🚨The timing of all changes discussed in this article remains at the sole discretion of LivePerson and is subject to change.
Introduction
At LivePerson, we are committed to providing you with the most advanced and reliable Generative AI capabilities. To ensure we continue leading the industry in innovation and performance, we are pleased to announce the migration of our AI-powered applications to Google’s latest Gemini models.
Change history
As we make updates to this article, we'll update the change history below. Check here to understand what's changed in this article and when.
Date | Description of change |
|---|
April 1, 2026 | Updated scheduled for products integrating KnowledgeAI (i.e., Conversation Assist and Conversation Builder); updated dates for aiStudio AI agents and added testing guidance for same |
March 16, 2026 | Updated the date for the cutover of system/default prompts for brands in the NA region to be March 29, 2026; this will still happen for brands in EMEA on March 28, 2026. |
March 12, 2026 | All APAC schedules set to "TBD." Removed APAC-specific note in "General migration process." |
March 12, 2026 | Clarified the language regarding models and inferencing in "General migration process." |
March 11, 2026 | APAC: Updated the cautionary guidance in the "General migration process" section to be APAC-specific; check the note with the red "stop" sign. Also, added APAC-specific schedules for all products; please see the product-specific sections in this article.
All regions: Updated the dates for Gemini-optimized prompts for Voice solutions to read "TBD," as these prompts are delayed. Also, added testing guidance for Routing AI agents built in Conversation Builder.
|
March 10, 2026 | Added partial migration schedule for Copilot Rewrite in the APAC region. |
Why LivePerson is migrating to Gemini
This transition is driven by our ongoing partnership with Google, which grants us early access to cutting-edge models. Our extensive benchmarking has shown that Gemini models consistently perform as well as—and in several key use cases, better than—our previous models. By moving to Gemini, we are not only ensuring the long-term stability of our services ahead of upcoming model retirements but also delivering faster, more accurate, and more efficient AI experiences to help you better serve your customers.
➡️ All the rigorous safeguards you rely on today will remain in equal or improved standing with Gemini. Specifically:
- Zero data retention: Your data is processed in-memory and is never stored by the model provider.
- No model training: Google is strictly prohibited from using your data to train their foundation models.
- Robust guardrails: All of our existing Generative AI security guardrails are fully supported to ensure sensitive data is protected.
For a comprehensive overview of our security architecture and data handling policies, we invite you to review our updated Generative AI Security & Privacy Whitepaper.
Impacted LivePerson products
- Automated Conversation Summaries
- Copilot Rewrite
- Copilot Translate
- KnowledgeAI integrations in Conversation Assist
- KnowledgeAI agents built in Conversation Builder
- Routing AI agents built in Conversation Builder
- AI agents built in aiStudio
Product-to-model mapping
Use case | Provider | Subscription | LLM |
|---|
Automated Conversation Summaries | google | summarization | gemini-2.5-flash |
Copilot Rewrite | google | rewrite | Gemini-2.5-flash-lite; TBD for APAC |
Copilot Translate* | google | TBD | gemini-2.5-flash-lite; TBD for APAC |
KnowledgeAI integrations in Conversation Assist | google | kai | gemini-2.5-flash-lite; TBD for APAC |
KnowledgeAI agents built in Conversation Builder | google | kai | gemini-2.5-flash-lite; TBD for APAC |
Routing AI agents built in Conversation Builder | google | TBD | gemini-2.5-flash |
AI agents built in aiStudio | google | TBD | gemini-2.5-flash |
*Copilot Translate already uses a Google Gemini model, and self-service configuration of prompts by brands isn't yet available. So, there is no call to action for you if you're using this product. However, as a part of this effort, we will be upgrading the model in use from Gemini 2.0 Flash.
There will be no changes to the Max Tokens used by the model.
General migration process
Step 1 - New system prompts are available; preview is available: LivePerson releases new system prompts that are optimized for Gemini models. Depending on the application in question, access these prompts in the dropdown list on the application's configuration page or via the Prompt Library. Starting on this date, you can preview the Gemini-optimized system prompts if you choose.
Step 2 - Optimized custom prompts are available; start testing. By the date for Step 2, you can find within the UI a version of your custom prompt that has been specially engineered by our data scientists for use with Gemini. Here too, depending on the application, access the prompt in the dropdown list on the application's configuration page or via the Prompt Library. The prompt isn't active. But we've taken the step to add the custom prompt to your account for you, so you don't have to perform this manual work. On this date, begin to test your custom prompt.
Step 3 - Cutover of system prompts: LivePerson automatically updates all system prompts (the default one and any other) to become ones that are optimized for Gemini. If your solution uses a system prompt (rather than a custom one), in effect, you are migrated to Gemini automatically at this time.
➡️ For some products, Step 3 happens before Step 2; see the product-specific guidance later in this article. It includes each product's schedule.
