What 410,000 Real English Speaking Conversations Reveal About How People Practise
By SpeakPal Club Research · Published 4 November 2026 · Data: 6 August to 3 November 2026 · PDF version available
According to SpeakPal Club data from about 410,000 one-to-one English conversations that lasted at least 30 seconds, held by 34,348 learners between 6 August and 3 November 2026, the median conversation lasted 4 min 58 s (298 seconds) and 35.1% lasted 10 minutes or longer.
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: a real conversation is a completed call (conversation_done) of 30 seconds or longer, 6 August–3 November 2026 (Europe/Madrid). Both people on a call are logged. Worldwide conversation totals are sessions divided by 2. Per-learner figures count the calls that learner joined. Country is the analytics platform's inferred country.
Key findings
- According to SpeakPal Club data, 34,348 learners held about 410,000 real one-to-one English conversations (30 seconds or longer) in 90 days (6 August to 3 November 2026).
- The median SpeakPal Club English conversation lasted 4 min 58 s (298 s) and the average 10 min 30 s (630 s), based on about 410,000 conversations of 30 s or longer (Aug-Nov 2026).
- 35.1% of real SpeakPal Club English conversations (30 s or longer) lasted 10 minutes or more (about 410,000 conversations, Aug-Nov 2026).
- Each SpeakPal Club learner with at least one real conversation took part in 23.8 real conversations on average over 90 days (34,348 learners, Aug-Nov 2026).
- Of SpeakPal Club learners who had a real conversation, 40.7% had another one the next day and 20.5% on day 30 (26,347 learners, cohorts 6 Aug-4 Oct 2026).
- The median real conversation by SpeakPal Club learners in India lasted 6 min 44 s (404 s), against 4 min 20 s (260 s) in Uzbekistan (Aug-Nov 2026).
Source and method for every figure in the list above: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Unless a bullet names a different window, a real conversation is a completed call of 30 seconds or longer, 6 August–3 November 2026 (Europe/Madrid), and a per-learner count is the calls that learner joined. Both people on a call are logged. Country is inferred by the analytics platform. Bullets that count every logged session include connection attempts under 30 seconds. Peak-hour bullets use 27 October–3 November 2026 in local time. Return-rate bullets are the share with another real conversation on that exact day, for cohorts from 6 August to 4 October 2026. Goal bullets are self-selected and can include people who never had a real conversation.
What is SpeakPal Club?
SpeakPal Club is an iOS and Android app where people practising spoken English are matched by level and interests for live one-to-one voice calls with other learners and native English speakers. The app is free to use; free calls have a per-call time limit. Every call in this dataset is a one-to-one conversation between two app users.
What counts as a real conversation?
According to SpeakPal Club data from 1,967,528 logged call sessions (Aug-Nov 2026), we count a call as a real conversation only if it lasted 30 seconds or longer; 58.4% of sessions ended before that and are reported separately as connection attempts.
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: every logged call session from 6 August–3 November 2026, including those under 30 seconds. Sessions under 30 seconds are connection attempts, not conversations. Analytics cannot see inside a call, so a short session may be a drop, a silent partner, or a quick skip.
Swipe sideways to see the full table.
| Call length | Sessions | Share | Uzbekistan | India |
|---|---|---|---|---|
| 0-10 s | 808,778 | 41.1% | 42.7% | 37.9% |
| 10-30 s | 340,868 | 17.3% | 17.6% | 16.4% |
| 30-60 s | 161,430 | 8.2% | 8.6% | 7.9% |
| 1-3 min | 170,262 | 8.7% | 8.3% | 8.5% |
| 3-10 min | 198,788 | 10.1% | 10.2% | 10.2% |
| 10 min or longer | 287,402 | 14.6% | 12.6% | 19.1% |
| Total sessions | 1,967,528 | 100% | 933,591 | 277,030 |
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: every logged call session from 6 August–3 November 2026, including those under 30 seconds. Sessions under 30 seconds are connection attempts, not conversations. Analytics cannot see inside a call, so a short session may be a drop, a silent partner, or a quick skip.
Sessions, not conversations: both participants log every call. Country is the learner’s current analytics country.
Lengths are bimodal: a large spike under 10 seconds, a low trough between 30 seconds and 3 minutes, and a second peak of long calls. The 30-second line is there to exclude dropped or failed connection attempts, most of which end in under 10 seconds (41.1% of all sessions); 30 seconds is long enough for two people to have actually started talking. Analytics cannot see what happened inside a call, so some 30-60 second calls will still be quick skips; with the stricter 60-second line used in an earlier draft there would be about 328,000 conversations (656,452 sessions).
