Does Gemini still cite your site with Google Search off?

What changed when Gemini let users turn off Google Search?

Gemini's web app now has a “Search Google” option. It lets users decide, one prompt at a time, whether the assistant checks the live web before it answers. Until now, Gemini made that decision itself. The option sits in the menu behind the plus icon next to the prompt box, below the upload choices and above image creation. Selecting it adds a check mark in the menu and a small “Search Google” chip to the prompt. Users spotted the change on social media before Google said anything, and it doesn't appear in Gemini's release notes yet. Google also hasn't said whether search is on by default or how widely the option is rolling out, so treat these details as early and likely to change.

If your business depends on being cited, the cost is easy to see. A prompt answered without Search has no web sources to link to. Publishers have already seen citations thin out in Google's AI Mode, and this adds a second way for links to disappear completely. The same applies to a dental clinic, a law firm or an online store. Ask Gemini “which accountants in Manchester specialize in freelancers?” with Search on, and it can find and cite current pages. Ask with Search off, and it can only name firms it already links to that question from training. There's no link, no click and no way for the user to check the claim on your site. If you want to see how another Google AI feature handles this, we've looked at whether Canvas in AI Mode links to its sources.

Our view: this isn't a disaster, but it is a useful wake-up call. The toggle doesn't create a new problem. It exposes an old one. AI assistants have always answered many questions from memory, and the businesses that appear in those answers are the ones the model has seen again and again across the web. What's new is that users can now pick the memory-only route on purpose, often for quick facts or advice where live results don't seem necessary. So organic visibility now has two jobs: Gemini has to be able to find you when it searches, and it has to know you when it doesn't. The rest of this article covers both jobs and how to measure each one.

Where do Gemini's answers come from when Search is off?

With Google Search off, Gemini answers from its training data: what it learned from a huge body of text up to its knowledge cutoff. Nothing is fetched in real time. It doesn't look anything up in the search index, doesn't list sources and doesn't link to your page. It may still mention your brand, but it's working from what it took in during training, not from what you published this morning. That has real consequences. Prices, opening hours, staff, new services and recent awards can all be out of date or missing. For a fast-moving software product, Gemini might describe the features you had a year ago. For a local clinic, it might not know a second location exists, and the user has no citation to click to find out.

So what does the model actually “know” about a business? Think in terms of entities, not pages. A language model learns that a name belongs to a certain kind of business in a certain place, linked to certain topics and a certain reputation, because many independent sources say so in the same way. A single page on your own site calling you the best employment lawyer in Bristol is a weak signal. Dozens of mentions in legal directories, local news, client reviews, conference listings and industry forums, all describing you the same way, make a strong one. The reverse matters too. When a model has seen little about a brand, it's more likely to leave it out. Worse, it may fill the gaps with confident guesses that sound right and aren't.

Timing is what catches most business owners off guard. Once Google indexes a page you publish today, Gemini can cite it in Search-on answers within days. In Search-off answers it can't help at all until a future model is trained on it. That can take many months, and nobody outside Google controls the schedule. That's why lasting authority matters so much here. Only content that's still accurate and still referenced a year from now builds up in the model's memory. Paid ads have no equivalent. There's no ad slot inside an answer given from memory, and no budget can buy your way into a model's training data. Either the web knows you well, or the answer goes to a business it knows better.

How do you stay visible in Gemini without search grounding?

To stay visible in Gemini when it answers without searching, you need to be a clearly defined, often-mentioned entity. That means the same name, description and specialty repeated across trusted sites, backed by solid content that others quote. Start with consistency, because it's the cheapest fix and the one most often skipped. Write one plain sentence that says who you are, what you do and where you do it. Then use it everywhere you control: your About page, Google Business Profile, LinkedIn, industry directories, professional association listings and partner pages. Keep your business name, address and phone number identical everywhere. Small mismatches, like a trading name on one site and a legal name on another, make it harder for any system to tell that two mentions are the same company.

Next, earn mentions on sites you don't control, because that's where most entity signals come from. Offer expert comment to trade publications, appear on niche podcasts, speak at local events that publish speaker pages, and get listed by suppliers and partners. Encourage detailed reviews on the platforms your customers actually use. What helps is your brand showing up next to the topics you want to be known for, again and again, on sites a model would treat as credible. A family law firm quoted in a regional newspaper's article on divorce costs builds the right link between name and topic. A pile of bought links on unrelated blogs doesn't, and it can hurt your regular Google rankings too. Relevance and repetition matter more than volume.

