Across 113 one-to-one class sessions with 16 adult learners, the topics that came up for the most different students were regional Mexican usage, conversation confidence, pronunciation and listening speed. The grammar that fills most textbooks came further down, and one famous pair, por against para, appeared in the sessions of only two students.
This is a record of what adults actually spent paid class time on, taken from the automatic summaries of the classes themselves rather than from a survey or a curriculum plan.
Key takeaways
- Regional Mexican usage came up for 15 of 16 students, in 52 of 113 sessions. It is the most widely shared sticking point we have.
- Conversation confidence came up for 11 students, in 42 sessions. It is not a grammar problem and it takes real class time.
- Pronunciation appeared in 80 of 113 sessions (70.8%), more than any other topic by volume.
- The subjunctive, which dominates intermediate textbooks, appeared in 28 sessions but for only 5 different students.
- Twenty-seven topics were tested. The full table, with session counts and student counts for each, is published as a CSV.
The short version
Adults do not get stuck where the syllabus says they will. They get stuck on the parts of a language that vary by country, on hearing fast speech, on finding a word in the two seconds they have, and on the fear of speaking at all. Each of those four filled more class sessions than any classic grammar pair we tested.
What the research already says, and what was missing
The idea that fear of speaking is a real, measurable part of language learning is forty years old. Horwitz, Horwitz and Cope introduced the Foreign Language Classroom Anxiety Scale in 1986 and split the problem into three parts: communication apprehension, fear of negative evaluation, and test anxiety (The Modern Language Journal, 70(2), 125 to 132). Decades of studies have used that scale since, mostly with university students filling in questionnaires.
What is thin is the other kind of evidence: an ordinary record of what paid adult lessons actually get spent on, week after week, outside a research setting. Lists of “the hardest parts of Spanish” are usually a teacher’s opinion, which is worth something but is not a count. This study is a count. It also lands near the older research from a different direction: the thing our record calls conversation confidence, showing up for 11 of 16 learners, is the classroom-time footprint of what that 1986 paper called communication apprehension.
What we coded
Every class we teach on Zoom produces an automatic summary. We pulled all 360 of those summaries from January 2025 to September 2026, kept the 113 that are one-to-one classes with an adult learner, and discarded the rest, which are content recordings and work meetings. The 113 summaries hold 123,155 words. Each one was tested against 27 fixed topic patterns, and a topic scores once per session.
The sample is concentrated and we are saying so first. One student accounts for 60 of the 113 sessions. A second accounts for 16. That is why the table below ranks by the number of different students rather than by the number of sessions: a thing one learner struggles with is an anecdote, and a thing eleven learners struggle with is a finding.
Regional usage is the most widely shared problem
Fifteen of 16 students hit Mexican and regional usage, in 52 of the 113 sessions. This is the gap between the Spanish a course teaches and the Spanish a person hears in Guadalajara. It shows up as vocabulary that changes across a border, as slang that a textbook will not print, and as the recurring question of whether a phrase is friendly or rude.
It is also the hardest thing to learn without a person. A dictionary can tell you what guero means. Only a Mexican can tell you who says it, to whom, and with what face.
Confidence takes class time, and it is not soft
Conversation confidence appeared in 42 sessions and for 11 different students. The summaries record it as hesitation, as falling back into English, as apologizing for a mistake before finishing the sentence. Adults arrive able to read Spanish they cannot say, and the block is real enough that it appears in the automatic record of the class without anyone naming it as a topic.
The instructional response we settled on is not encouragement. It is design: a first class that ends with one thing the student can say to a real person that week, and conversation sessions with a fixed topic so nobody spends thirty minutes deciding what to talk about.
The full taxonomy
Twenty-seven topics were tested and every one of them appeared in at least two sessions. The table below shows the most widespread, plus por against para for contrast. Session counts are useful for knowing what fills a teacher’s week. Student counts are what should drive a curriculum.
| Topic | Students (of 16) | Sessions (of 113) | Share of sessions |
|---|---|---|---|
| Mexican and regional usage | 15 | 52 | 46.0% |
| Conversation confidence | 11 | 42 | 37.2% |
| Pronunciation and accents | 10 | 80 | 70.8% |
| Listening speed | 10 | 40 | 35.4% |
| Vocabulary gaps (how do you say) | 9 | 54 | 47.8% |
| Preterite against imperfect | 8 | 28 | 24.8% |
| Stem changing verbs | 7 | 26 | 23.0% |
| Reflexive verbs | 7 | 21 | 18.6% |
| Medical and workplace Spanish | 6 | 67 | 59.3% |
| The subjunctive | 5 | 28 | 24.8% |
| Ser against estar | 5 | 13 | 11.5% |
| Por against para | 2 | 3 | 2.7% |
The complete list of 27 topics is in the CSV.
