Snapchat: multimodal Gemini in the My AI chatbot — over 2.5x engagement growth in the U.S.

• Snap Partners with Google Cloud to Power Multi-Modal Generative AI Experiences Within My AI — Snap Inc. (newsroom), 2024-09-24
• Gemini at Work: how customers use Google Cloud AI products (Thomas Kurian) — Google (blog), 2024-09-24
• Snap Inc. Partners with Google Cloud to Power Multi-Modal Generative AI Experiences Within its My AI Chatbot — Google Cloud Press Corner, 2024-09-24
• Snapchat taps Google's Gemini to power its chatbot's generative AI features — TechCrunch, 2024-09-24

Background

Snap Inc. is the maker of Snapchat, one of the world's largest messaging apps: per the Google Cloud press release, the app reaches more than 850 million monthly active users, predominantly a young audience. My AI is Snapchat's built-in chatbot and, in Snap's own wording, "one of the largest consumer chatbots available today." It launched in February 2023 — and, importantly for understanding this case, was initially powered by OpenAI's ChatGPT.

Snap's relationship with Google Cloud is far older than generative AI: Snapchat first launched on Google Cloud infrastructure back in 2011, making the partnership more than a decade old. On September 24, 2024, the companies announced its expansion: Gemini's multimodal capabilities on Vertex AI became the foundation of the next generation of My AI. The announcement came in sync with Google Cloud CEO Thomas Kurian's "Gemini at Work" blog post, timed to Google's global event of the same name. In the post, Kurian sorts nearly 50 customer examples into six types of AI agents — customer, employee, data, security, and creative — and Snap features in the "customer agents" section, with a video demo in the event program. In other words, for Google this was not a routine integration but one of its headline public examples of consumer Gemini.

What makes the case more interesting than most "we plugged an LLM into a chatbot" stories: the camera is Snapchat's primary interface. Users of this app communicate in snaps and videos, not paragraphs of text. So multimodality here is not a feature from a marketing checklist but a natural extension of the product: a bot that understands snaps plugs into the audience's existing behavior instead of demanding new behavior.

The model-switching history is a story of its own. In a year and a half, My AI went from ChatGPT (February 2023 launch) to Gemini (September 2024) — a vivid illustration of how large B2C platforms treat model selection: not "marrying a vendor forever" but "picking the model for the specific capability." For Google, this became a showcase case: it demonstrates Gemini working at consumer scale, with interactions counted in millions per day.

Problem

A text-only chatbot inside a camera-first app is a half-measure. Snapchat users express themselves in snaps and videos; forcing them to type long text queries means tearing the bot out of the app's native mode of communication. For My AI to be useful to this specific audience, it needs to understand images and video as fluently as text: from a photo of a street sign in a foreign language to a video of a snack shelf.

The second part of the problem is scale and economics. My AI is one of the largest consumer chatbots in the world, and Snapchat reaches more than 850 million monthly users. At that scale, every model improvement is multiplied across millions of daily interactions — but so are latency and inference costs. The service needed multimodal responses with low latency and an acceptable cost per query, or the feature's unit economics would not add up.

The third layer is trust and safety. Snapchat's audience is young, and from its launch My AI has been surrounded by public debate about potential risks to users, especially children — TechCrunch noted this in its coverage of the Gemini move, stressing that as the bot's capabilities grow, so do the stakes. For Snap this meant the new generation of My AI had to rest on a model with built-in accuracy and safety mechanisms, not merely the most powerful one available.

Solution

Snap deployed Gemini on Vertex AI inside My AI. The announcement's wording is precise and worth quoting in full: the company leverages "the strong multimodal capabilities of Gemini on Vertex AI, particularly the technology's ability to understand and operate across different types of information like text, audio, image, video, and code, to offer more engaging and innovative features for our Snapchat community through My AI."

In practice this looks like scenarios embedded into the familiar "shoot and send" gesture. A user can photograph a street sign while traveling and ask My AI to translate it. They can take a video of different snack offerings and ask which one is the healthiest option. What used to require a separate translator app or a search session now happens inside a conversation with the bot — with the same motion a user sends a snap to a friend. The multimodal capabilities were rolled out to all U.S. Snapchatters.

Choosing Vertex AI as the platform continues an infrastructure story: Snapchat has run on Google Cloud since 2011, so this was the path of least resistance for integration, operations, and scaling. The Google Cloud press release separately stresses that Gemini models provide "built-in accuracy and safety features" and serve millions of multimodal interactions daily — for a product with a young audience and a history of public questions around My AI, that was a meaningful selection argument.

