AI in Education: What the Google/Ipsos Survey Actually Shows
Editorial update: November 2, 2026. This article summarizes a Google/Ipsos survey and offers a practical evaluation checklist for education tools. Survey responses describe reported use and attitudes; they do not establish that an AI product improves learning outcomes.
What the Google/Ipsos survey reported
In a January 15, 2026 article, Google's Ben Gomes described the third annual “Our Life with AI” study, conducted with Ipsos in late 2025 across 21 countries and 21,000 participants.
Google reported that 74% of AI users used it to learn something new or understand a complex topic. It also reported AI use among 85% of students aged 18 and over and 81% of teachers, compared with 66% of the global public surveyed. The student figure should not be generalized to children or to every school system.
These results indicate substantial reported adoption within the surveyed groups. They are not a controlled comparison of teaching methods, and they do not validate a particular API provider, classroom deployment or improvement in test scores.
Turn an education use case into a measurable task
The following workflow is our implementation guidance, rather than a finding of the survey. Start with a narrow task: for example, drafting comprehension questions from a teacher-approved passage. Keep the passage, target reading level and scoring criteria fixed while comparing outputs.
Before building a full application, define what a teacher should be able to verify:
| Criterion | A concrete check |
|---|---|
| Grounding | Each question can be answered from the supplied passage |
| Answer quality | The proposed answer matches the passage and has no invented facts |
| Difficulty | Vocabulary and reasoning match the intended reading level |
| Usefulness | A teacher can accept, revise or reject each question |
| Reliability | Timeouts and invalid output produce a visible retry or review step |
Use synthetic passages or material approved for the evaluation. Begin without student names, private submissions or other personal records. That makes it easier to review outputs and share failures while deciding what information the eventual workflow needs.
Measure the whole workflow
For a small evaluation set, record model identifier, prompt version, request settings, latency, token usage and teacher review decisions. Measure how much editing the draft needs, not just how quickly the API responds. Compare against a baseline, such as the teacher's existing question-writing process, before describing a time saving.
If you use CCAPI, consult the API documentation for the endpoint and request format, and the pricing page for the selected model. Confirm those details when implementing; this article does not prescribe an unverified package or SDK. Keep API credentials on the server and make the review step part of the product.
What to conclude
The survey is a useful reason to investigate learning workflows. Whether a specific tool helps a particular group of learners requires its own evaluation, appropriate permissions and teacher judgment. Adoption percentages should not be presented as evidence of higher scores, lower costs or faster API responses.
Read Google's original survey article for its reported findings and context. This revision removes unsupported institutional case studies, outcome and latency figures, and a fabricated ccapi-sdk example. The original CCAPI publication date and URL are retained.