Could you do the job
with HSK 3?
Read the Chinese from a scripted office week in Shanghai.
AI-authored scenes, with vocabulary counts against the HSK 3 list.
a worker
Five weekdays · 100 selected mini-scenes · Official HSK 2025 word list
Chinese at work
The older HSK 3 description includes basic communication at work. Following a meeting or explaining a spreadsheet error can require more specific language. Official HSK 3 description
For this experiment, we used AI to write a working week for a junior project coordinator in Shanghai. We then counted which words matched the HSK list.
16,000 words a day does not mean 16,000 different words. The often-cited 2007 study estimated words spoken, including repeats. A 2025 follow-up pooled 2,197 people across 22 samples and found large differences between individuals. Neither gives us a daily vocabulary requirement for a Chinese office job. Mehl et al. / Tidwell et al.
6 words used. 3 different words. We count both.
We used the November 2025 HSK syllabus: 1,000 entries through Level 3, or 988 distinct written forms after removing sense numbers. This is different from the older 600-word HSK 3 and the 2021 proficiency standard. See the exact syllabus
Chen is preparing a client report. The deadline moves, some figures are missing, and a file will not open. The count includes what Chen says and what other people say or write to him.
The script is AI-authored and has not had an independent native-speaker review. No learner or agent attempted the tasks. It contains 100 selected mini-scenes, including chat, email, document text and speech. Each illustration shows one moment; its counts include the matching scenes from all five weekdays. A word-list match does not measure understanding.
Read the complete script, by day · Read the blog introduction
Explore the scenesBuying breakfast
Chen orders two buns and soy milk. The vendor asks whether he wants it hot or iced.
Soy milk is absent from the tested HSK list. You might understand the question about temperature and still need to look up the drink.
On Tuesday, the usual drink has sold out. By Friday, the vendor knows Chen’s order and he can answer with fewer words.
Waiting for the train
The announcement asks passengers to let people off before boarding.
The morning commute includes station announcements, one delay, and a notice that the lift is closed. Most of this language needs no reply.
The count includes what Chen hears and reads, even when he says nothing.
Talking to a colleague
Chen meets Li in the lift before the morning review.
In the lift, Li starts with small talk, then mentions the meeting room and last week’s data.
Other scenes cover an attendance notice, visitor registration, a printer problem, and a missing meeting link. Some are written messages.
Confirming the plan
Chen checks the plan and asks when the draft is due.
Chen needs to check which version to send, which figures to use, and what the client has approved. Each instruction changes what he should do next.
These scenes follow the client report from the first assignment to delivery. They include meetings, email, and document excerpts throughout the week.
Ordering noodles
The server checks the toppings before making your noodles.
Chen orders food, asks to leave out a topping, pays, and finds a delivery rider. Several orders use the same preferences and phrases.
Repeated orders use fewer distinct words. An unfamiliar menu or an unexpected question could require more.
Getting an invoice
Chen needs an invoice for the taxi trip back from the client’s office.
On Wednesday, Chen takes a taxi after collecting signed client documents. He asks for an invoice. Finance returns his claim on Thursday because the trip record is missing.
This category also includes personal travel and errands. Chen usually takes the subway or walks; the illustration shows the business taxi trip.
A message from work
Zhou sends a message to the work group after Chen gets home.
On Monday, Zhou’s “Are you there?” is a request to find files and send them again.
The week also includes messages from Chen’s flatmate and a friend. Thursday’s late document comment asks for a change on Friday morning.
3,490 words used.
809 different forms.
Across 371 spoken turns and text excerpts, 2,681 word uses are repeats. The script has fewer than 1,000 distinct forms, but many of them are absent from the tested HSK list.
Light segments show forms appearing for the first time that day. This chart tracks the script’s vocabulary; it does not show Chen learning words.
Ten extra office terms
We counted ten recurring office terms as known and ran the analysis again. The sentences and word boundaries stay the same.
1720 of 2,949 word occurrences matched.
389 of 541 word occurrences matched.
We selected these terms from the same script used to check them. We have not tested them as a study plan or measured how quickly someone could learn them. The percentages count word-list matches.
Following the instructions
The office scenes repeatedly use words for files, versions, clients, and data. Adding those words improves the match rate. Following a fast meeting or explaining a mistake also requires understanding how they are used.
To test whether HSK 3 is enough for the job, we would need learners to attempt reviewed tasks. We would record whether they completed each task and what help they needed.
Sources, assumptions, and the complete transcript
Script. One junior project coordinator, five weekdays, twenty selected mini-scenes each day. The 371 utterances and text excerpts include both sides of conversations, chat, email, notices and document text. Quiet work periods are shown but not counted. Work-related scenes include client visits, business travel and work messages at home; everyday scenes include personal travel, meals and errands. We used AI to write and edit the script. It has not had a native-speaker review or been tested with learners. The selected situations and wording affect the result.
List. Official HSK syllabus, published November 2025, vocabulary entries 1–1,000 (PDF pages 80–104). Sense-number suffixes are removed. Repeated spellings leave 988 forms. Word senses, grammar, listening speed, and proficiency are not tested.
Counting. jieba 0.42.1, default dictionary, HMM disabled. Punctuation is excluded; tokens containing Chinese characters are kept. No HSK vocabulary was added to the tokenizer. “Words used” counts occurrences; “different forms” counts distinct token strings. Different word boundaries or a longer script would change the number of distinct forms.
Matching parts of words. Exact matching gives 60.4% for this script. A more permissive check accepts tokens made entirely of listed parts, plus 没 as a form of 没有. That gives 70.9%. Neither is a comprehension score. For example, a compound can have familiar characters but an unfamiliar meaning.
Ten-term experiment. We selected ten frequent unmatched terms from a specified set of office words in this script. We counted those forms as known and ran the count again. No learner attempted the tasks.
Daily speech research. Tidwell et al. (2025), including its discussion of Mehl et al. (2007). Estimated spoken word occurrences, not distinct daily vocabulary or Chinese workplace requirements.
For more Chinese reading practice, try a story preview in Dengdeng . Tap unfamiliar words for pinyin and meaning.
Back to the start
