Gemini Notebook Can Now Turn Your Sources Into Interactive Websites
On 16 November 2026 Google gave Gemini Notebook a 'make me a site' feature: one sentence turns the sources in a notebook into an interactive website — visual explainers, data dashboards, or project trackers — opened in a new tab or shared with others. The web rollout is global from today for users 18+, including Workspace subscribers; mobile is coming soon.

In short: your notes no longer have to sit in a notebook — they can become a website
On 16 November 2026, Google announced on its official blog that Gemini Notebook (formerly NotebookLM) has gained a new ability: turning the sources in a notebook directly into an interactive website. In Google's words, it takes a single sentence — type "make me a site" in the notebook's chat, describe what you want, and the rest is handled for you. The announcement was written by Maria Ines Aranguren, a software engineer at Google Labs, and published on 16 November 2026 at 16:00 UTC. We read the full post.
What kinds of sites can it make? Google names three. The first is visual explainers for exam prep: turn a pile of chemistry notes into a clickable, interactive periodic table, or history notes into a timeline. The second is interactive dashboards for exploring data: turn a few sales documents into a quarterly sales dashboard. The third is project management trackers: tasks and deadlines laid out in a table you can actually use. Google's demo video adds two details: the periodic table page ends with a five-question quiz, and the sales dashboard can be filtered by region, showing revenue against quota and pipeline charts.
Each of the three is worth a closer look. The periodic table is about more than looking nice — it is about being clickable: tap an element to see its properties, then take a few quiz questions, so revision turns from flipping through notes into doing something. The sales dashboard is about filtering: slice by region and it is immediately clear which area is behind; put revenue next to quota and half the argument in the quarterly meeting disappears. The tracker is about lining things up: tasks, owners, and deadlines in one table, so a meeting no longer starts with digging through chat logs to find out where things stand. The idea is the same in all three: the sources do not change — what changes is how you work with them, from reading to clicking, filtering, and checking.
How "interactive" is it, really? Judging by Google's demo, there are at least three layers: click for detail (tap an element for its description), filter for a slice (cut the dashboard by region), and test yourself (the five questions at the end of the periodic table). That is a real step up from exporting a document as a static web page — but do not mistake it for a full application. There is no login, no database, and no backend where users enter data and get computed results. Think of it as a "living source page" rather than a "website that runs a business." For studying, reporting, and getting a team on the same page, that is enough; for collecting feedback, taking orders, or running complex logic, it is not there yet.
What happens after it is built? Click "Open in a new tab" to open and interact with it, or share it with others. The rollout: from 16 November, on the web globally, for Gemini Notebook users aged 18 and over, including Google Workspace subscribers. The mobile app is "coming soon," with no date given.
Two details in the rollout are worth pausing on. First, Google Workspace subscribers are included, which means this is not just a personal tool from day one: a teacher can turn revision material into an interactive page for students, and a team can turn research into a dashboard in shared space — schools and companies are its first real users. Second, the 18-and-over limit sits awkwardly against the exam-prep pitch — the school students who need study tools most are not covered, at least not in the first wave. Put together, the picture is clear: the first audience is adults who learn and work, not students of every age.
Getting started takes three steps. First, create a notebook and put your sources in — papers, notes, spreadsheets, links; the fuller the material, the more the generated page has to work with. Second, type "make me a site" in the chat and describe what you want: a study page for exam prep, a sales dashboard for the team, or a task tracker; which numbers to highlight and what interactions to include — the more specific you are, the closer the result. Third, click through it yourself before sharing: are the numbers right, is the content really from your sources, do the links point where they should? One extra sentence in the description saves a round of fixes later.
The demo's second example deserves its own mention: five sales documents become one dashboard. The point is not "five" but "synthesis" — the model first has to line the numbers up across documents: which one is this quarter's, which is last quarter's, whether definitions match, whether a region is missing — and only then draw the charts. For a person, that is the most time-consuming part: opening five files, cross-checking, then consolidating. For a model, it is the work it is best at — and also the work where it can quietly go wrong: the same metric named differently in two documents may be treated as two metrics; a missing region may simply be skipped without a warning. So the first thing to do with a generated dashboard is not to admire it, but to spot-check two or three numbers against your own documents.
