Oct 16 edition/Reporting & analysis
BusinessResearch

BusinessMarkets, money & strategy

OpenAI launches a free AI course for Brazilian teachers on Teachers' Day and promotes ChatGPT's Study Mode

OpenAI marked Brazil's Teachers' Day by releasing a free Portuguese-language AI for Educators course and promoting ChatGPT's Study Mode, a prompt-based tutoring setting that questions students rather than giving answers. Evidence that it improves learning remains mixed.

From the OpenAI News article of October 15, 2026 and cited sources: On October 15, 2026, OpenAI released a free AI for Educators course on OpenAI Academy in Brazilian Portuguese. [1] [2] [8] [3]

SOURCES[1] Teaching students to think and learn through questions —… openai.com[2] OpenAI homenageia professores em campanha… — Exame exame.com[8] OpenAI aborda uso da IA nas salas de aula — Meio & Mensagem meioemensagem.com.br

AI-assisted, not human-edited. Methodology Corrections Who owns this site

KEY FACTS
What happened
On October 15, 2026, OpenAI released a free AI for Educators course on OpenAI Academy in Brazilian Portuguese. [1] [2] [8] [3]
Original source
OpenAI News article, company source
Brief by
AT Pulse research desk AI-assisted; publisher and contact on the desk page
When
October 15, 2026 article published
Evidence
16 sources across 13 sites

Reach: not measured yet

THE CORE IDEAS4 TAKEAWAYS
01

On October 15, 2026, OpenAI released a free AI for Educators course on OpenAI Academy in Brazilian Portuguese. It is aimed at primary, secondary and university teachers and covers lesson planning, activities, assessments and communications. A humorous campaign film set in a first-year university classroom accompanied the launch. Brazil has about 50 million monthly ChatGPT users. [1] [2] [8] [3]

02

Study Mode is not a separately trained model. It is a layer of system instructions over general ChatGPT models that guides students with questions, hints and checks for understanding. Students can switch back to regular ChatGPT at any time, and any errors in the underlying model carry over. [6] [12] [9]

03

OpenAI's own preliminary study of more than 300 university students found grades about 15% higher in microeconomics and no significant difference in neuroscience. These results come from OpenAI and have not been independently replicated, so they show gains are not consistent across subjects. [2] [11]

04

A peer-reviewed field experiment with nearly 1,000 high-school math students in Turkey found that unrestricted GPT-4 raised practice scores but lowered later exam scores. A tutor-style version that guided students instead of giving answers avoided that drop. [4] [5] [15]

WHY IT MATTERS

Peer-reviewed research shows that unrestricted chatbots can undermine learning, while tutor-style prompting limits the harm.

Read the full assessment

Because Study Mode is a toggle rather than a lock, schools that need guardrails enforced will likely need their own policies or platform-level controls.

I've written the dossier below.

Executive brief

OpenAI's Teachers' Day push in Brazil leans on Study Mode, a ChatGPT feature that asks students questions instead of handing them answers. The best evidence for it is mixed. OpenAI's own study of 300+ university students found roughly 15% higher grades in microeconomics and no significant difference in neuroscience (Exame). On October 15, 2026, OpenAI released its free "AI for Educators" course in Brazilian Portuguese (OpenAI). The news is about distribution and marketing in a market of about 50 million monthly ChatGPT users (Forbes Brasil), not a new model.

What changed and event timeline

  1. Guardrails study lands in PNAS

    In a field experiment with nearly 1,000 Turkish high-school math students, unrestricted GPT-4 raised practice scores but lowered later exam scores. A tutor-style version avoided the drop (;).

  2. Study Mode launches

    It is built on system instructions, not a separately trained model. It was open to Free, Plus, Pro and Team users, with ChatGPT Edu to follow ().

  3. OpenAI Academy in Portuguese

    Three free tracks launched: ChatGPT fundamentals, prompting, and text/image generation ().

  4. Measurement tools announced

    Stanford SCALE described a Learning Outcomes Measurement Suite that tracks learning over time. It cited earlier Study Mode gains as "promising" and said a randomized trial was still running ().

  5. Teachers' Day launch in Brazil

    OpenAI released "IA para Educadores" (AI for Educators) and ran a humorous campaign film set in a first-year university classroom (;).

Capabilities and access

  • Course: "AI for Educators" is free on OpenAI Academy.
  • Study Mode: guides students with questions, hints and checks for understanding. It is switched on inside ChatGPT (AI Forradalom).
  • Model version: the reviewed sources don't name the model behind Study Mode in October 2026.
Read the full section
  • Course: "AI for Educators" is free on OpenAI Academy. It targets primary, secondary and higher-education teachers and covers lesson planning, activities, assessments and communications (Exame).
  • Study Mode: guides students with questions, hints and checks for understanding. It is switched on inside ChatGPT (AI Forradalom).
  • Model version: the reviewed sources don't name the model behind Study Mode in October 2026. At launch, behaviour differed between the paid o3 and free GPT-4o tiers (Leon Furze).