Equally important is what Snap did not do: the company did not rewrite My AI "all-Google" and did not abandon its multi-vendor strategy. My AI launched on ChatGPT in February 2023; since September 2024 the multimodal scenarios run on Gemini — Snap has historically combined models from different providers across parts of the product. The case thus records not a "defection to Google's camp" but a targeted decision: for multimodality at consumer scale, they picked the model that at that moment best combined capability, latency, cost, and safety.

On Google's side, the case sits in a broader frame: in the "Gemini at Work" blog post, Thomas Kurian classifies My AI as an example of a "customer agent" — an agent that helps make recommendations and answer customer questions. Snap CEO Evan Spiegel, per TechCrunch, described the partnership as reinforcing "everything that's so important to serving our community": users can "learn so much more about the world, do it really quickly in the moment." In the Google Cloud press release, Spiegel adds the mission frame: Snap empowers people to express themselves, live in the moment, learn about the world, and have fun together — and in that logic a multimodal assistant is not a separate product but an amplifier of each of the four. Kurian, for his part, calls Snap "an early leader in digital communication" now "at the forefront of using generative AI to build agents creating new value for its community."

Result

Per Snap and Google Cloud, after deploying Gemini in My AI, engagement within Snapping to My AI grew by over 2.5x in the United States. Kurian's wording in the Google blog: "Snap deployed the multimodal capability of Gemini within their 'My AI' chatbot and has since seen over 2.5x as much engagement within Snapping to My AI in the United States."

Framing that matters. This is the case's only published metric. It was measured in a narrow window of August 27 – September 2, 2024 — stated directly in the footnote of Snap's announcement, marked "Snap Inc. internal data." "Snapping to My AI" means sending snaps to the bot, so the metric tracks growth of multimodal usage specifically, not all My AI activity. Absolute audience figures for the bot were not disclosed: "2.5x" of an unknown base could mean explosive growth or modest absolute numbers. No impact on revenue, retention, or time-in-app has been published.

Editorial analysis. First: the size of the effect here is less interesting than its mechanics. The 2.5x+ growth happened not because the model got "smarter at answering in text," but because the bot learned to accept input in the format Snapchat's audience already communicates in — snaps. This is an unusually clean confirmation of the principle that modality must match the product's interface: the same LLM functionality packaged into a text box would almost certainly not have produced such a jump. Second: a one-week measurement window right after launch is the classic novelty spike; the steady-state engagement level could have turned out higher or notably lower, but the companies never published it — in late 2024 or since. Silence after a loud announcement is itself information for anyone evaluating the case.

Third: strategically, the case illustrates where platforms of Snap's scale stand in the model market — they migrate between vendors (ChatGPT in 2023, Gemini in 2024) faster than the enterprise segment, because their integrations are thinner and their motivation is a specific capability rather than ecosystem loyalty. For model providers this means consumer flagships are a winnable — and losable — segment; for product teams, that architecting for model replaceability pays off at every such transition.

Finally, note the disclosure asymmetry: the announcement includes an engagement figure but says nothing about inference cost, latency, or the impact of multimodal queries (noticeably more expensive than text) on the feature's economics — even though at 850M+ monthly users those are the parameters that decide whether the feature survives. The public part of the case answers "did it take off" but not "does it add up." For a reader mapping this case onto their own product, that means the scenarios and mechanics are transferable, but the financial model will have to be built from scratch on your own data.

Lessons learned

1. Modality must match the product's interface: in a camera-first app it was multimodality, not better text answers, that unlocked usage growth.
2. Read the footnotes: the headline 2.5x was measured in a one-week window right after launch — an early spike signal, not a proven steady state; the absence of follow-up metric publications is a signal too.
3. Relative metrics without absolute baselines (how many users, how many sessions) are standard in vendor case studies; demand the base before making decisions for your own product.
4. A consumer AI assistant grows on concrete scenarios (translate a sign, compare snacks), not on an abstract 'chat with a bot' — scenarios must fit the user's existing gesture.
5. Large B2C platforms pick models per capability, not per brand: My AI went from ChatGPT to Gemini in eighteen months — architect for model replaceability.
6. Infrastructure history carries weight: 10+ years of Snapchat on Google Cloud made Vertex AI the path of least resistance — all else equal, the vendor already in production wins.
7. For products with young audiences, a model's built-in safety mechanisms are a selection criterion on par with capability: the public context around My AI left Snap no room for a 'raw' solution.

Frequently asked questions