Finally, publish lasting content that other people use as a reference. Some will disagree, but one excellent page that other sites cite beats twenty thin blog posts written to hit a keyword list. Pick the questions your customers ask most and answer them better than anyone else. Include real details from your own work, and keep the page up to date. Add clear author bios, an honest About page and Organization schema that lists your official profiles. Put a question in each heading and answer it straight away. That's the same format that helps you win People Also Ask spots and AI citations. For a quick read on how ready your current pages are for AI answers, run them through our free GEO score checker.

How do you get cited by Gemini when Search is on?

To get cited by Gemini when Search is on, your pages must be crawlable and indexed by Google, and each passage should answer a single question cleanly. When Gemini searches, it draws on Google Search results. If Googlebot can't reach, render or index a page, it can't be cited, however good it is. Check the basics in Search Console: the indexing reports, your robots.txt rules, stray noindex tags left behind after a redesign, and pages that need JavaScript to show their main text. Canonical tags pointing to the wrong URL are another quiet problem. Titles and descriptions won't make or break a citation. They do affect whether people click when your link appears, which is why it pays to know how titles, descriptions and noindex affect AI citations.

There's a second crawler setting to understand: Google-Extended. It's a robots.txt token, separate from Googlebot. It controls whether Google can use your content to train future Gemini models and to support answers in Gemini apps and Google's developer AI platform. Blocking it doesn't affect your rankings in regular Google Search. Some publishers block it to protect content they sell, and for them that can be a sensible business decision. For most clinics, firms, stores and service businesses, though, blocking Google-Extended works against you. It keeps you out of the training data behind Search-off answers and can also limit how often you're cited when Search is on. In return, you're protecting content that exists to be found. Check your robots.txt this week, because some plugins and templates add the block without telling you.

Once the technical side is clean, shape the content so AI systems can pull passages out easily. Put the answer first: open each section with one or two sentences that settle the question, then explain. Phrase headings the way people ask questions, and add a short FAQ where it fits. Include the concrete details AI systems tend to favor, such as price ranges, timeframes, locations, steps and specific names. Mark up articles, FAQs, products and local business details with structured data so the meaning is clear. Show a visible “last updated” date, and actually update the page when facts change, because for time-sensitive questions Gemini tends to prefer current sources. None of this is unusual. It's the same organic SEO work that wins rankings, with a sharper focus on passages that can be quoted on their own.

How should you track Gemini referrals and brand mentions?

The simplest way to track Gemini is to give it its own channel in GA4 and pair that with regular manual prompt tests in both modes. Clicks from Gemini usually arrive as referrals from gemini.google.com. In GA4, create a custom channel group called something like “AI assistants”. Give it a rule that matches that domain along with chatgpt.com, perplexity.ai, claude.ai and copilot.microsoft.com, and put it above the default Referral channel so the traffic isn't buried. Then compare its engagement and conversions with organic search. Expect undercounting, because some app clicks and copied links arrive with no referrer and land in Direct. Search-off answers send almost no clicks at all, so this report only measures the Search-on half of your Gemini visibility.

For the Search-off half, you need a set of test prompts. List 10 to 20 questions your customers really ask before buying, the ones that should lead to you if the answer is fair. Once a month, run each one in Gemini twice, once with Search Google on and once with it off. Record three things each time: whether you're mentioned, whether you're linked and whether what's said about you is accurate. Note which competitors show up instead. Results vary between sessions and accounts, so look for trends over several months rather than reacting to a single run. If Gemini keeps getting a fact wrong, such as an old address or a service you dropped, you know which mentions across the web need fixing. The same test works in other assistants. If Bing matters in your market, here's how to get cited in Bing's AI Mode.

Don't ignore the indirect signals. When Gemini mentions you from memory, the result is often demand, not a click. Someone Gemini tells that your firm handles cross-border inheritance may simply search for your name a few days later. Watch branded search impressions in Search Console and your direct traffic, and add a “How did you hear about us?” field with AI assistants as an option to your contact forms. Over a year, these numbers show whether your organic authority is growing. They also make the clearest case for this work over pay-per-click, which stops the day the budget does. Our comparison of how SEO and Google Ads stack up on cost per lead puts numbers on that trade-off. Artiql runs this kind of organic program on autopilot. If you'd like a plan built for both Gemini modes, book a short demo and we'll tailor one, and your pricing, to your business.

Frequently asked questions