NotebookLM's June Update Makes It Worth a Second Look

NotebookLM has had a busy June, and most UK businesses have not noticed yet.

Google shipped two significant updates to its AI research tool in the first three weeks of June 2026. The larger arrived on 08/06/2026, when the company overhauled the core engine and added a cluster of capabilities that shift NotebookLM from a smart note-taker into something closer to a proper research workstation. The second update landed on 22/06/2026 and added memory retention: the ability to track what you have asked and found across sessions, without touching any other Google service.

What changed on 08/06/2026

Google moved NotebookLM onto Gemini 3.5 and paired it with a new reasoning layer called Antigravity. More concretely, every notebook now has its own secure cloud computer. That computer can write and run code in the background while you interact with the tool in plain English, working from a library of more than 100 built-in software skills covering tasks like statistical analysis, data cleaning, and structured document processing.

Output formats expanded considerably. Notebooks can now produce PDF reports, Excel spreadsheets, PowerPoint decks, CSV files, JSON data, charts, and visualisations, all downloadable from the studio panel. Previously, extracting structured output meant copying text manually. Now you describe the format you need and download the result.

The agentic source discovery is arguably the most useful change for professionals who start a project with a question rather than a stack of documents. You can describe your research in the chat, and NotebookLM will search the web for relevant sources and propose adding them to your notebook. You stay in control: nothing is added without your review. But it removes the friction of having to pre-assemble a document library before you can begin.

In internal evaluations published alongside the update, Google reported that the upgraded system achieved a win rate of over 65% against its prior baseline across core evaluation dimensions, including a 78% win rate on advanced web research tasks.

What changed on 22/06/2026

The more recent update is smaller but addresses something that has held many professionals back from committing to the tool. The new memory feature keeps a record of past interactions within NotebookLM, so you pick up context where you left off rather than re-explaining your project each session.

Critically, this memory is sandboxed from the rest of Google's services. It does not access Gmail, Drive, Calendar, or any other connected account. Users can view, manage, or delete it at any time.

Why the privacy architecture matters for UK firms

For UK businesses, that sandboxed design carries real weight. Many teams have been cautious about feeding client documents or commercially sensitive material into consumer AI tools, partly because of uncertainty about data retention and model training. NotebookLM holds SOC 2 Type II and ISO 27001 certification, and Google's standard Workspace terms specify that data uploaded to notebooks is not used to train its models.

Combined with the memory feature sitting entirely within the platform boundary, this gives professionals more confidence that what they upload stays where they put it.

That does not mean every compliance concern is resolved. Organisations in regulated sectors should review Google's data processing terms before uploading sensitive material, and those subject to strict GDPR obligations may still prefer on-premises tooling. But the privacy posture here is meaningfully better than many informal AI tools that have already crept into workplace use.

Who benefits most

The profile of users likely to get the most from these updates is fairly broad. Legal teams reviewing documents or tracking regulatory changes will find the comparison and sourcing features useful. HR teams synthesising employment law updates, finance professionals parsing HMRC guidance, and consultancies building research dossiers for clients all fit naturally. Researchers dealing with large bodies of academic or technical material, who previously spent hours assembling comparison tables manually, will likely see the clearest immediate return.

The free tier covers a meaningful level of capability. The Pro plan sits at approximately £16 per month.

What Adevious AI makes of it

We have been watching NotebookLM develop for some time. For most of its history it was a useful but narrow tool: you uploaded documents, it helped you query them. The June 2026 updates close several of the gaps that limited broader adoption.

The cloud computer, expanded output formats, and agentic sourcing collectively make it a tool capable of carrying a meaningful portion of a research workflow rather than just assisting at one stage. The memory feature, modest as it sounds, changes the daily experience in a way that matters. Research rarely happens in a single session. Keeping context between sessions without re-loading or re-explaining your project removes the kind of friction that causes people to abandon tools after an initial trial.

If you tried NotebookLM six months ago and found it limited, it is worth another look.

Your takeaway this week

Choose one document-heavy task at your business. A regulation you need to summarise for the team, a set of supplier proposals to compare, or a body of client feedback to organise. Upload the relevant documents to a new NotebookLM notebook, ask for a comparison matrix or summary report, and see whether the output saves you an hour. It takes about two minutes to set up.

If you want to talk through how NotebookLM fits into a broader knowledge workflow at your organisation, the team at Adevious AI would be happy to help. Get in touch at adevious.co.uk.