If you keep seeing “Gemini” in the news and want to know whether it changes anything for your website, here is the short answer. Google Gemini is Google’s new family of AI models, and it does not change how Google ranks pages today. But it now powers Bard, it already makes Google’s AI answers in Search faster, and Google says it is coming to more of Search soon.
Google announced Gemini on 6 December 2023. In this article I explain what Gemini is, where you can use it right now, what Google has said about Search, and what I think site owners and marketers should do (and not do) this month.
What Is Google Gemini?
Gemini is a large language model, or more correctly a family of them. A large language model is software trained on huge amounts of data to understand requests and write answers. Gemini was built by Google DeepMind, the AI lab Google formed by joining Google Brain and DeepMind earlier this year.
Google calls this first version Gemini 1.0. It is the model that now sits behind several Google products, starting with Bard and the Pixel 8 Pro phone.
The big idea Google keeps repeating is that Gemini is “natively multimodal”. In plain words, it was trained from the start on different types of information together: text, images, audio, video and code. Older approaches often trained separate models and joined them later. Google says training them together helps Gemini understand and reason across different kinds of input at the same time.
The Three Sizes: Ultra, Pro and Nano
Gemini comes in three sizes, each made for a different job:
- Gemini Ultra is the largest model, made for very complex tasks. It is not available to the public yet.
- Gemini Pro is the middle model, made to handle a wide range of tasks at scale. This is the one now inside Bard.
- Gemini Nano is the smallest model, made to run on the device itself, without sending data to Google’s servers.
This split matters more than it first looks. Google doesn’t have to use one huge, expensive model for everything. It can put a small model on your phone, a mid-size one in a free chatbot and a big one in paid or business products.
What Google Claims About Performance
Google says Gemini Ultra scored 90.0% on MMLU, a well-known test that covers 57 subjects such as maths, history, law and medicine. Google says this makes Ultra the first model to beat human experts on that test.
Google also says Ultra beat the previous best results on 30 of 32 widely used academic benchmarks. These are Google’s own numbers, from Google’s own testing. Benchmarks are useful, but they are not the same as how a model feels in daily use, so I treat them as a starting point, not a verdict.
Where You Can Use Gemini Right Now
Gemini is rolling out in stages. Here is what is live at launch and what is still to come.
Bard Now Runs on Gemini Pro
The most visible change is in Bard, Google’s AI chatbot. Bard now uses a specially tuned version of Gemini Pro. Google calls it the biggest upgrade to Bard since it launched, with better reasoning, planning and understanding.
The upgrade is available in English in more than 170 countries and territories. For now it works with text prompts. Google says support for other types of input and more languages is coming.
If you tried Bard when it first opened and gave up on it, this is a good time to try it again. In my own tests of AI chatbots, I always rerun the same set of prompts after a big model change, because the quality jump can be large.
Pixel 8 Pro Gets Gemini Nano
The Pixel 8 Pro is the first phone built to run Gemini Nano. At launch it powers two features:
- Summarize in the Recorder app, which gives you a summary of recorded conversations and meetings, even without a network connection.
- Smart Reply in Gboard, Google’s keyboard, which suggests replies in messaging apps. It starts with WhatsApp, Line and KakaoTalk, with more apps planned for next year.
Android developers can also build with Gemini Nano through AICore, a system feature in Android 14, starting on the Pixel 8 Pro.
Gemini Pro for Developers and Businesses
From 13 December 2023, developers and business customers can use Gemini Pro through the Gemini API. There are two ways in:
- Google AI Studio, a free, web-based tool for testing prompts and building apps with an API key.
- Vertex AI on Google Cloud, a managed platform with more controls for security, privacy and data.
For most marketers this is not something you will touch directly. But it matters because the SEO and content tools you already pay for may start to offer Gemini as an option next to the models they use today.
What Is Coming Next: Ultra and Bard Advanced
Gemini Ultra is still going through safety checks. Google says this includes red-teaming (where outside experts try to make the model behave badly) and more fine-tuning based on human feedback.
Google plans to give Ultra first to selected customers, developers, partners and safety experts. A wider release to developers and businesses is planned for early next year. Google also plans to launch Bard Advanced early next year, a new experience that starts with Gemini Ultra.
What Google Gemini Means for Search
This is the part most readers of this site care about. Google said three things that matter for search:
- Google is already testing Gemini in the Search Generative Experience (SGE).
- In those tests, Gemini made SGE 40% faster in English in the US, along with quality improvements.
- Gemini is coming to more Google products, including Search, Ads, Chrome and Duet AI, in the coming months.
Notice what Google did not say. It did not announce a ranking update. It did not say Gemini changes how pages are crawled, indexed or ranked in the normal blue links. The announcement is about the AI layer on top of Search, not the core ranking systems.
Why a Faster SGE Matters
SGE is Google’s experiment that shows an AI-written snapshot at the top of some search results. It is still opt-in through Search Labs. If you have not tried it yet, my first look at SGE in Search Labs explains how it works and what the answers look like.
One of the most common complaints about SGE has been speed. You type a query, the normal results load, and then you wait for the AI snapshot to appear. A 40% cut in waiting time removes one of the biggest barriers to Google rolling SGE out more widely.
My honest view: speed was one of the things holding SGE back as a product. If it gets fast enough to feel like normal search, the case for showing AI answers to more people gets stronger. That is the real signal in this announcement for SEO, even if nothing changes in your traffic this week.
