As things stand, there is no direct route in. Google DeepMind introduced Genie 4 on October 27, 2026 as a research preview. The model generates interactive worlds in real time, but it is not a search engine. Google describes neither an index you can submit pages to nor sources that get linked. A brand only appears there deliberately when someone writes it into the prompt or supplies it as an image. Whether the model shows brands on its own, Google doesn't say. What matters more for brands is the setup behind it: Gemini, Google's language model, plans and monitors the simulation. What such systems know about your brand therefore still depends on text that Gemini is allowed to read.
- Since October 27, 2026, Genie 4 has been open only to selected testers.
- Google names no program through which brands get into the model.
- According to Google, legible text such as a brand name often only appears when it is in the prompt. The model doesn't reproduce real places exactly.
- Gemini directs the scene. Brand work therefore remains work on text: your website has to be readable and answer your audience's questions.
- The Google-Extended switch decides whether Google may use your content for Gemini. Google doesn't mention Genie 4 in that context.
What can Genie 4 do, and where does it stop?
A world model is an AI system that learns how an environment works. According to t3n, this lets it simulate the consequences of actions in advance instead of just generating the next word or image. Genie 4 delivers music, ambient sound and dialogue along with the world. Users move through the scene by keyboard and change it with text prompts: different weather, a new camera angle, an additional person (Google DeepMind).
Genie 4 works with models from the Gemini family. Gemini 3.8 Flash serves as a guiding agent. It observes the scene, breaks goals into individual steps and checks the new frames to see whether each step succeeded. Another Gemini model, Gemini Omni, keeps the world's appearance and behavior realistic, according to DeepMind.
Google also lists the model's limitations. Details and objects can drift over the course of a session. A session lasts a few minutes, and the model isn't built for long ones. Two points matter most for brands: Genie 4 currently can't simulate real places exactly, and clearly legible text often only appears when it is in the description of the world.
Where can a brand appear in Genie 4?
Five routes are conceivable. Only two of them are in your own hands.
| Route | What Google says | Example |
|---|---|---|
| Brand name in the prompt | Legible text often only appears when it is in the description of the world | "A small café with the sign 'Rösterei Nord' above the door": the name is in the prompt, but Google doesn't promise it will render without errors |
| Your own image as a starting point | Genie 4 accepts text and image prompts. In Google's demos, a reference image serves as the first frame of the scene | A photo of your own shop as the starting image. The longer the session runs, the more details can deviate |
| Real place | The model currently can't simulate real places exactly | "Our branch in downtown Hamburg" won't produce a reliable replica |
| From the model's knowledge, unprompted | Google doesn't say whether or which brands the model shows on its own | "A busy shopping street": whether real logos appear there isn't documented |
| As a cited source | Genie generates image and sound. Citations or links don't appear in Google's description | A simulated scene about coffee roasting doesn't point to your website |
The first two routes require someone to use your brand on purpose: you, your agency or creatives. Google's test groups use the model the same way. Filmmakers try out scenes in advance, and game developers build explorable prototypes in seconds (Google DeepMind). That makes Genie 4 first and foremost a tool for your own production, not a channel where you get found.
What changes for AI systems that people ask questions?
The real clue is in Google's outlook. DeepMind describes the direction as "hybrid world models". Language and code models set the structure and the timeline, while Genie generates frame by frame how the world looks, sounds and reacts. A person's request therefore goes first to a language model, which interprets it and plans.
This leads to our own assessment, on which Google itself has announced nothing: if people start asking for a scene and not just an answer, the key position is held by the same model family that already writes answers today. Whoever is known there as a term with clear attributes has the better hand. There is a second strand too. Google sees such worlds as training environments for robots and AI agents. Systems that act are therefore increasingly learning in simulated environments.
What can you do now?
- Check Google-Extended in your robots.txt. Google-Extended isn't a separate crawler but a switch. It controls whether Google may use crawled content to train future Gemini models. It also governs grounding in the Gemini apps and on Vertex AI, meaning search content fed into the answer (Google Search Central). It has no effect on Google Search. If you want Gemini to know your brand, no block may apply. The rule that allows it is below this list.
- Describe the brand in words. Name, products, locations and defining attributes belong on the website as text, not just in the logo or the product photo. Some AI agents read the accessibility tree instead of the image, the structure the browser builds from the HTML. Others work with screenshots (Search Engine Journal, June 2026). Why this helps people too is explained in What is good for humans, the machine understands too.
- Answer your audience's questions. A language model can only pass on what is clearly answered somewhere. Our Answer Check compares every question from scouting with your website. It shows on which page and in which section the answer is, or what is missing from it.
- Add Gemini to your measurement. Regularly ask Gemini a fixed list of 20 to 50 real questions too, and note each time: mentioned, cited or not present at all. Which metrics matter is covered in How do I measure my visibility in AI answers?
- Prepare images for your own simulations. For Project Genie, Google recommends high-quality images with the character in the center of the frame in its prompt guide, which is still written for Genie 3. If you want to see your brand name, write it word for word into the description of the world.
If Google-Extended falls under a blanket block on AI bots, allow it explicitly:
User-agent: Google-Extended
Allow: /According to Google's robots.txt specification, only the group that most specifically matches a crawler applies to it. It ignores all other groups. So once Google-Extended is named in the file, User-agent: * no longer applies to this token. How to separate search, training and fetching cleanly is shown in Which AI crawlers should I allow?
Can I already use Genie 4?
+
Not yet, unless you are one of the selected testers. According to Google DeepMind, Google AI Ultra subscribers in Project Genie and early-access partners of the Gemini API and Google AI Studio are next. Google names no date or list of countries. A waitlist is already open.
Can I submit or book my brand in Genie 4?
+
Google's announcement mentions no such offer. It says nothing about a directory for brands or ad placements in the simulated worlds.
Does blocking Google-Extended also affect Genie 4?
+
Google doesn't say. Its documentation names the Gemini apps and Vertex AI for Google-Extended. Genie 4 builds on capabilities of the Gemini family but doesn't appear in that list.
Can people tell that a Genie scene with my brand is AI-generated?
+
According to Google, all video and audio from Genie 4 will include SynthID. That is an invisible digital watermark that identifies AI-generated content.