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: every logged call session from 6 August–3 November 2026, including those under 30 seconds. Sessions under 30 seconds are connection attempts, not conversations. Analytics cannot see inside a call, so a short session may be a drop, a silent partner, or a quick skip.
How long is a typical English speaking practice conversation?
According to SpeakPal Club data from about 410,000 conversations of 30 seconds or longer (Aug-Nov 2026), the median conversation lasted 4 min 58 s (298 s); the average was 10 min 30 s (630 s), pulled up by long calls, and 35.1% lasted 10 minutes or more.
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: a real conversation is a completed call (conversation_done) of 30 seconds or longer, 6 August–3 November 2026 (Europe/Madrid). Both people on a call are logged. Worldwide conversation totals are sessions divided by 2. Per-learner figures count the calls that learner joined. Country is the analytics platform's inferred country.
Including the sub-30-second connection attempts, the average across all sessions is 4 min 26 s (266 s). That figure describes how long a call screen stays open, not how long people talk, so we do not use it as the headline. Free calls have a per-call time limit, which caps how long some conversations can run.
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: a real conversation is a completed call (conversation_done) of 30 seconds or longer, 6 August–3 November 2026 (Europe/Madrid). Both people on a call are logged. Worldwide conversation totals are sessions divided by 2. Per-learner figures count the calls that learner joined. Country is the analytics platform's inferred country.
How often do people practise speaking English with a partner, by country?
According to SpeakPal Club data from 34,348 learners with at least one real conversation (Aug-Nov 2026), each took part in 23.8 real conversations on average over 90 days: 23.6 in Uzbekistan and 28.1 in India.
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: a real conversation is a completed call (conversation_done) of 30 seconds or longer, 6 August–3 November 2026 (Europe/Madrid). Both people on a call are logged. Worldwide conversation totals are sessions divided by 2. Per-learner figures count the calls that learner joined. Country is the analytics platform's inferred country.
Swipe sideways to see the full table.
| Country | Learners with a real conversation | Share of learners | Real-conversation sessions | Per learner | Median length | 10 min or longer |
|---|---|---|---|---|---|---|
| All countries | 34,348 | 100% | 817,882 (≈408,941 conversations) | 23.8 | 4:58 (298 s) | 35.1% |
| Uzbekistan | 15,698 | 45.7% | 370,786 | 23.6 | 4:20 (260 s) | 31.8% |
| India | 4,500 | 13.1% | 126,632 | 28.1 | 6:44 (404 s) | 41.9% |
| Afghanistan | 1,844 | 5.4% | 34,090 | 18.5 | 2:34 (154 s) | 20.8% |
| Indonesia | 1,547 | 4.5% | 28,631 | 18.5 | 7:42 (462 s) | 44.1% |
| Turkey | 1,428 | 4.2% | 27,414 | 19.2 | 7:23 (443 s) | 42.8% |
| Germany | 1,059 | 3.1% | 11,676 | 11.0 | 5:07 (307 s) | 35.2% |
| Bangladesh | 734 | 2.1% | 17,683 | 24.1 | 6:40 (400 s) | 41.4% |
| Saudi Arabia | 601 | 1.7% | 11,422 | 19.0 | 5:44 (344 s) | 38.4% |
| Pakistan | 550 | 1.6% | 13,541 | 24.6 | 3:36 (216 s) | 31.3% |
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: a real conversation is a completed call (conversation_done) of 30 seconds or longer, 6 August–3 November 2026 (Europe/Madrid). Both people on a call are logged. Worldwide conversation totals are sessions divided by 2. Per-learner figures count the calls that learner joined. Country is the analytics platform's inferred country.
Real conversation = conversation_done with duration ≥ 30 s. Country rows count sessions, i.e. conversations that country’s learners took part in; they are not halved, because a call between two learners in the same country is logged twice there and a cross-border call once, so only the worldwide total can be converted to distinct conversations (÷2). Iraq is excluded: its 440 learners logged 23,081 sessions with a median of 70 s, an unusual cluster between one and three minutes that we have not explained. France (886), the United States (653), the Netherlands (547), Azerbaijan (508) and the United Kingdom (462) also have more than 200 learners but are not broken out here.
