Quick answer: Google now lets Gemini in Google Docs ground its writing in a Gemini Notebook (formerly NotebookLM), with inline citations back to your sources. Grounding narrows what the AI draws from, but the output is still only as reliable as the sources in the notebook. The same principle shapes how ChatGPT, Gemini, and Perplexity describe your firm: they answer from whatever sources they can find about you.

It's 11:40 p.m. An associate has fourteen tabs open: a deposition transcript, two precedent contracts, the client's playbook, and a memo due on a partner's desk by 8 a.m. Every paragraph means another trip to another tab, another copy and paste, and another chance to lose track of where a sentence came from.

What changed in Google Docs?

Gemini Notebook can now serve as a context source in Google Docs, so Gemini grounds your drafts in a curated set of sources without switching tabs or copying and pasting between tools. Type "@" in the Gemini side panel or bottom bar in Docs, select a notebook, and Gemini adds inline citations so you can check facts without leaving the document. The details are in Google's announcement on the Workspace Updates blog.

If Gemini Notebook sounds unfamiliar, you probably know it by its old name. Google began renaming NotebookLM to Gemini Notebook on July 16, 2026.

Most coverage will stop at "this saves time." For law firms, the more useful lesson is what the feature can't do.

Why does grounding matter?

From the outside, the hesitation around AI in legal work doesn't seem to be about speed. It seems to be about accuracy. A model working from general training data can produce language that sounds right and a citation that points nowhere.

Grounding changes the question the model answers. Instead of drawing on everything it was trained on, the model draws on the documents you hand it. That's a smaller, more checkable question.

Does grounding make AI output reliable?

Not on its own. Grounding moves the risk. It doesn't remove it.

I'm not a lawyer, and how a firm manages its matter files is a call for its attorneys. What I can speak to is how grounded AI behaves, because the same mechanic sits at the center of my work.

Picture two notebooks built on the same topic. The first holds only current, final documents. The second holds those same documents plus an outdated version, an abandoned draft, and a file from the other side.

Ask both the same question and you may get two different answers. Both will arrive with tidy inline citations. In the second case, those citations only prove the answer came from the notebook. They don't prove it came from the right document.

Citations make checking faster. They don't make it optional. A clean source library produces output you can trust. A cluttered one produces confident output you can't.

How does this apply to your firm's AI visibility?

Here's where this stops being a Google Docs tip.

Say a general counsel asks ChatGPT, Perplexity, or Google's AI Overviews which firm handles AI contract disputes in Dallas. With web search on, those tools do a version of the same thing: they pull from sources they can find and trust, then assemble an answer. Either your firm is in that source pile or it isn't. If it is, the facts the tool finds are either accurate or they aren't.

When we run AI visibility audits for firms, one of the first things we do is ask AI tools basic questions about the firm. When was it founded? What does it specialize in? Who are the partners? We ask each question twice, once with web search off and once with it on, because the answers can differ. Then we look at where the answers came from.

What we're really auditing is the firm's public notebook: directory profiles, bar listings, attorney bios, the Google Business Profile, published articles, and press releases. Suppose the partner bio says one thing, an old directory listing says another, and the website offers nothing structured for a machine to read. Then the AI is working from a messy notebook. It can still produce a confident answer. That answer might simply be wrong, or it might name a competitor whose sources were cleaner.

What should a firm do this quarter?

How your team uses this feature for legal work is a decision for your attorneys, your IT team, and your professional responsibility policies.  If you're working on an AI adoption plan more broadly, that's something we help businesses with too. The part I can help with is the external one: what the web says about your firm.

Apply the same discipline there:

• Make your firm's name, address, and phone number identical everywhere they appear.
• Give attorney bios structure: credentials, bar admissions, publications, and recognitions, stated plainly.
• Add schema markup so machines can read what people already see on the page.
• Keep your Google Business Profile Q&A current, because AI tools repeat stale answers.

None of this is glamorous. It's the same work as cleaning up a source library, just done in public.

Frequently asked questions

What is Gemini Notebook?
Gemini Notebook is Google's research and source-grounding tool, formerly called NotebookLM. Google began renaming it on July 16, 2026.

How do you use a Gemini Notebook in Google Docs?
In Google Docs, type "@" in the Gemini side panel or bottom bar and select a notebook. Gemini then grounds its writing in that notebook's sources and adds inline citations you can check.

Does grounding stop AI from making mistakes?
No. Grounding limits the AI to the sources you provide, but the output reflects the quality of those sources. Outdated or incorrect documents in a notebook can still produce confident, cited, wrong answers, so every citation still needs checking.

How can a firm find out what AI tools say about it?
Ask ChatGPT, Gemini, and Perplexity basic factual questions about the firm, such as when it was founded and what it specializes in. Ask each question with web search off and on, and note which sources the answers come from. An AI visibility audit does this systematically across many questions and platforms.

The notebook always wins

Grounded AI looks like a real step forward for anyone drafting from a pile of source documents. The inline citations alone could save hours of back-and-forth between the people writing and the people reviewing.

The lesson underneath the feature is older than the technology: output reflects input. A draft reflects the notebook behind it. What AI tells a prospective client about your firm reflects the sources you've left for it to find.

If you'd like to know what ChatGPT, Gemini, and Perplexity currently say about your firm, and which sources they're drawing from, we'll run an AI visibility check and walk you through what we find.