Technical analysis for researchers and developers

  • Architecture: Study Mode is a system-prompt layer over general ChatGPT models. It draws on learning-science ideas such as active learning, metacognition and feedback.
  • Reproducibility: the PNAS "GPT Tutor" results show that prompt-level safeguards can be tested independently against both a no-AI control and an unrestricted-AI arm (Wharton).
  • Evaluation: SCALE warns that test scores are a narrow signal and favours measuring learning over time in real settings (Stanford SCALE).
Read the full section
  • Architecture: Study Mode is a system-prompt layer over general ChatGPT models. It draws on learning-science ideas such as active learning, metacognition and feedback. OpenAI said it plans to build this behaviour into core models later and acknowledged that results may be inconsistent until then (The Decoder).
  • Reproducibility: the PNAS "GPT Tutor" results show that prompt-level safeguards can be tested independently against both a no-AI control and an unrestricted-AI arm (Wharton).
  • Evaluation: SCALE warns that test scores are a narrow signal and favours measuring learning over time in real settings (Stanford SCALE).

Claims and evidence

No independent reporting confirms what the Brazil course actually contains beyond OpenAI's description.

Read the full section
  • About 15% higher microeconomics grades with Study Mode; no significant effect in neuroscience; 300+ students; March 2026. These are OpenAI's own preliminary results, not independently replicated (Exame; OpenAI).
  • Designed with educators and pedagogy experts from 40+ institutions. This is OpenAI's own account (MIT Technology Review).
  • Brazilian teachers spend a third of their time on admin, 40% above the global average. This figure appears only in OpenAI's announcement as summarized in search results; its underlying data source is not identified (OpenAI).
  • Unrestricted AI can hurt later performance; tutor-style prompts reduce the harm. This comes from a peer-reviewed study, not OpenAI (IDEAS/RePEc).

No independent reporting confirms what the Brazil course actually contains beyond OpenAI's description.

Context and prior work

  • OpenAI has an earlier Brazil education case study with Arco Educação, which used GPT-4 (OpenAI).
  • It also offers a free ChatGPT version for teachers (OpenAI).
  • Its Learning Lab partners include Arizona State University, UCL Knowledge Lab and MIT Media Lab (Stanford SCALE).
Read the full section
  • OpenAI has an earlier Brazil education case study with Arco Educação, which used GPT-4 (OpenAI).
  • It also offers a free ChatGPT version for teachers (OpenAI).
  • Its Learning Lab partners include Arizona State University, UCL Knowledge Lab and MIT Media Lab (Stanford SCALE).
  • Bastani et al. remain the main independent reference for the "AI as a crutch" problem (SSRN).

Limitations, safety and contested findings

  • Easy to bypass: students can switch back to regular ChatGPT, and OpenAI's head of education acknowledged this (MIT Technology Review).
  • Accuracy: Study Mode is not trained only on curated academic material, so errors carry over from the general model (MIT Technology Review).
  • Criticism from educators: Leon Furze reported about six pedagogical problems in roughly 10 minutes of use, plus math errors found by others, and called the feature mainly a positioning move (Leon Furze).
Read the full section
  • Easy to bypass: students can switch back to regular ChatGPT, and OpenAI's head of education acknowledged this (MIT Technology Review).
  • Accuracy: Study Mode is not trained only on curated academic material, so errors carry over from the general model (MIT Technology Review).
  • Criticism from educators: Leon Furze reported about six pedagogical problems in roughly 10 minutes of use, plus math errors found by others, and called the feature mainly a positioning move (Leon Furze).
  • Mixed results: the null result in neuroscience shows gains are not consistent across subjects (Exame).

Business and practitioner implications

  • Market strategy: OpenAI is using free courses and local-language marketing to embed ChatGPT in Brazilian classrooms (Propmark).
  • For edtech builders: guided-tutor prompting has documented benefit (Wharton). But Study Mode alone is a toggle, not a lock.
  • For buyers: treat OpenAI's learning-gain figures as preliminary until the randomized trial and SCALE measurement suite publish results.
Read the full section
  • Market strategy: OpenAI is using free courses and local-language marketing to embed ChatGPT in Brazilian classrooms (Propmark).
  • For edtech builders: guided-tutor prompting has documented benefit (Wharton). But Study Mode alone is a toggle, not a lock. Institutions needing enforced guardrails need policy or controls at the platform level.
  • For buyers: treat OpenAI's learning-gain figures as preliminary until the randomized trial and SCALE measurement suite publish results.
FOLLOW THE EVIDENCE

The source trail.

Sources (16)
A LITTLE LESS NOISE. A LOT MORE CONTEXT.

Stay curious.
Follow the evidence.

Independent perspectives, the original sources, and room for the questions that don't have easy answers.

How we build the brief
Connect with us

Find us where you already read.