From "understanding sources" to "publishing sources"
This makes more sense in the NotebookLM timeline. The tool first appeared as Project Tailwind at Google I/O 2023, positioned as a "notebook for research grounded in your own sources": upload papers, notes, spreadsheets, and links, and let AI summarise, compare, and quiz you on them. It was renamed Gemini Notebook in July 2026. This update is a bigger jump than it looks: it turns a tool that helped you understand sources into one that helps you publish them. The wall between research and publishing just got knocked down by a single prompt.
Nor is this the first time it has changed how its output arrives. Its most talked-about feature before this was Audio Overviews: turning a pile of sources into a podcast-style conversation between two hosts, so you could absorb a paper's key points on a commute. Now it turns sources into a website. The logic is the same both times: the input is always "your own sources," and the output is simply another medium — first audio, now the web. The benefit is consistent too: what is on the page can be traced back to a document you uploaded, not invented by the model. For regular NotebookLM users, this update feels less like a leap than the obvious next step: the sources were already in the notebook, so giving them a shareable form was only a matter of time.
For people who do vibe coding, the interesting part of this story is not "yet another AI that makes websites." AI that makes websites is everywhere now. What is special about Gemini Notebook is that its pages grow out of "your sources": the content is constrained by what you uploaded, not made up from nothing. That has been NotebookLM's selling point since day one — answers you can trace back to the document they came from. This update extends that promise from answering questions to publishing: what you hand over is no longer a chat transcript, but a page other people can open and click. In other words, the bar is moving: writing a good prompt is no longer scarce. What is scarce is a set of sources worth turning into a website, and the judgment to decide what the page should look like.
Compared with tools like v0 and Lovable, which generate websites from a single sentence, Gemini Notebook starts from sources rather than a blank prompt. Those tools are strong at producing a good-looking page from nothing; their weak spot is that you have to fill in the content and verify it yourself. Notebook is the mirror image — the content comes first, and the page is simply the content in another shape. For students, researchers, and anyone who has to present data to a team, that order makes more sense: real material first, presentation second. The ceiling is equally clear: this is not a general website builder, but a "dress your sources up as a web page" feature. Do not expect it to build a complete product site for you.
There is another layer for content creators. Making a study guide used to mean four steps across four tools: writing, layout, graphics, and building the page. Now, once the sources are ready, one sentence produces a page you can send out. What you save is not the "make a website" step, but the time lost shuffling between tools. The reverse is equally true: when everyone can produce a decent-looking page with one sentence, the page itself stops being the advantage. Whether your sources are original, whether your explanations land, whether your data is real — the parts that cannot be generated are what create the gap. AI has flattened the cost of execution; it has raised the cost of judgment.
Zooming out, "turning documents into a shareable page" is not Gemini Notebook's idea alone — Notion, Canva, and Gamma all do versions of it, and Google's own Gemini app and AI Studio can generate a site from one sentence. Notebook's position is a little different: it does not compete on how fancy the page looks, but on the thread connecting the content to its sources — every part of the page can be traced back to a document you uploaded. That thread is its real moat, and also its boundary: the more solid the sources, the more trustworthy the page; if the sources are empty, the prettiest page is still an empty shell. People who vibe-code will recognise the trade-off: the tool turns an idea into something visible, but the person has to make sure the idea is worth seeing.
What the announcement does not say matters more than what it does
Let us list what is missing first. One: how long a generated site stays up — Google does not say. Two: whether people who receive a shared link need a Google account to view it — not stated. Three: whether the site updates when the sources in the notebook change — not stated. Four: accuracy — the announcement contains no accuracy validation, so whether the numbers and conclusions on the page are faithful to the originals is something users have to test themselves. Five: price, usage limits, and which model writes the sites — none of it mentioned. Six: whether you can edit the code or export it afterwards — also not mentioned. Seven: the mobile app gets only the words "coming soon," with no timeline.
Those gaps do not make the feature itself invalid, but they decide what it can be used for right now. The fitting uses are clear: study material, internal dashboards, one-off sharing — things you make, use, and move on from. The unfitting uses are equally clear: pages published formally to the outside world, sources containing sensitive data, and pages that need long-term maintenance and updates. When in doubt, ask yourself three questions: who is responsible if a number on this page is wrong, can the link be taken back once sent, and does the source material contain anything that should not be public? Google has not answered these questions; that does not mean they do not exist — it means the judgment is left to the user.
