- OS/app guidance: Remind users how to disable common assistants:
- Gmail/Docs Smart Compose: Turn off writing suggestions; accepting suggestions normally uses Tab/Enter, so disabled mode prevents accidental acceptance. (support.google.com)
- Copilot on Windows: Copilot Vision text editing is explicitly opt‑in in Insider builds; instruct candidates to keep it off during proctored runs. (blogs.windows.com)
- Grammarly: Provide an in‑app switch to suspend Grammarly for the test domain/window, since it’s designed to work broadly across apps. (grammarly.com)
4) Test content and fairness guardrails
- Use neutral, non‑trigger prompts to reduce topical bias.
- Rotate passages and randomize variants per user to prevent memorization.
- Where possible, surface identical prompts in both modes (with and without suggestions) to compare Composition WPM apples‑to‑apples.
5) Transparent scoring UI
Show three numbers on the results card:
- Composition WPM (human‑only)
- Suggestion Acceptance Rate (and Accepts per Minute)
- Assisted WPM (total output)
Add an explanation: “Composition WPM reflects your own keystrokes. Suggestion Acceptance Rate reflects how much AI you accepted.” This makes the trade‑off visible to learners, teachers, and hiring managers.
Actionable tips for test‑takers
- Practice in both modes. Use no‑suggestions mode to build raw speed and accuracy; then re‑enable assistance to learn when suggestions genuinely help rather than distract.
- Learn the acceptance keys—and when not to press them. In Gmail and Docs, Tab/Enter accept suggestions; train yourself to ignore irrelevant grayed‑out text. (support.google.com)
- Audit your own SAR. If your Suggestion Acceptance Rate is high but Composition WPM is low, invest in foundational drills (home‑row fluency, punctuation bursts, numbers lines).
Guidance for teams and hiring managers
- Request both metrics. Ask candidates for Composition WPM and SAR rather than a single headline WPM. A 75 WPM Composition with 10% SAR shows very different skills than 75 WPM with 60% SAR.
- Define role‑appropriate thresholds. Customer support might value higher Assisted WPM (consistent phrasing), while journalism or programming screens should weight Composition WPM more heavily.
Why this matters now
Two trends make this urgent. First, inline AI is woven into our daily tools—from Gmail/Docs to Windows itself—with growing, in‑flow editing features. Second, user sentiment around overreach is rising; high‑profile controversies over AI‑driven writing aids (and their rapid rollbacks) show how quickly the landscape is evolving. Tests that ignore these realities risk misclassifying skills and incentivizing “press Tab” over learning to compose. (blogs.windows.com)
A modern typing test shouldn’t fight assistive tech; it should make assistance legible. By separating human keystrokes from accepted suggestions, reporting both Composition WPM and Suggestion Acceptance Rate, and offering a no‑suggestions mode, you’ll give learners a fair score, coaches better diagnostics, and employers clearer signals—without pretending the Tab key doesn’t exist.
