Pakistani teachers are foraying into the world of artificial intelligence. Here’s a look at the trends
| I |
n February 2024, just over a year after the launch of ChatGPT, we launched the first-ever teacher training programme for teaching with AI. The #AIDotEDU Crash Course invited applicants from across the country for a three-day online bootcamp on critical AI literacy and to help teachers make sense of the tools and its capabilities, capably navigate pedagogy and understand the big picture on AI and related skills—particularly to understand and appreciate this important new frontier in teaching and learning in the 21st Century. Six months earlier, we had already done a crash course for over 300 university professors.
In September 2025 and in August 2026, we repeated the #AIDotEDU Crash Course—with some modifications and improvements—attracting 2,088 applicants from 197 schools.
These massive online open courses—coupled with data collection that preceded them—gave us an unprecedented access into how teachers were thinking about AI; how they were preparing themselves; and what they were actually doing in or outside the classroom. What we have gained is a rare panel dataset that captures how teachers in Pakistan have fared over the last three years in the most consequential revolution in education and how they have tried to equip themselves with the knowhow of this technology and its (ab)uses.
The big picture
The good part first: the number of teachers who had used Gen AI increased from 64 percent in 2024 to 93 percent in 2026; those reporting receiving some kind of training increased from 32 percent to 69 percent; those who see students’ abuse of AI increased from 26 percent to 35 percent; and the gap between public (or low-fee private) schools and elite schools decreased on both training and use of AI.
Now, the not-so-good part: belief in the importance and intention of using AI has remained steady at 80 percent since 2024; only 20 percent report their schools having a policy on AI (another 33 percent are unaware); assessment redesign is flat at 54 percent over the years; and teachers’ use of AI is limited generally to lower-order (productivity-enhancing) tasks such as lesson planning (83 percent), rather than higher-order thinking and critiquing tasks (25 percent).
This means that the system has converted intent into action but not action into learning. The bottlenecks have moved from teachers to institutions; institutional training is reinforcing the productivity rung; fear is the engine of training and trying. The next likely gains are in governance (i.e. institutional policies) and pedagogy (how to implement)—both cheap.
Let’s analyse the numbers.
Trends in teachers’ AI readiness
Five key trends are obvious:
Trend 1: Data clearly suggests that belief was never a bottleneck; capacity and permission was.
In 2024, 80 percent of teachers believed that AI would reshape learning; that assessment has remained steady. What has changed over the period is the teachers’ capacity and training (32 to 69 percent) and use (64 to 93 percent, even larger if we cap the bigger elite schools). However, the plans for student facing use have remained stagnant (79 to 82 percent). The data also suggests that fear has been a motor—driving teachers to learn and use—rather than a brake. Those apprehending student misuse are 1.43 times more likely to be planning a student facing use; more likely to plan assessment redesign; and more likely to receive training.
Trend 2: Adoption and use of AI stop at the teacher’s desk, focusing on their own productivity gains rather than student use.
We asked teachers how they use AI in 2026. The four uses with highest percentage are all teacher productivity uses: lesson planning (83 percent); practice assessments (71 percent); use in the classroom (63 percent); and developing video content (53 percent). The three tasks aimed at helping student use AI more effectively are: pre-read with GPT (30 percent); improve their work (28 percent); and critique AI generated output (25 percent).
In 2024, 80 percent of teachers believed that AI would reshape learning. The assessment has remained steady; what has changed is the teachers’ capacity, training and use.
Asking students to critique AI output is the most valuable task a school can do with this technology; it is the least practiced. This is not unique to Pakistan. Even in the US, 54 percent of teachers use AI for planning and admin tasks while 23 percent use it in the classroom. This is where most of the attention is going now globally. Pakistani academia needs to double down on adopting higher end uses.
Trend 3: Governance has lagged behind.
19.5 percent teachers reported that their school has a written policy on AI. One in three did not know if their school had one, which is as consequential as not having one. Zero percent of public-school teachers reported a written policy in 2026. This is the biggest deficit. It is important because leaving how and when to use AI to a teacher is a double-edged sword.
While not telling teachers what to do may allow experimentation and innovation, in the absence of proper pedagogical training, teachers are more likely to use (or allow its use) improperly than properly. The long-term harm from misuse is far more than the short-term gains from allowing diversity of use. A written policy establishes a floor.
For example, if AI’s use is going to be more restrictive in primary and secondary years than in high school, it is better to say so explicitly so that teachers, in their enthusiasm, don’t err.
Trend 4: Willingness to acquire training has increased; willingness (among teachers) to pay for it has nearly tripled.
Teachers in schools with least institutional support are most willing to pay for training themselves. Low-fee private school teachers are most likely to pay (38 percent); public school teachers are the least likely (17.6 percent), even though most teachers still attend a school-organised (free or partner-run) training event.
Overall schools organise training for 38 percent of teachers but pay 19 percent of the time. Many convene, but not yet fund. Teachers who have attended school-paid trainings are 2.6 times more likely to pay for more training themselves. Where schools organised a free-event, they are half as likely to pay for more training.
School investment signals value; free convening signals obligation. Trainings still lag in high-impact areas of pedagogy, which should be the focus of future investments in training.
Trend 5: Data suggests four different archetypes of teachers: frontier practitioners; convinced planners; sceptical planners; and peripheral.
Frontier practitioners (25 percent) have student facing use planned; 91 percent have students critique AI; 71 percent are designing assessments. They spend 10.4 hours a week using AI and are most likely to be in a school with a written policy. But they are least institutionally trained: they are self-motivated and self-trained.
Convinced planners (39 percent) believe AI reshapes learning (100 percent); use AI for lesson planning (100 percent); do not use AI for student tasks; spend about 5 hours per week. They have mostly received school-organised training. This is the primary product of institutional training.
Sceptical planners (20 percent) do not believe AI reshapes learning (0 percent) but use it for lesson planning anyway (100 percent). They are most trained and most willing to pay. They use AI as a productivity hack and not a learning tool.
Peripheral users (16 percent) do not use AI for much, not even lesson planning; they are least encouraged by schools; they are the youngest; and least willing to pay for training.
Important lessons
Selling the why is not necessary; creating capability and permission is the real bottleneck. To some extent, the low-hanging fruits of productivity have been harvested. Governance and school policy are easy to execute but higher-end uses of AI is where the greatest opportunity lies. Focus should be on what schools are teaching themselves (mostly, low-level productivity hacks) and improve upon it.
Pakistan does not lag too far behind other countries. However, the gap is beginning to widen. Veterans (20+ years of experience) are the most eager learners and implementers, and use of generic tools such as ChatGPT (94 percent), Claude (54 percent), and NotebookLM (34 percent) lead the more specialist ones such as MagicSchool (16 percent) and Khanmigo (16 percent).
Schools must take the lead in writing and publishing a policy; funding training; rewriting the training brief to focus on high-end; protecting 6-10 hours a week of AI use (which is where depth of use truly arrives); and focus on redesigning assessments.
Most importantly, schools must systematically evaluate AI use in the classroom, run experiments, evaluate impact on learning, and collect data and evidence. We must create a genuinely local body of evidence for the Pakistani school context. Teachers must become action researchers within their own contexts, supported by a credible research framework that they adhere to.
All schools can and must participate and contribute to the national repository of AI learning evidence. Because the results of studying three years of AI use by teachers show that teachers are not the constraint, institutions are.
The writer is the founder and CEO of Learning Innovation and Teaching Excellence (LITE) Lab and STEMx – a STEM School for the World. He is also the curator of the Holistic Science Teaching Fellowship and #AIDotEDU and #GenAI4Univ. He can be reached at [email protected].