Gemini in Search Is Not the Same as Bard
People often mix these up, so here is a simple way to keep them apart:
- Bard is a separate chatbot at its own address. People go there on purpose to chat.
- SGE sits inside Google Search. It answers inside the results page and shows links to web pages next to its answer.
- Gemini is the model underneath. It can power both, plus phones, Chrome, ads and business tools.
For websites, SGE is the one to watch. It is part of the place where your organic traffic comes from today, and it links out to sources. Bard is a different product with a different job.
What About Ads and Chrome?
Google has only said Gemini will come to Ads and Chrome “in the coming months”. There are no details yet on what that looks like. I would not plan budgets or projects around guesses here. When Google shares details, we will know more.
What Gemini Does Not Change for SEO
When a big AI launch happens, I usually see two bad reactions. Some people panic and want to rewrite everything. Others ignore it completely. Neither is right. Here is what stays the same:
- Google’s guidance on AI content. Google’s position, set out in its February 2023 guidance on AI-generated content, is that it rewards helpful content however it is made. Using AI mainly to manipulate rankings is still spam.
- The helpful content system. Google’s guidance after the September 2023 helpful content update still applies. Pages made for people, with real experience and expertise, are what Google says it wants to rank.
- Technical basics. Crawling, indexing, internal links, page speed and structured data work exactly as they did last week.
- Your reporting. There is no new report in Search Console for SGE or Gemini. SGE clicks are not shown separately.
A better AI model makes it easier for anyone to produce generic text. That makes truly useful, first-hand content more valuable, not less. In my experience, the sites that hold up best through these shifts are the ones that answer questions better than a summary could.
How to Prepare Your Site: A Practical Plan
You do not need to do anything urgent because of Gemini. But this is a good moment to get ready for more AI answers in Search. Here is the process I recommend:
- Try SGE for your own key queries. Turn on SGE in Search Labs where it is available to you. Search 10 to 20 queries that bring you the most traffic and note where an AI snapshot appears.
- Check whether your pages are cited. Look at the links shown next to each snapshot. Note who is cited and what those pages do better, such as clearer answers or more specific detail.
- Sort your queries by type. Simple fact questions are the most likely to be answered directly by AI. Comparisons, local needs, buying decisions and “how do I” tasks with many steps usually still need a click.
- Improve the pages that matter most. Answer the main question early and clearly. Add what an AI summary can’t: your own testing, photos of real work, clear prices, opinions with reasons and up-to-date details.
- Set a baseline in Search Console. Save clicks, impressions and click-through rate for your top pages now. If SGE grows, you will be able to compare against a clean “before” period.
- Decide your AI crawler policy. Choose whether you want your content used to improve Bard and Vertex AI models, and set robots.txt to match.
A Note on Google-Extended
Many site owners ask me whether they should block Gemini. The control Google offers is Google-Extended, a robots.txt user agent token (a name you can write rules for) that Google introduced in September 2023. It lets you choose whether your content is used to help improve Bard and Vertex AI generative APIs, including future generations of the models behind them.
Blocking it looks like this:
User-agent: Google-Extended
Disallow: /Keep in mind that Google-Extended is separate from Googlebot. Blocking Google-Extended does not stop Googlebot from crawling your pages for Search. I explain the details and trade-offs in my guide to blocking Bard and Vertex AI training with Google-Extended.
My own view: for most business sites, the bigger traffic question is SGE inside Search, not Bard training. I would make this choice calmly and on purpose, not as a reaction to a headline.
Gemini vs Other AI Models: Does It Matter to You?
Gemini is clearly Google’s answer to OpenAI’s GPT-4, which powers the paid version of ChatGPT. Google’s own benchmark numbers suggest Ultra competes at the top level. But Ultra is not public yet, so nobody outside Google can properly compare it today.
For a marketer, the model race matters less than two practical questions:
- Where do your customers get answers? Google Search is still where most of them start. Changes to Search matter more than chatbot scores.
- Which tool helps your team work better? If you use AI for research, outlines or first drafts, test the new Bard against the tools you already use with your real tasks.
When I compare AI tools for content work, I use the same short list of real tasks each time: summarising a long document, suggesting an outline for a topic I know well, and explaining a technical idea in simple words. Then I check every fact. A model that writes smoothly but gets facts wrong costs more time than it saves.
Common Mistakes to Avoid
- Assuming a ranking update happened. If your traffic moved this week, look at the usual causes first: seasonality, recent site changes and announced updates.
- Publishing more AI text because the models got better. Better writing quality does not fix a lack of original value. Mass-produced pages are still the risk.
- Trusting launch benchmarks as proof. Wait until you can test models on your own work.
- Blocking Googlebot by mistake. If you add AI crawler rules, double-check that you have not blocked Googlebot or your whole site.
- Ignoring SGE because it is “just an experiment”. Google is clearly investing in it. A few hours of testing now is cheap.
Conclusion: Google Gemini Is a Signal, Not a Ranking Change
Google Gemini is a big step for Google’s AI products. It already powers Bard and the Pixel 8 Pro, and it made SGE 40% faster in Google’s US English tests. But it is not a ranking update, and nothing in Google’s guidance on content or technical SEO has changed.
The real takeaway is direction. Google is making its AI answers faster and better, and it plans to bring Gemini to Search, Ads and Chrome soon. Your next step is simple: turn on SGE, search your most valuable queries, and see where your pages stand. If you are new to Bard itself, my first look at Bard for SEOs is a good place to start, and then try it again with Gemini Pro to see the difference.
