Canvas
in AI Labs: The Future of Structured Planning Powered by AI in
Search
The search box has been the primary interface between humans and
digital information for nearly three decades. Type a query, receive a list of
links, click through to sources, and construct understanding from what you
find there. The process is so familiar it has become invisible, a reflex as
automatic as reaching for a light switch. Google’s Canvas feature, introduced
as part of AI Mode in mid-2025, represents the most significant departure
from that paradigm since the search engine itself was
invented.
Canvas is not an incremental improvement to search. It is a
different model of the human-information relationship: one in which the
search interface is also a workspace, where queries become projects, where
information retrieved is immediately organizable, and where the AI that
retrieved the information continues to assist with structuring and developing
it. The implications extend well beyond productivity enhancement into
questions about how we think, how we plan, and what it means for a digital
tool to participate in the construction of knowledge rather than merely
facilitating access to it.
What Canvas Actually Does
Canvas is a persistent side-panel workspace within the AI Mode
search interface. Unlike a standard search result page, which presents
information and then disappears when the user navigates away, Canvas
maintains state across a session and, in some implementations, across
sessions. A user researching a topic can build a structured document within
Canvas as they search, pulling in AI-synthesised information from successive
queries, organising it into sections, and developing it into a plan, an
outline, or a working document.
The inputs Canvas accepts go beyond text. Users can upload PDFs,
images, and handwritten notes, which the AI processes and integrates into the
workspace. A student can upload a reading list and ask Canvas to organise the
sources by theme. A project manager can sketch a timeline and ask Canvas to
flesh it out with suggested milestones. A researcher can upload a draft and
ask Canvas to identify gaps against a literature search. As The
Verge reported on Canvas’s multimodal capabilities, the feature’s
ability to maintain context across inputs is what distinguishes it from a
simple AI chat interface.
Lifewire’s
review of Canvas in AI Mode characterises the feature as a
transformation of search from information retrieval toward something closer
to a productivity platform. That framing is accurate but incomplete: Canvas
is not trying to be a competitor to Notion or Google Docs. It is trying to
make the information retrieval and organisation process continuous,
eliminating the friction of moving between research and
synthesis.
The Cognitive Shift
The design philosophy behind Canvas reflects a view of how
research and planning actually work that is more accurate than the
traditional search model. Information retrieval and information organisation
are not sequential processes that happen one after the other; they are
iterative and interleaved. A researcher who finds a useful source typically
wants to note its relevance immediately, in context, rather than collecting a
list of tabs to organise later. A planner developing a project timeline wants
to see how new information changes the shape of the plan as they discover it,
not after they have finished searching.
Canvas makes this interleaving possible by treating the search
interface itself as a planning environment. The AI does not just retrieve
information in response to queries; it responds to the structure the user is
building, suggesting how new material fits into existing frameworks and
flagging tensions between different pieces of information.
A recent arXiv paper on
two-dimensional interfaces for AI-assisted cognition provides
theoretical grounding for this design direction. The paper argues that notebook
and canvas models of human-computer interaction are better suited to complex
knowledge work than the linear chat interfaces that dominate current AI tool
design, because they support the spatial organisation of information in ways
that reflect how working memory actually handles complex
tasks.
The Data Implications
Canvas creates a new category of user data that Google has not
previously had access to: the structure of how users organise information,
not just what they search for. When a user builds a research outline in
Canvas, they are revealing not only their information needs but their mental
models, their conceptual frameworks, and the connections they draw between
ideas. This is significantly richer behavioural data than search query logs,
and its commercial and privacy implications deserve
attention.
The concerns this raises are related to those examined in AI
and Emotional Manipulation: when AI systems have detailed knowledge
of how users think and organise information, the potential for manipulation,
whether through targeted advertising, personalised persuasion, or subtle
framing of AI-generated content, expands significantly. A Canvas workspace
is, among other things, a detailed map of a user’s current project and their
approach to it.
Google’s privacy disclosures for Canvas are not yet as specific as
the data implications warrant. What is retained, for how long, and under what
circumstances it is used for model training or advertising personalisation
are questions that users of the feature cannot currently answer from publicly
available documentation.
Canvas and Educational Use
The educational applications of Canvas are simultaneously its most
compelling and most contested use case. For students conducting research,
Canvas offers genuine assistance: it can help structure a literature review,
identify connections between sources, generate outlines from reading
materials, and provide AI-generated first drafts of sections that students
can then develop and revise.
The pedagogical concern is familiar from previous rounds of
AI-in-education debate: tools that do cognitive work on behalf of students
may reduce the development of the cognitive skills that the work is intended
to build. The difference with Canvas is one of degree. Previous AI writing
tools could generate text; Canvas can generate the entire research and
structuring process, potentially eliminating the intellectual labour that
makes research educationally valuable.
As examined in Teaching
Tomorrow: Can AI Rebuild the Curriculum from Scratch?, the
appropriate integration of AI tools into educational contexts requires
deliberate curriculum design that distinguishes between tasks where AI
assistance supports learning and tasks where it substitutes for it. Canvas is
powerful enough that this distinction requires explicit attention from
educators who deploy it in academic settings.
Competitive Implications
Canvas extends Google’s ambition in AI Mode from answering queries
to hosting the entire research and planning workflow. That ambition places
Google in direct competition with a much wider range of software categories
than search traditionally addresses: note-taking tools, project management
platforms, document editors, and research management
software.
As explored in Inside
Google’s AI-First Shift, Google’s internal culture is shifting toward
efficiency-first product development that attempts to deliver more capability
with less computational overhead. Canvas is a product expression of that
shift: it attempts to make search more valuable by making it stickier and
more integrated into users’ workflows, increasing session depth and data
richness without proportionate increases in computational
cost.
The platform risk for productivity software companies is real. If
Canvas becomes the default research and planning interface for users who
would previously have moved between a search engine and a separate
productivity tool, the market for standalone research and note-taking
applications contracts. That competitive pressure is one reason why Canvas’s
data practices and market power implications are attracting regulatory
attention alongside its user-facing features.
About the Author
Stuart Kerr is the Technology Correspondent for LiveAIWire,
covering artificial intelligence, ethics, and the ways technology is
reshaping everyday life.