Pakistan AI Skills Survey 2026: What Learners Actually Told Us

What we found

1. Only one respondent in 60 (1.7%) named English-only material as the biggest thing stopping them learning AI — even though 70% would rather learn in Urdu, Roman Urdu or a mix of Urdu and English. Preferring to learn in Urdu and being blocked by English are not the same thing, and this survey was run by a company that sells an Urdu-English course.
2. The barriers people did name were guidance and money: 27% said they do not know where to start or whom to trust, and 25% said courses cost too much. 30% said nothing has stopped them at all.
3. 73% would pay PKR 15,000 or less for a complete AI agents course, and 10% would pay nothing at all. Only 18% chose the PKR 15,000–30,000 band that DSP's own course sits in, and 8.3% would pay more than PKR 30,000.
4. 58% want this for freelancing income — more than a job (5.0%), their own business (25%), curiosity (3.3%) and teaching others (8.3%) put together.
5. This is a sample of people already close to AI, not a sample of Pakistan: 83% were already learning it, 68% had built something, and every respondent had used ChatGPT, Claude or Gemini. 37% named Claude against 10% for ChatGPT — a split that reflects where the survey was distributed, not the country.
6. 62% of respondents live in Pakistan, 18% in the Gulf and 12% in the UK, US, Canada or Australia.

Cite as: Pakistan AI Skills Survey 2026, Digital Services Program. Free to quote and reuse with attribution (CC BY 4.0). Download the full counts as CSV.

How the survey was run

• Self-selected online questionnaire of ten closed questions, offered in Urdu and English side by side, open from 2026-09-12 to 2026-09-17. 60 responses were received and all are counted; none were removed or weighted. The first response, on 12 September 2026, was a test submission made by DSP while verifying that the collection pipeline worked; it is counted with the rest.
• Distributed through Digital Services Program's own audience - its Facebook page, WhatsApp groups, LinkedIn and YouTube community - and open to anyone with the link. People already interested in learning AI are therefore over-represented, and the sample is not representative of Pakistan as a whole.
• The survey collected no name and no phone number. Email was optional and used only to send respondents the results. Responses are reported only in aggregate; the CSV on this page holds the counts, not individual answers.
• Respondents could skip a question, so each table's percentages are of the people who answered that question. Every figure on this page, in the CSV and in the structured data is computed from the same counts.

Fielded 2026-09-12 to 2026-09-17. 60 responses. The questionnaire is still readable in full on the survey page, in both languages, so you can see exactly how each question was worded.

Every question, every answer

Counts are given alongside percentages so you can check the arithmetic. Respondents could skip questions, so each table's percentages are of those who answered that question, not of all 60.

What this does and does not show

This is a self-selected online sample, not a random sample of Pakistan. People already interested in AI are over-represented, and anyone without internet access is absent from it entirely. It cannot tell you what the average Pakistani thinks about AI.

What it can tell you is what people who are trying to learn AI say is stopping them — and on that question, as far as we can find, no other published data exists for Urdu speakers at all. We would rather publish it with its limits stated than not publish it.

If you want to run it again, or run it properly with a representative sample, the questions are public and you are welcome to reuse them. Tell us and we will link to your results from this page, including if they disagree with ours.

Who ran it

Sardar Ghaffar, founder of Digital Services Program, an AI agents training company in Islamabad that teaches DSP AI Agent Mastery — Zero to Master in Urdu and English. We sell a course, and a finding that language is a barrier suits us — which is exactly why the wording, the method and every raw count are on this page. Check them.