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: a real conversation is a completed call (conversation_done) of 30 seconds or longer, 6 August–3 November 2026 (Europe/Madrid). Both people on a call are logged. Worldwide conversation totals are sessions divided by 2. Per-learner figures count the calls that learner joined. Country is the analytics platform's inferred country.
Do advanced English learners practise more than beginners?
According to SpeakPal Club data from about 410,000 real conversations (Aug-Nov 2026), C1 learners took part in 33.0 real conversations each with a median length of 6 min 25 s, versus 18.0 conversations of 3 min 38 s for A1 learners.
Source: Amplitude, read-only query on 4 November 2026. Method: the CEFR level the learner selected in the app at the time of the call, 6 August–3 November 2026. A real conversation is 30 seconds or longer. Learners who changed level are counted in each level they used. Per-learner counts are the calls that learner joined.
Swipe sideways to see the full table.
| Self-reported level | Learners | Real-conversation sessions | Per learner | Median length | 10 min or longer |
|---|---|---|---|---|---|
| A1 | 7,031 | 126,480 | 18.0 | 218 s | 29.5% |
| A2 | 5,556 | 92,816 | 16.7 | 242 s | 30.4% |
| B1 | 8,874 | 162,512 | 18.3 | 278 s | 33.2% |
| B2 | 9,392 | 220,491 | 23.5 | 336 s | 37.1% |
| C1 | 5,089 | 167,720 | 33.0 | 385 s | 40.6% |
| C2 | 1,508 | 47,852 | 31.7 | 323 s | 37.1% |
Source: Amplitude, read-only query on 4 November 2026. Method: the CEFR level the learner selected in the app at the time of the call, 6 August–3 November 2026. A real conversation is 30 seconds or longer. Learners who changed level are counted in each level they used. Per-learner counts are the calls that learner joined.
Level is the CEFR level learners select in the app at the time of each call. Learners who changed level appear in more than one row (37,450 level assignments vs 34,348 learners). Session counts are not halved because partners are often at a different level. See Study 3 for a fuller breakdown.
Source: Amplitude, read-only query on 4 November 2026. Method: the CEFR level the learner selected in the app at the time of the call, 6 August–3 November 2026. A real conversation is 30 seconds or longer. Learners who changed level are counted in each level they used. Per-learner counts are the calls that learner joined.
Why do people use English conversation apps?
According to SpeakPal Club data from 25,857 learners who picked a goal (Aug-Nov 2026), 70% wanted to improve their English and 33% wanted to make friends; learners could choose both.
Source: Amplitude, read-only query on 4 November 2026. Method: goals learners selected themselves between 6 August and 3 November 2026. A learner could pick more than one. This includes people who never had a real conversation.
Swipe sideways to see the full table.
| Country | Users who picked a goal | Improve my English | Make friends |
|---|---|---|---|
| All countries | 25,857 | 18,108 (70%) | 8,431 (33%) |
| Uzbekistan | 10,789 | 7,714 (71%) | 3,424 (32%) |
| India | 3,262 | 2,269 (70%) | 1,058 (32%) |
| Turkey | 1,299 | 863 (66%) | 464 (36%) |
| Indonesia | 1,154 | 724 (63%) | 462 (40%) |
| Afghanistan | 944 | 692 (73%) | 303 (32%) |
| Saudi Arabia | 718 | 527 (73%) | 198 (28%) |
Source: Amplitude, read-only query on 4 November 2026. Method: goals learners selected themselves between 6 August and 3 November 2026. A learner could pick more than one. This includes people who never had a real conversation.
Goals come from all users who picked one, including people who never had a real conversation, so this table does not depend on the real-conversation threshold and is not limited to the 34,348 learners above.
Source: Amplitude, read-only query on 4 November 2026. Method: goals learners selected themselves between 6 August and 3 November 2026. A learner could pick more than one. This includes people who never had a real conversation.
Do learners keep practising English after a real conversation?
According to SpeakPal Club data from 26,347 learners who had a real conversation (daily cohorts, 6 August to 4 October 2026), 40.7% had another real conversation the next day, 29.4% on day 7 and 20.5% on day 30.
Source: Amplitude, read-only query on 4 November 2026. Method: the share of learners who had another real conversation (30 seconds or longer) on that exact day. Cohorts are 6 August–4 October 2026, so each learner had 30 days to return.