The textbook order does not match the trouble
A standard course sequences grammar by structure: present tense, then past tenses, then the subjunctive. Our record sequences trouble differently. Pronunciation appears in 71% of sessions, more than any other topic, and it never shows up as a finished unit the way a tense does. Listening speed appears for ten of the 16 students. The subjunctive, treated as the summit of intermediate Spanish, came up for five students, which is real and much narrower than its reputation.
The honest reading is that grammar topics are episodes and the other four are conditions. A student finishes the preterite. Nobody finishes listening.
How another teacher could use this
If you teach adults, the ranking suggests putting regional variation and listening into the weekly structure rather than saving them for an advanced module, and treating confidence as a design problem for the lesson plan rather than a personality trait of the student. If you are a learner choosing a course, it suggests one question worth asking: what happens in week one, and will I say anything out loud to another human being in it?
How we counted this
Source: Zoom AI Companion meeting summaries, delivered by email after each class, pulled from the school’s own mailbox on September 20, 2026.
Window: January 30, 2025 to September 18, 2026.
Sample: 360 summaries pulled. 113 classified as one-to-one classes with an adult learner, holding 123,155 words. 247 excluded as content recordings, group sessions or work meetings. 16 distinct students.
Classification: a summary counts as a class when it names a student from the school roster who is not a member of staff, and at least three distinct teaching terms appear in the text. The student credited is the roster name mentioned most often, so a name that appears once inside a practice dialogue is not counted as a separate learner.
Coding: 27 fixed regular expressions, one hit per session maximum. A session can carry several topics, so the session column does not sum to 113. The patterns are crude on purpose: they can be rerun, and anyone can see what they match.
Limits: this is a small, concentrated sample. One student accounts for 53% of the sessions, which is why the ranking is by distinct students. The summaries are written by an automatic system that records what was discussed, not an assessment of what the student got wrong, so a topic count is a measure of class time and attention rather than of error rates. All students are referred to in aggregate and no names or content appear.
Data: the full taxonomy (CSV). Free to reuse with attribution.
Questions people ask about this data
What is the hardest part of Spanish for English speakers?
In our classroom record, it is not a grammar rule. The topics that recur for the most different learners are regional variation in Mexican Spanish, confidence in speaking at all, pronunciation, and understanding speech at native speed. Grammar pairs like preterite against imperfect matter, and they affect fewer students than the four above.
Is the subjunctive really that hard?
It came up in 28 of 113 sessions but for only five of our 16 students, because it belongs to a level most learners have not reached. When it does arrive it takes several sessions. Its reputation comes from being difficult for the people who get to it, not from being a problem for everyone.
How do you measure what confuses a student?
Imperfectly. We count what the automatic class summary records as being discussed, which is a proxy for attention rather than a direct measure of error. A better instrument would be a coded transcript of student speech with errors marked, which is on our list and is a much larger job.
Why publish a sample this small?
Because it does not otherwise exist. Public data on adult language learning comes from apps and from classroom studies of university students. Nobody publishes what happens in paid one-to-one lessons with working adults, so 113 sessions with the limits stated is more useful than nothing, as long as nobody treats it as a population estimate.
Can I cite this?
Yes, with attribution to Speak Better Spanish and a link to this page. Questions to hannah@speakbetterspanish.com.
References
- Horwitz, E. K., Horwitz, M. B. and Cope, J. Foreign Language Classroom Anxiety. The Modern Language Journal, 70(2), 1986, 125 to 132.
- Cambridge English. Guided learning hours per CEFR level.
- Speak Better Spanish. Learner sticking points, September 20, 2026.
Latest updates
- September 20, 2026, later the same day: Added the background section, the outside references and the data visuals. Figures unchanged.
- September 20, 2026: First publication, coded the same day from the full summary corpus.
Recent Comments