Step 4 - Cutover of custom prompts: On the date for Step 4, if you haven't applied your optimized custom prompts to your Production solution, here's what happens: In all of your custom prompts in Production, LivePerson automatically updates the Provider, Subscription, and LLM settings, so they point to the preferred Gemini model for the given product (see the product-to-model mapping above). But the rest of the prompt, including the prompt instructions, remains unchanged. In effect, this is a forced cutover to Gemini for all custom prompts. Importantly, it means that while the prompt in use in Production uses a Gemini model, the prompt might not be well-suited for Gemini. And this can negatively affect performance.
⚠️ Avoid this automatic cutover by testing the optimized custom prompts that we added to your account and manually switching to them in Production before this date.
FAQs
Does Google Gemini keep my data in-region?
Yes. Data residency is a priority, and all data processed via Gemini remains within your designated region. For detailed information on our privacy and security protections, please visit the LivePerson Trust Center to explore our Generative AI safety standards.
I have lots of custom prompts in the Prompt Library. Which ones will LivePerson optimize for Gemini?
LivePerson will optimize all custom prompts that were used at least once in Production between January 9, 2026 and February 8, 2026, inclusive. This ensures you receive Gemini-optimized versions of all of your active prompts.
➡️ Only prompts where the usage resulted in 2,000 tokens or more expended were optimized.
If you created a new prompt after February 8, 2026, and you'd like it to be optimized for Gemini, please contact your LivePerson representative.
A single prompt in the Prompt Library can have multiple versions. Which version of my custom prompt will you optimize?
LivePerson will optimize only the latest version of a custom prompt. (See the previous FAQ for additional constraints.)
While there's an editing history for each prompt in the Prompt Library, this functions more as a failsafe. It only allows you to copy what you need from an older version if you ever need it.
When will I receive my Gemini-optimized custom prompt from my LivePerson representative?
You won't be receiving this prompt from your LivePerson representative. LivePerson will add it to your account in the system for your manual selection and use according to your product's schedule. (Product-specific schedules are discussed below.)
Will LivePerson be updating my Alpha account?
LivePerson will not be automatically cutting over any Alpha accounts to Gemini models.
However, once Gemini-optimized system prompts are made available for your product, they can be used in both Alpha and Production. So, please take care to manually migrate your Alpha account to Gemini if you have one. This requires that you manually select and use a Gemini-optimized system prompt.
Tip: The names of all Gemini-optimized system prompts that use a Gemini model are prefixed with the word "Gemini."
How do I apply my Gemini-optimized custom prompt in Production?
To avoid the forced cutover, apply your Gemini-optimized custom prompt in Production as soon as possible after you perform testing as described below.
- For all solutions except AI agents built in aiStudio, there are no special instructions needed to accomplish this. In your Production solution, simply replace the Azure-optimized custom prompt in use with your Gemini-optimized prompt.
- In the case of AI agents built in aiStudio, LivePerson will update the relevant flow for you.
General documentation resources
Prompt engineering with Gemini
We’ve developed a series of best practices for prompt engineering with Gemini to help you maximize the potential of Google's Gemini models.
LivePerson's Prompt Library
If you need help using the Prompt Library, see these articles (and others in the same section):
Get support
If you have questions regarding your specific implementation, please contact your LivePerson account representative.
Product - Automated Conversation Summaries
Schedule for EMEA
- March 4, 2026: New system prompts are available in the system; preview is available.
- March 10, 2026: Optimized custom prompts are available in the system; start testing.
- March 28, 2026: Cutover of system prompts.
- April 11, 2026: Cutover of custom prompts.
Schedule for NA
- March 4, 2026: New system prompts are available in the system; preview is available.
- March 10, 2026: Optimized custom prompts are available in the system; start testing.
- March 29, 2026: Cutover of system prompts.
- April 12, 2026: Cutover of custom prompts.
Schedule for APAC
- TBD: New system prompts are available in the system; preview is available.
- TBD: Optimized custom prompts are available in the system; start testing.
- TBD: Cutover of default prompts and custom prompts.
Name of Gemini-optimized system prompts
- Gemini - Structured style summary - Summarization
- Gemini - Paragraph style summary - Summarization
Testing
Use the Test and tune feature to test the Gemini-optimized system or custom prompt and compare it to your current prompt. Focus the bulk of your testing time here:
If desired, you can also test the end-to-end flow by creating a "test" skill and configuring your solution like so:
Doc resources
Product - Copilot Rewrite
Schedule for EMEA
- March 4, 2026: New system prompts are available in the system; preview is available.
- March 10, 2026: Optimized custom prompts are available in the system; start testing.
- March 28, 2026: Cutover of system prompts.
- April 11, 2026: Cutover of custom prompts.
Schedule for NA
- March 4, 2026: New system prompts are available in the system; preview is available.
- March 10, 2026: Optimized custom prompts are available in the system; start testing.
- March 29, 2026: Cutover of system prompts.
- April 12, 2026: Cutover of custom prompts.
Schedule for APAC
- TBD: New system prompts are available in the system; preview is available.
- TBD: Optimized custom prompts are available in the system; start testing.
- TBD: Cutover of default prompts and custom prompts.
