By
Stuart Kerr, Technology Correspondent,
LiveAIWire

Google’s rollout of AI Mode in Search, announced at Google I/O in
May 2025, extended the company’s AI Overviews feature into a fuller, more
conversational search experience, one that increasingly resembles a study
companion rather than a results page. Under the hood, AI Mode uses what
Google calls a query fan-out technique, breaking a single question into
several subtopics and issuing multiple searches simultaneously to assemble a
more complete answer than a single query would return. For students, that
shift matters because it changes what a search actually returns: not a list
of links to sift through, but a synthesised, multimodal response that can
incorporate text, images, and increasingly live camera input through Search
Live, Google’s extension of its Project Astra visual reasoning technology
into Search itself.

The classroom implications of that shift are still being worked
out, but the direction is clear. Google reports that people are already
coming to Search with longer, more complex, and more multimodal questions
than before, and that usage of AI Overviews drives a measurable increase in
return visits for the types of queries where it appears. A search experience
that can accept a photograph of a diagram or a handwritten equation, reason
about it, and respond with an explanation tailored to the question is a
materially different tool for homework help than a conventional keyword
search, and students appear to be adopting it accordingly.

From Search Feature to Study Tool

The multimodal capability is only part of what has changed. Alongside
AI Mode, Google has been building a parallel set of education-specific tools
under the LearnLM banner, its family of models fine-tuned specifically for
pedagogy rather than general-purpose reasoning. LearnLM is now integrated
directly into Gemini 2.5, and its influence is most visible in two features
aimed squarely at students: Guided Learning, a Socratic-style tutoring mode
that prompts students toward understanding through questions rather than
handing over direct answers, and Learn Your Way, a research experiment that
transforms static textbook material into a set of interactive, multimodal
formats a student can choose between.

The evidence that this approach actually improves learning outcomes,
rather than simply making study feel more engaging, comes from a randomised
controlled study Google conducted with 60 students aged 15 to 18 in the
Chicago area. Students were given up to 40 minutes to study a textbook
chapter on adolescent brain development, split between a group using Learn
Your Way and a group using a standard digital PDF reader. The results,
published by Google Research in September 2025, showed the Learn
Your Way group scored 9 percent higher on an assessment taken immediately
after the study session, and 11 percent higher on a retention test taken
three to five days later, 78 percent average versus 67 percent for the
control group. Student sentiment tracked the outcomes: all of the students
who used Learn Your Way reported feeling more comfortable going into the
assessment, compared with 70 percent of the control group, and 93 percent
said they would want to use the tool again for future learning.

What Multimodality Actually Adds to Studying

The theoretical basis for combining multiple formats, text, audio,
slides, mind maps, quizzes, rather than relying on a single mode of
presentation, rests on dual coding theory, the long-standing finding in
cognitive psychology that forging connections between different
representations of the same material strengthens the underlying mental
model a learner builds. Google’s implementation breaks source material into
digestible sections augmented with generated images and embedded questions,
offers narrated slide presentations spanning the full material, generates
simulated audio conversations between an AI tutor and a student modelling
common misconceptions, and produces hierarchical mind maps that let a
student zoom between the big picture and the specific detail.

None of these individual formats is new to educational technology.
What AI Mode and the LearnLM tools built on top of it change is the cost of
producing them. Generating a narrated slide deck, a set of section quizzes,
and a mind map from a single textbook chapter previously required
substantial teacher time or dedicated instructional design resources.
Automating that generation, while personalising the content to a specific
student’s stated grade level and interests, is what allows the approach to
scale to individual students rather than remaining a resource only
well-funded schools could provide.

The Adoption Numbers

Google reports that in 2025, more than a million educators and
students received AI training through Google for Education, with more than
100,000 earning Gemini certifications through the platform’s Learning
Centre. Gemini in Classroom, which lets teachers generate quizzes, lesson
materials, and differentiated activities grounded in their own class
content, is now available free of charge to Google Workspace for Education
users. That combination, free access bundled into an education suite many
schools already use, is likely a more significant driver of classroom
adoption than the underlying model capability itself, since it removes the
procurement and cost barriers that have slowed adoption of other education
technology.

The competitive context matters here too. Google’s Guided Learning
launched within roughly a week of OpenAI’s Study Mode and follows
Anthropic’s Claude for Education, launched earlier in 2025. All three
approaches share the same basic pedagogical premise, that an AI tutor should
guide a student toward an answer rather than simply supplying one, but
Google’s specific advantage lies in the breadth of its existing content
ecosystem. Search, YouTube, Classroom, and NotebookLM together give Google’s
tools access to a volume and diversity of educational content, and in
Classroom’s case, direct visibility into what a specific class is actually
studying, that a standalone chatbot cannot easily replicate.

What Remains Unresolved

The efficacy data so far comes from a single study, a single subject
area, and a single age range, and Google itself has acknowledged that
population diversity, subject-domain coverage, and replication across
different contexts are the open questions that will determine whether the
early results generalise. A nine-to-eleven-point improvement over a
standard digital reader in one Chicago classroom is a promising signal, not
a settled finding, and the honest reading of the evidence is that multimodal
AI study tools look genuinely useful rather than definitively
transformative at this stage.

The concerns that apply to AI in education generally apply here too.
Over-scaffolding, where a tool provides so much structured support that
students practise recognising the right answer rather than recalling it
unaided, is a real risk that Google’s own researchers have flagged.
Personalisation built on assumptions about learning styles or cultural
context that do not hold for every student remains a design risk rather
than a solved problem, and opacity, where a student or teacher cannot easily
tell why a tool generated a particular explanation or changed its approach,
is a legitimate transparency concern for any adaptive system used in a
classroom setting. Multimodal AI search and study tools are moving from
laboratory demonstration into everyday classroom use faster than the
evidence base evaluating their long-term effects is accumulating, and that
gap is worth watching as adoption scales.

About the Author

Stuart Kerr is the Technology Correspondent for LiveAIWire. He
writes about artificial intelligence, emerging technology, and the forces
reshaping work, business, and society.