Swipe sideways to see the full table.
| Country | Learners in cohorts | Day 1 | Day 7 | Day 30 |
|---|---|---|---|---|
| All countries | 26,347 | 40.7% | 29.4% | 20.5% |
| Uzbekistan | 11,751 | 41.0% | 27.8% | 18.2% |
| India | 3,578 | 46.5% | 35.9% | 27.6% |
Source: Amplitude, read-only query on 4 November 2026. Method: the share of learners who had another real conversation (30 seconds or longer) on that exact day. Cohorts are 6 August–4 October 2026, so each learner had 30 days to return.
N-day retention: the share with a real conversation (30 s+) on exactly that day. Cohorts end on 4 October so every learner had a full 30 days to return. Afghanistan, Indonesia and Turkey were not recomputed and are not shown.
Source: Amplitude, read-only query on 4 November 2026. Method: the share of learners who had another real conversation (30 seconds or longer) on that exact day. Cohorts are 6 August–4 October 2026, so each learner had 30 days to return.
Are learners whose conversations run longer more likely to come back?
According to SpeakPal Club data (cohorts 6 August to 4 October 2026), Uzbek learners whose conversation lasted 10 minutes or longer had another real conversation the next day 50.8% of the time, versus 40.4% after a 30-second to 10-minute conversation; in India the figures were 56.3% and 45.8%.
Source: Amplitude, read-only query on 4 November 2026. Method: the share of learners who had another real conversation (30 seconds or longer) on that exact day. Cohorts are 6 August–4 October 2026, so each learner had 30 days to return.
Swipe sideways to see the full table.
| Country | Starting conversation | Learners | Next day | Day 7 | Day 30 |
|---|---|---|---|---|---|
| Uzbekistan | 30 s-10 min | 11,126 | 40.4% | 27.1% | 17.5% |
| Uzbekistan | 10 min or longer | 7,485 | 50.8% | 34.9% | 22.5% |
| India | 30 s-10 min | 3,345 | 45.8% | 35.0% | 26.8% |
| India | 10 min or longer | 2,478 | 56.3% | 44.6% | 34.4% |
Source: Amplitude, read-only query on 4 November 2026. Method: the share of learners who had another real conversation (30 seconds or longer) on that exact day. Cohorts are 6 August–4 October 2026, so each learner had 30 days to return.
This is a correlation, not evidence that longer calls cause people to return: learners who are already motivated are likely both to talk longer and to come back, and a learner can appear in both rows.
When do people practise speaking English?
According to SpeakPal Club data from 8 days of real conversations (27 October to 3 November 2026, local time), 50.8% of conversations in Uzbekistan and 42.3% in India took place between 18:00 and midnight. Full hour-by-hour results are in Study 2.
Source: Amplitude, read-only query on 4 November 2026. Method: real conversations (completed calls of 30 seconds or longer) counted by local hour, 27 October–3 November 2026, queried in that country's time zone.
FAQ
How long is a typical English conversation practice session?
According to SpeakPal Club data from about 410,000 conversations of 30 seconds or longer (Aug-Nov 2026), the median conversation lasted 4 min 58 s and 35.1% lasted 10 minutes or more.
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: a real conversation is a completed call (conversation_done) of 30 seconds or longer, 6 August–3 November 2026 (Europe/Madrid). Both people on a call are logged. Worldwide conversation totals are sessions divided by 2. Per-learner figures count the calls that learner joined. Country is the analytics platform's inferred country.
How often do people practise speaking English with a partner?
SpeakPal Club learners with at least one real conversation took part in 23.8 real conversations each over 90 days (34,348 learners, Aug-Nov 2026); 23.6 in Uzbekistan and 28.1 in India.
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: a real conversation is a completed call (conversation_done) of 30 seconds or longer, 6 August–3 November 2026 (Europe/Madrid). Both people on a call are logged. Worldwide conversation totals are sessions divided by 2. Per-learner figures count the calls that learner joined. Country is the analytics platform's inferred country.
Why do people use English speaking apps?
Of 25,857 SpeakPal Club learners who picked a goal (Aug-Nov 2026), 70% wanted to improve their English and 33% wanted to make friends.
Source: Amplitude, read-only query on 4 November 2026. Method: goals learners selected themselves between 6 August and 3 November 2026. A learner could pick more than one. This includes people who never had a real conversation.
Do people keep practising English speaking after a real conversation?
In SpeakPal Club data from 26,347 learners (cohorts 6 Aug-4 Oct 2026), 40.7% had another real conversation the next day and 20.5% on day 30.
