Salesforce Patents a Way to Share AI Prompts Across Team Channels
When someone on your team figures out the perfect AI prompt, right now it lives in their head or a random notes doc. Salesforce is patenting a way to bottle that prompt and pass it around like a shared recipe.
What Salesforce's shared-prompt system actually does
You're in a team chat channel and a coworker has written a really effective instruction for the AI assistant, something like a polished prompt that summarizes customer feedback in a specific format. Today, sharing that means copy-pasting it into a message and hoping everyone keeps track of it.
Salesforce's patent covers a system where that prompt becomes a shareable object inside the platform. One person creates it, posts it to a channel (or multiple channels), and other people can click to run it themselves. When someone else uses it, the AI processes the prompt together with their own account data, so the result is personalized to them.
The output then shows up right in the channel, visible to whoever ran it. It keeps the useful AI work inside the collaboration tool instead of scattering it across personal documents or email threads.
receiving, from a first user profile in a first virtual space of a group-based communication platform, a first request to share a prompt, wherein the prompt includes a instruction for a generative machine-learned model; …
Translation: A user asks the chat app to share an AI instruction.
How the prompt travels from one workspace to another
The patent describes a workflow built on top of a group-based communication platform (think Slack or Salesforce's own Slack-adjacent products). Here's the sequence:
- A user in one channel (the patent calls it a "virtual space") asks the platform to package a prompt as a shareable message.
- That message can be posted to a different channel, crossing workspace boundaries.
- A second user in that channel sees the message and requests to run the prompt.
- The platform feeds that user's own profile data plus the original prompt text into a generative machine-learned model (an AI like a large language model).
- The AI's response is displayed back in the channel, visible through that second user's interface.
The critical design choice is that the prompt and the user's data are combined at runtime. That means the same prompt can produce different, personalized outputs for each person who uses it, rather than just replaying a static result.
The patent also covers the cross-channel posting step specifically, where a user in one space can authorize the message to appear in a second, separate space. That's the infrastructure for making a prompt discoverable beyond the team that originally wrote it.
The communication platform may receive a request from a second user profile to access and/or utilize the sharable prompt.
Translation: Another user decides to use the shared prompt.
What this means for AI tools inside Slack and Salesforce
Salesforce's run of AI-in-collaboration filings suggests the company sees the chat channel itself as the natural place for AI outputs to live, not a separate tool or tab. If this system ships in something like Slack, it changes how teams actually reuse AI work: instead of someone being the informal keeper of good prompts, the platform becomes the library.
For everyday users, that could mean less time reinventing the wheel every time you need AI help on a routine task. For Salesforce as a business, it also deepens how embedded AI becomes in day-to-day team workflows, which makes the platform stickier.
Salesforce's 13th filing we've tracked in Language AI since May follows a faster database search and a model-ranking dashboard.
Claim 1 is broader than it might first appear. It covers any system that receives a prompt-share request, packages it into a message, posts that message to a second virtual space on a user's request, and then pipes a third user's profile data plus the prompt into a machine-learned model. That sequence is not tied to any specific AI model, any specific data type, or even any specific platform.
In practice, that breadth could put a wide perimeter around something that feels like a very ordinary workflow. Forwarding a message and then having an AI run it with the recipient's data is not a complicated idea. The question any patent examiner would weigh is whether bundling those steps into a single claim covers something that wasn't already obvious to engineers building collaboration-plus-AI features.
As a reader, the honest takeaway is that this filing is less about a technical invention and more about staking out territory in a space where every major collaboration platform is building similar features. The claim is clear and well-constructed, but its value will depend entirely on what prior art examiners find.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings
11 drawing sheets from US 2026/0260077 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →