Agentic education · Learning science

Actionable companion · Cognitive load & attention

From AI evidence to responsible practice: ChatGPT in education: global reactions to AI innovations

AI-assisted editorial implementation guide, separate from the original research. Proposed actions are not findings or recommendations attributed to the source authors. No independent peer review of this guide is asserted.

Read original paperSearch paper agentCC BY 4.0

Evidence and context

Title and abstract screening; stage labels describe reported settings, not verified numerical ages. Design labels require full-text appraisal. Reported stage(s): Age / stage not established. Verify the participants, intervention and comparator before applying this study in a new setting.

Original authors’ abstract, reproduced under the linked source license:

The release and rapid diffusion of ChatGPT have caught the attention of educators worldwide. Some educators are enthusiastic about its potential to support learning. Others are concerned about how it might circumvent learning opportunities or contribute to misinformation. To better understand reactions about ChatGPT concerning education, we analyzed Twitter data (16,830,997 tweets from 5,541,457 users). Based on topic modeling and sentiment analysis, we provide an overview of global perceptions and reactions to ChatGPT regarding education. ChatGPT triggered a massive response on Twitter, with education being the most tweeted content topic. Topics ranged from specific (e.g., cheating) to broad (e.g., opportunities), which were discussed with mixed sentiment. We traced that authority decisions may influence public opinions. We discussed that the average reaction on Twitter (e.g., using ChatGPT to cheat in exams) differs from discussions in which education and teaching–learning researchers are likely to be more interested (e.g., ChatGPT as an intelligent learning partner). This study provides insights into people's reactions when new groundbreaking technology is released and implications for scientific and policy communication in rapidly changing circumstances.

What kind of evidence is this?

Supporting AI learning science · Design requires full-text appraisal

Stage: Age / stage not established. Screening labels describe reported context and study-design language. They are not a risk-of-bias rating or proof of effectiveness for an age band.

Study methods · source passage

Using the Twitter API for Academic Research, we collected 16,830,997 tweets posted from November 30, 2022, to January 31, 2023. We chose this rollout period to get an initial reaction from the public before many had spent much time thinking about or using ChatGPT. The data collection procedure was as follows: first, we queried the Tweets that mentioned ChatGPT. Second, we identified and collected all conversations with either a sparking tweet (i.e., tweets that start a discussion) or a reply mentioning ChatGPT. A conversation is defined as the sparking tweet and all other tweets directly replying to the sparking tweet. Notably, a conversation needs to include at least two different users. This led to 127,749 unique conversations. Notably, we found no mentions of ChatGPT on Twitter before November 30, 2022.

Limitations and uncertainty · source passage

The results of this study must be interpreted considering at least the following limitations. First, although we have thoroughly prepared the data for the analyses, problems arise due to bots. Bots are computer programs that perform automated tasks, such as automatically retweeting tweets with a specific hashtag. Bots are challenging to detect because their behavior constantly changes as technology advances40. We used heuristics and text-based approaches to identify and remove tweets posted by bots from our data but cannot guarantee that tweets from bots are still present in the data used. Second, due to our limitation of Tweets to those in English, our geographical focus is around 57% North America. Thus, our findings may not be generalizable to other regions. Third, insight into the sample used for this study is limited by the information available through the Twitter API. For instance, we could not accurately determine whether politicians, academics, entrepreneurs, or opportunists tweeted. However, we were able to provide some general insights into the sample by analyzing, for instance, the sample’s experience with Twitter (operationalized by Twitter sign-up date) or the users’ reach (operationalized by number of followers; see Figs. D and E and Table A in the appendix). Fourth, we analyzed only the first two months of conversations on Twitter after the release of ChatGPT. This means that subsequent discussions based on more experience with ChatGPT (educational stakeholders had more time to figure out strengths and weaknesses and use ChatGPT for, e.g., teaching–learning scenarios) or that included GPT4 were not considered. However, it was precisely the approach of this study to capture the unvarnished, rather unreflective reaction of humans to groundbreaking technological innovations. Moreover, this study goes beyond previous findings on human reactions to ChatGPT, some of which only focused on the first few days after the release of ChatGPT (e.g.31).

Design the system before the pilot

Use this study to inform a learning mechanism or implementation question. It is not direct evidence that autonomous agents improve learning. Keep the proposed agent behavior separate from what the source actually tested.

  • Limit simultaneous instructions and unnecessary dialogue. Compare task performance and a suitable mental-effort measure; easier-feeling work is not automatically better learning.

A 90-day district pilot

  1. Days 1–15 · Appraise. Have the curriculum lead and a teacher read the methods and limitations. Record the study design, sample, comparator, outcomes and evidence gaps. Define a local question that the study can help inform.
  2. Days 16–30 · Design. Treat the paper as background evidence until its population is verified in the full text. Do not assign an age-specific intervention or extend adult results to children. Name an accountable owner and set the baseline, success threshold and budget before implementation.
  3. Days 31–60 · Test. Start with a small voluntary cohort, preserve an accessible non-AI route and require teacher review. Use synthetic or de-identified examples; approve data handling and tool terms through district processes.
  4. Days 61–90 · Decide. Extract the participant ages or education stage, recruitment, task and comparator before selecting a local pilot. If these cannot be established, retain the paper for design discussion rather than age-band efficacy claims. Compare with baseline, document uncertainty and stop or revise if learning worsens, access is unequal or critical errors persist.

The board decision brief

Report the learning problem, evidence quality, participating population, total cost including review time, measured benefit, errors, access differences and remaining uncertainties. Make a dated recommendation to stop, revise or expand. A positive perception survey alone does not establish learning improvement.

Attribution and reuse

Tim Fütterer; Christian Fischer; Anastasiia Alekseeva; Xiaobin Chen; Tamara Tate; Mark Warschauer; Peter Gerjets. ChatGPT in education: global reactions to AI innovations. Scientific Reports (2023-09-15). https://doi.org/10.1038/s41598-023-42227-6

Companion prepared 2026-11-01. Original authorship and DOI are preserved. The abstract is an attributed source extract; the implementation plan is new AI-assisted editorial material, offered under the same CC BY 4.0 license. This guide has not undergone independent peer review.