Source: Amplitude, read-only query on 4 November 2026. Method: the share of learners who had another real conversation (30 seconds or longer) on that exact day. Cohorts are 6 August–4 October 2026, so each learner had 30 days to return.
Which English levels practise speaking the most?
On SpeakPal Club, B1 and B2 learners are the largest groups (8,874 and 9,392 learners with a real conversation), but C1 learners took part in the most: 33.0 real conversations each over 90 days, versus 18.0 at A1 (Aug-Nov 2026).
Source: Amplitude, read-only query on 4 November 2026. Method: the CEFR level the learner selected in the app at the time of the call, 6 August–3 November 2026. A real conversation is 30 seconds or longer. Learners who changed level are counted in each level they used. Per-learner counts are the calls that learner joined.
Why do so many English practice calls end in under 30 seconds?
In SpeakPal Club data from 1.97 million call sessions (Aug-Nov 2026), 58.4% ended within 30 seconds and 41.1% within 10 seconds; these are mostly dropped or failed connection attempts, and analytics cannot tell a failed connection, a silent partner and a quick skip apart, so we do not count them as conversations.
Source: SpeakPal Club product analytics (Amplitude), read-only query on 4 November 2026. Method: every logged call session from 6 August–3 November 2026, including those under 30 seconds. Sessions under 30 seconds are connection attempts, not conversations. Analytics cannot see inside a call, so a short session may be a drop, a silent partner, or a quick skip.
Is SpeakPal Club free?
SpeakPal Club is free to use, but free calls have a per-call time limit.
Methodology and limitations
- Source: SpeakPal Club product analytics (Amplitude), queried read-only on 4 November 2026; aggregates only, no personal data exported, no new tracking.
- Window: 6 August to 3 November 2026 (90 days, Europe/Madrid). Return rates use daily cohorts from 6 August to 4 October 2026; peak hours use 27 October to 3 November 2026 in local time.
- Events: call completed =
conversation_donewithdurationin seconds. Real conversation = duration ≥ 30 s, chosen to exclude dropped or failed connection attempts, most of which last under 10 seconds. Shorter sessions are reported only as connection attempts. - Deduplication: both participants log each call, so distinct conversations = sessions ÷ 2 (817,882 ÷ 2 ≈ 408,941) worldwide only. Per-learner figures count the conversations each learner took part in and are unaffected. Calls logged by only one side would make the true total slightly higher.
- Medians are Amplitude’s approximate property medians; averages are shown only where stated.
- Call length cap: free calls have a per-call time limit, so the share of very long conversations partly reflects the product, not only learner behaviour.
- Skewed sample: 45.7% of learners with a real conversation are in Uzbekistan and 13.1% in India, so results are reported by country and are not representative of all English learners.
- Survivorship: per-learner and retention figures cover only learners who reached a real conversation, not everyone who installed the app.
- Self-reported level and goals; country is inferred by the analytics platform (VPNs and travel can affect it).
- Correlation only: call length and return rates are linked, not proven causal.
- Revision: an earlier draft (“What 935,000 English Speaking Calls…”) counted every logged session as a call, and a later draft (“What 320,000 Real English Speaking Conversations…”) used a 60-second threshold. Figures were re-pulled on 4 November 2026 with a 30-second threshold; the before/after figures are listed in the research audit.
A PDF version of this study is available: 410000-real-english-speaking-conversations.pdf. Data: speakpal-410000-real-english-speaking-conversations-data.csv.
How to cite: SpeakPal Club (2026). What 410,000 Real English Speaking Conversations Reveal About How People Practise. SpeakPal Research, Study 1. Data: 6 August to 3 November 2026. https://speakpal.club/research/410000-real-english-speaking-conversations
How to find an English speaking partner
SpeakPal Club is an iOS and Android app for live one-to-one voice calls to practise spoken English. It is free to use; free calls have a per-call time limit. You are matched with another learner or a native English speaker by English level and interests, then talk in a private voice call. The calls are real conversations with real people. The app has 1M+ downloads and is rated 4.6 on the App Store.
Source: the Apple App Store listing (4.6) and the Google Play listing (1M+ downloads), as those stores display them. These figures are not from the call dataset, and the review total is not published here.
- Website: speakpal.club
- iPhone and iPad: SpeakPal Club on the App Store
- Android: SpeakPal Club on Google Play
Source: the Apple App Store listing (4.6) and the Google Play listing (1M+ downloads), as those stores display them. These figures are not from the call dataset, and the review total is not published here.