One practical warning: the more convincing a generated page looks, the more it deserves a minute of checking. One wrong number on a periodic table and a student memorises the wrong fact; one wrong bar on a sales dashboard and the quarterly meeting goes in the wrong direction. Source grounding lowers the chance of fabrication, but it does not reduce it to zero — turning sources into a page involves choices, transitions, and layout, and each step can drift. The good news is that checking is cheap: your sources are right there in the notebook, so you can verify them item by item. The bad habit is forwarding immediately to save time; the good habit is clicking through first, then sharing.
Four things are worth watching next. First, when the mobile app actually arrives — "coming soon" in a Google announcement can mean weeks or much longer. Second, whether a site follows its sources when they change, or has to be regenerated each time — that decides whether it works for long-running projects. Third, what access controls a shared page has: who can view it, whether it can be revoked, and whether search engines can index it. Fourth, whether Workspace admins get controls — once schools and companies adopt it at scale, permissions and compliance questions will follow immediately. The answers to these four will decide whether this stays a "nice demo" or becomes part of everyday workflows.
The "share" action itself is worth thinking about. Sharing research used to mean sending a fifty-page document nobody would finish, or a screenshot of a chat transcript nobody could interact with. Now you send a link, and the other person opens it, filters the data, takes the quiz, and walks the timeline — information shifts from "pushed at you" to "explored by you." For teams, that changes how review works: it is no longer "you present, I listen" but "we open it together and look." The precondition is that permissions and expiry are clear — Google has not explained them yet, so before sharing, open the link in a second account or a colleague's browser and confirm the other side sees what you intend them to see.
Our take: pages are cheap now; sources are not
The real significance of this update is not how good Gemini Notebook is at making websites, but that it has pulled "making a website" into a research tool's workflow. The old flow was research, organise, then find a tool to build a page; now, halfway through the research, you already have a page you can share. For ordinary users that is convenience; for people who build products it is a signal — when the cost of generating a page falls to a single sentence, competition returns to the content itself: how good your sources are, and how sound your judgment is.
A concrete vibe-coding example makes it clear. A developer with a notebook of user interviews — twenty transcripts, a few screenshots, a list of needs — used to spend half a day turning that into a findings page for the team: writing, drawing, building. Now one sentence produces an interactive page the team can open, filter feedback, and check against the original interviews. But a developer with no interviews, no data, and only an idea gets a page that looks polished and says nothing — no prompt, however well written, can conjure research that does not exist. The tool has paved the road from sources to page, but the stretch from zero to sources still has to be walked by a person.
One more thing is worth noting: NotebookLM's direction has stayed remarkably restrained. It has not rushed to become a chatbot, nor a website builder; it has kept circling "your own sources": first summaries and Q&A, then audio overviews, now generated websites. Each step hands the same material back in a new form. The advantage is trust — everything on the page has a provenance. The disadvantage is limited imagination — it can never build what no source supports. For vibe coders, that is a useful reminder: the hard part was never building the thing, but having something worth building. Sources, judgment, taste — those are the parts one sentence from an AI cannot replace. Tools will keep multiplying and getting cheaper; sources and judgment will not.
Comments (0)
Sources
Related articles
On 19 November 2026, PostgreSQL 19 shipped — two months late, after a flood of AI-reported vulnerabilities rewrote the schedule. It brings REPACK (CONCURRENTLY), pg_plan_advice, and WAIT FOR LSN: tables can be reorganized while they run, query plans can be pinned, and standbys wait for your writes.
On 17 November 2026, Microsoft launched MAI Code 2 and Code 2 Flash: the larger model has 77 billion active parameters and a one-million-token context, averaging 63.2% across four benchmarks — close to GPT-6.1 Sol and Claude Opus 5.5 — at $0.309 per task. Read the three caveats first: promotional pricing, the lowest score of the three, and no long-horizon work.
On 11 November 2026, AWS open-sourced UAEF: adapters turn traces from LangGraph, LangChain, Strands and Bedrock AgentCore into one format, then 36 metrics across 7 dimensions score them, flagging regressions past 5%. The judge model is hardened too — scores must cite verbatim evidence, and never decide safety.