Name of Gemini-optimized system prompt
- Gemini - Copilot:Rewrite - Default
Testing
Use the Test and tune feature to test the Gemini-optimized system or custom prompt and compare it to your current prompt. Focus the bulk of your testing time here:
If desired, you can also test the end-to-end flow by creating a "test" skill and configuring your solution like so:
Doc resources
- See this Copilot Rewrite article for info on how to specify and test prompts.
- See the general documentation resources listed later in this article.
Product - KnowledgeAI integrations in Conversation Assist
Schedule for EMEA
- March 4, 2026: New system prompts for Messaging are available in the system; preview is available.
- March 10, 2026: Optimized custom prompts are available in the system; start testing.
- April 4, 2026: Updated system prompts are available in the system; preview is available.
- April 11, 2026: Cutover of system prompts and custom prompts.
Schedule for NA
- March 4, 2026: New system prompts for Messaging are available in the system; preview is available.
- March 10, 2026: Optimized custom prompts are available in the system; start testing.
- April 5, 2026: Updated system prompts are available in the system; preview is available.
- April 12, 2026: Cutover of system prompts and custom prompts.
Schedule for APAC
- TBD: New system prompts for Messaging are available in the system; preview is available.
- TBD: Optimized custom prompts are available in the system; start testing.
- TBD: Cutover of default prompts and custom prompts.
Name of Gemini-optimized system prompt
- Gemini - Enrichment Factual EN - Conversation Assist (Messaging)
- Gemini - Enrichment Factual EN - Conversation Assist (Voice)
Testing
Use KnowledgeAI's Answer Tester to test the Gemini-optimized system or custom prompt and compare it to your current prompt. The Answer Tester is designed for rapid, iterative development, so focus the bulk of your testing time here:
If desired, you can also test the end-to-end flow by creating a "test" skill and configuring a dedicated knowledge base-level recommendation rule in Conversation Assist; make sure it's used for testing alone.
Doc resources
Product - KnowledgeAI agents built in Conversation Builder
Schedule for EMEA
- March 4, 2026: New system prompts for Messaging are available in the system; preview is available.
- March 10, 2026: Optimized custom prompts are available in the system; start testing.
- April 2, 2026: Updated system prompts are available in the system; preview is available.
- April 11, 2026: Cutover of system prompts and custom prompts.
Schedule for NA
- March 4, 2026: New system prompts for Messaging are available in the system; preview is available.
- March 10, 2026: Optimized custom prompts are available in the system; start testing.
- April 2, 2026: Updated system prompts are available in the system; preview is available.
- April 12, 2026: Cutover of system prompts and custom prompts.
Schedule for APAC
- TBD: New system prompts for Messaging are available in the system; preview is available.
- TBD: Optimized custom prompts are available in the system; start testing.
- TBD: New system prompts for Voice are available in the system; preview is available.
- TBD: Cutover of default prompts and custom prompts.
Name of Gemini-optimized system prompt
- Gemini - Enrichment Factual EN - Messaging bot
- Gemini - Enrichment Factual EN - Voice bot
Testing
Use KnowledgeAI's Answer Tester to test the Gemini-optimized system (or custom) prompt and compare it to your current prompt. The Answer Tester is designed for rapid, iterative development, so focus the bulk of your testing time here.
Need an example? See our testing discussion in the previous section on KnowledgeAI integrations in Conversation Assist. The testing process in the Answer Tester works the same.
If desired, you can also test the end-to-end flow by creating a test bot and deploying it on a "test" skill.
Doc resources
Product - Routing AI agents built in Conversation Builder
Schedule for EMEA
- April 15, 2026: New system prompts and optimized custom prompts are available in the system; you can preview the system prompts if desired, and you can start testing the custom prompts.
- May 15, 2026: Cutover of system prompts and custom prompts.
Schedule for NA
- April 16, 2026: New system prompts and optimized custom prompts are available in the system; you can preview the system prompts if desired, and you can start testing the custom prompts.
- May 16, 2026: Cutover of system prompts and custom prompts.
Schedule for APAC
- TBD: New system prompts and optimized custom prompts are available in the system; you can preview the system prompts if desired, and you can start testing the custom prompts.
- TBD: Cutover of system prompts and custom prompts.
Name of Gemini-optimized system prompt
Testing
Create a test copy of the Routing AI agent by exporting it to JSON and importing the JSON back into Conversation Builder. In the test copy, change the prompt in use to be a Gemini-optimized one. Then test the behavior.
Doc resources
Product - AI agents built in aiStudio
Schedule for EMEA
- April 15, 2026: aiStudio prompts are available to test
- May 15, 2026: Cutover of aiStudio prompts
Schedule for NA
- April 16, 2026: aiStudio prompts are available to test
- May 16, 2026: Cutover of aiStudio prompts
Schedule for APAC
- TBD: aiStudio prompts are available to test
- TBD: Cutover of aiStudio prompts
Naming convention for Gemini flows
These are all prefixed with "Gemini."
Testing
Perform testing within aiStudio (use the flow tester and start a conversation, use simulations to batch test queries, etc.). If you find any issues, contact your LivePerson account representative, so we can engage our Data Science team to examine the issue and collaborate on a solution.