Google Just Renamed a Bot That Ignores Your Robots.txt. Your Dealership Site Has Until August.

On July 16, Google renamed one of the fastest-growing products it has ever shipped. NotebookLM — the research tool where you dump in documents and web pages and then interrogate them — is now Gemini Notebook. Josh Woodward, the VP who runs Google Labs and the Gemini app, wrote that it "remains a standalone product focused on being your premier research tool, but it will now do more across the Google ecosystem." Google says more than 30 million people and over 600,000 organizations use it, and that notebooks will soon be brought "directly into AI Mode in Search."

That's the press-release version, and on its own it's a branding story you could safely ignore.

Here's the part that actually touches your website. Buried in Google's crawler documentation, the tool's fetching agent got renamed too — from Google-NotebookLM to Google-GeminiNotebook. Google's docs list the old name as "supported until August 2026." Search Engine Journal flagged the consequence on July 18: anyone who wrote a rule against the old string has a few weeks before that rule quietly stops matching anything at all.

And there's a second detail in that documentation that matters more than the rename. This fetcher never obeyed your robots.txt in the first place.

The category nobody at your store has heard of

Google sorts the things that request pages from your server into buckets. There are common crawlers — Googlebot and friends — that read robots.txt and follow it. There are special-case crawlers. And then there's a third bucket called user-triggered fetchers, where Gemini Notebook lives alongside Google Read Aloud, Google Site Verifier, and Feedfetcher.

Google's own description of the bucket is a single sentence, and it's the whole ballgame: "Because the fetch was requested by a user, these fetchers generally ignore robots.txt rules."

The logic isn't sinister. A crawler goes hunting across the open web on Google's initiative — so you get a vote, and robots.txt is that vote. A user-triggered fetch happens because a specific human pasted a specific URL and pressed a button. Google's documentation describes the Gemini Notebook fetcher as requesting "individual URLs that Gemini Notebook users have provided as sources for their projects." Functionally, that's closer to a person opening your page than to a bot indexing it.

So if your robots.txt has a tidy list of AI agents you've decided to disallow, understand what that list does and doesn't cover. It does not cover this. It never did.

Two mistakes, pointing in opposite directions

The first mistake is the boring one, and a real number of dealer sites are sitting on it right now. If you — or more likely your website vendor, or whoever configured your firewall or CDN two years ago when "block the AI bots" was the reflex — wrote a rule targeting Google-NotebookLM, that rule expires by itself in August. Nobody will email you. The dashboard will still show a rule. It will simply stop matching, because the string it's looking for no longer exists.

That's a hardcoded-value trap, and it's the kind of thing that sits undetected for a year. If you genuinely intend to block this fetch, you now have to block two strings, and you have to know that today.

The second mistake is the more expensive one, and it's the reason I'd think hard before writing either rule.

We've been here before. In 1994, Yahoo built a human-curated directory on the premise that people should approve what gets into the index. It worked beautifully until the web outgrew it, and Google's crawler — which asked nobody's permission — ate the market. Every few years our industry rediscovers the same instinct: something new is reading my content, therefore I should stop it. Dealers spent 2024 and 2025 adding GPTBot and ClaudeBot to robots.txt, and a meaningful share of them ended up invisible to the exact tools their shoppers had started using to build a short list.

Now apply that instinct to this specific fetch and look at what you'd actually be blocking.

Somebody sat down, opened a research tool, and typed in your URL. Not a category page. Not a competitor's. Yours. That's not a scraper strip-mining the web for training data — that's the single highest-intent signal a website gets. Someone is doing homework on your dealership, and they've asked Google to go read your page so they can ask questions about it.

Block that, and the tool comes back empty. The shopper doesn't see a firewall message and think what a principled dealership. They add another rooftop's URL and keep going.

What this means for your dealership

Think about how a car actually gets bought now. A shopper is deciding between three stores. They're comparing a CPO listing at your place against two others, trying to figure out what's included, what the fees look like, whether the warranty is real. Increasingly, that comparison doesn't happen in fifteen browser tabs — it happens inside a tool where they paste the three URLs and ask, which of these is the better deal, and what am I missing?

Whichever page can't be read loses by default. Not because it lost the comparison — because it wasn't in it.

This shift changes what "your website" has to be good at. For a decade the job was ranking: get in front of the shopper. That job still exists. But there's a second job now, and it's being legible — your pricing, your fees, your warranty terms, your hours, your inventory details all sitting in clean text that a machine can pull and summarize accurately when a shopper hands it your URL.

Two things follow from that, and they're both uncomfortable for the average dealer site.

The first is that vagueness now costs you directly. A VDP where the price is an image, the fees live behind a "click for details" toggle, and the disclaimer is a wall of gray 9-point type used to be merely annoying. Now it produces a summary that says the pricing is unclear — sitting right next to a competitor whose page summarized cleanly. The machine doesn't editorialize. It just reports what it could and couldn't find, and "couldn't find" reads as evasive.

The second is that this is heading into Search. Google says notebooks are coming directly into AI Mode. The behavior that today lives in a separate research app used by 30 million people is being folded into the surface where car shopping actually starts. A fetch pattern you can currently dismiss as niche is on its way to the front door.

What to do about it

• Audit your firewall, CDN, and WAF for both strings — this week. Search your rules for Google-NotebookLM and Google-GeminiNotebook. You need to know which state you're in, because after August the old rule is decorative. On most dealer sites this config lives with the website vendor, not with you — so this is a ticket you open, not a setting you find.
• Ask your vendor a direct question and get it in writing: what is our default bot policy? Many platforms ship aggressive bot-blocking on by default and never tell the dealer which agents are on the list. You may be blocking things you'd never have chosen to block. "I'll check" is not an answer; ask for the actual list.
• Don't add a blanket block on user-triggered fetches. Understand what you'd be turning away: a person who specifically asked for your page. If you have a real reason — a bandwidth problem, a content-licensing position — that's a business decision, make it deliberately. Just don't inherit it from a vendor default or a two-year-old panic.
• Make one VDP survive the paste test. Take a live vehicle URL, drop it into a research tool, and ask it: what's the out-the-door price, what fees apply, what's the warranty, what's included? Read the answer as a shopper would. If it's thin, wrong, or hedging, that's not the AI failing — that's your page.
• Move the money facts into real text. Price, fees, doc fee, warranty terms, and CPO inclusions belong in crawlable HTML, not baked into images, not hidden behind interactions, not stranded in a PDF. This is the same work that makes a page accessible and the same work that makes it rank. It just has a third payoff now.
• Stop treating robots.txt as your AI policy. It governs one category of one kind of visitor. It says nothing about user-triggered fetches, and it's not a strategy. The strategy is deciding what you want said about your store when a machine reads it — and then making sure the page says it.

The bottom line

The rename is a housekeeping item with a real deadline attached: if you blocked the old agent, that block dies in August, and nobody is going to tell you. Go look.

But the reason this story is worth ten minutes of your week isn't the string change. It's that Google documented, in plain language, that a growing class of AI reads your site regardless of what your robots.txt says — because a human asked it to. That's not an attack surface. That's a shopper. The dealers who spend the next year trying to build a taller wall are going to discover they've walled out their own customers, the same way the whitelist crowd did in 2024. The ones who win will spend that year making their pages ruthlessly clear, because clarity is now the thing being graded.

If you don't know what your site says when an AI reads it — or whether your vendor is quietly blocking the tools your shoppers are using — that's worth finding out before August. Run your dealership through our AI compatibility test, and we'll show you exactly what the machines can and can't see.