Amplitude Release Notes

On a survey’s Responses tab you can add user properties as extra columns. Open Columns, search, and pick any project user properties (DAC-restricted ones are hidden). Those columns show in the table and are included in both immediate and async CSV exports.

WHAT IT IS

Description:

On a survey’s Responses tab you can add user properties as extra columns. Open Columns, search, and pick any project user properties (DAC-restricted ones are hidden). Those columns show in the table and are included in both immediate and async CSV exports.

Why:

People analyzing survey responses usually need context like plan, country, or account alongside the answers. Before this, they had to leave the table and join user properties elsewhere. Now they can read and export responses with those properties in one place.

Where:

Survey → Responses → Columns. Same selection is used for CSV export.

Amplitude now makes it much easier to maintain a shared understanding of your taxonomy by automatically generating descriptions from both Amplitude usage and AI context, while still providing transparency about what’s been verified and what has been generated.

WHAT IT IS

Description: Amplitude now makes it much easier to maintain a shared understanding of your taxonomy by automatically generating descriptions from both Amplitude usage and AI context, while still providing transparency about what’s been verified and what has been generated. Customers still have full control over the metadata, modify any definitions they want, and maintain verified descriptions while still taking advantage of automatic improvements.

Why: Most customers have thousands of events and properties with empty or stale descriptions and no realistic way to backfill them by hand, which blocks them from trusting AI answers about their data. Previously, the only path was documenting the taxonomy manually, one entity at a time, and it drifted from the team's spec as soon as it was written. Amplitude now drafts those descriptions automatically, provides transparency on the source, and flips to a model where you can bring in a lot of context automatically and modify a few things.

How we generate them: There are two primary inputs: First, files uploaded to the project under Settings -> AI Controls (a data dictionary, a tracking plan, a product spec), which we treat as the most authoritative input. Second, the project's own Amplitude data, meaning how the entity is actually used: the charts, dashboards and cohorts it appears in, how popular and recently use those are, the events and properties it co-occurs with, its event and query volume, and for a property, the values it takes. We only suggest what we can trace back to one of those, so an entity with thin usage and no matching file gets no suggestion at all. Because uploaded files outrank anything we can infer on our own, getting a customer to upload their data dictionary is the highest-leverage thing we can do for description quality, and one covering both events and properties beats an event-only or partial reference.

Where it's used: These descriptions don't just sit in Data. They already enrich the context behind Global Agent, new event search in dropdowns, and Data Health Score, so customers benefit immediately with no extra work.

Availability: All paid plans with AI features enabled.

Not supported: Every other property type other than event and user properties, custom events, labeled events, and transformations. Only files scoped to the project are used, not organization-level files.

Available in EUDC: Yes

When: Available now.

Public Docs: https://amplitude.com/docs/data/event-property-descriptions#ai-generated-descriptions

WHAT IT IS

Description:

Users can now delete and restore events, event properties, and user properties through the MCP.

Why:

Taxonomy cleanup was the one governance workflow that couldn't finish inside an MCP conversation. Agents could already read the tracking plan, create events, and update metadata, so users could get all the way to "these 40 events are redundant" — and then had to leave the chat and delete them by hand in the UI. The added delete and restore tools close that gap for faster and easier clean up.

Where:

Available to our internal agents, such as Global Agent, through Data Assistant Agent, as well on our public MCP server

Available in EUDC:

Yes

When:

Now!

RELEASE DETAILS

RELEASE ID

taxonomy-mcp-taxonomy-delete-and-restore

SHIPPED

Sep 12, 2026

AREA

Data

Guides and surveys can now spread their audience evenly over time instead of hitting everyone at once.

WHAT IT IS

Description: Guides and surveys can now spread their audience evenly over time instead of hitting everyone at once. Turn on Even distribution in Targeting and pick a period (30/60/90/180/365 days). Each user gets their own 14-day eligibility window placed at a stable, deterministic point inside that period, so the audience trickles in evenly and each user still only sees the guide once per period.

Why: With recurrence alone, every eligible user becomes eligible at the same moment. Customers running a quarterly NPS survey get a spike of responses in the first few days, nothing for the rest of the quarter, and a sample that's skewed toward whoever happened to be active that week. Spreading the window gives a steady response stream and a more representative sample, without anyone having to hand-manage waves of targeting.

Where: Guide or survey → Setup → Targeting → Even distribution. Requires the same entitlement as recurrence limits.

When: Now!

Documentation: https://amplitude.com/docs/guides-and-surveys/setup-and-target#distribute-eligibility-evenly

Users of all tiers now have the ability to migrate event data among projects and organizations they control, as well as to pull in data from Mixpanel and Posthog.

WHAT IT IS

Description:
Users of all tiers now have the ability to migrate event data among projects and organizations they control, as well as to pull in data from Mixpanel and Posthog.

Where:
A new "Migrations" panel under the Data tab

Available in EUDC:
Yes

When:
Now

Documentation:
https://amplitude.com/docs/data/self-service-data-migrations

RELEASE DETAILS

RELEASE ID

self-service-data-migrations

SHIPPED

Sep 08, 2026

AREA

Data

Guides and Surveys can now select elements inside open shadow DOMs

WHAT IT IS

Description: Guides and Surveys can now select elements inside open shadow DOMs

Why: We already supported targeting elements within a shadow DOM, but now you can just use our click selector like any other element.

Where: Guides and Surveys element recorder and click trigger configuration (web only)

Available in EUDC: Yes

When: Now

Documentation: https://amplitude.com/docs/guides-and-surveys/form-factors#target-elements-in-shadow-dom

RELEASE DETAILS

RELEASE ID

better-shadow-dom-support-for-guides-and-surveys

SHIPPED

Sep 04, 2026

AREA

Guides and Surveys

WHAT IT IS

Description: You can now schedule when an experiment or flag stops, not just when it starts. Set a start time, an end time, or both. At the end time, delivery stops, the analysis window freezes, and the experiment moves to Completed (Pending decision).

What's included:

• Start and end in one modal — check Start time, End time, or both; a Runs 14 days summary confirms the window before you commit
• Auto-stop at the end time — experiments stop delivering variants, flags turn off
• Analysis window freezes at the scheduled stop, so results don't keep extending past the intended end
• Status moves to Completed (Pending decision) and the active-experiment quota slot is released for the next test
• Owners get notified when a scheduled start or stop fires
• Schedule details in the status pill tooltip — see exactly when it starts and stops

Why: Teams had to stop experiments by hand. Miss the window and variants keep shipping past the planned end — stretching analysis and holding an active-experiment slot. Time-boxed tests now end on schedule.

Where: Experiment or flag page → Start Experiment → Schedule. Once running, use Edit schedule.

Available in EUDC: Yes

When: Now!

Docs: https://amplitude.com/docs/feature-experiment/workflow/experiment-test#schedule-your-experiment

@amplitude/session-replay-react-native has moved from beta to GA.

WHAT IT IS

Description: @amplitude/session-replay-react-native has moved from beta to GA.

Details:

• Capture replays from React Native apps on iOS and Android
• Works on both RN architectures (New Architecture / TurboModules and legacy), with no configuration needed
• Privacy built in with mask levels, <AmpMaskView>, and opt-out support
• Fits existing setups: standalone, with the Amplitude Analytics plugin, or alongside Segment

Why: React Native teams can now confidently adopt Session Replay with a stable, production-ready standalone SDK across both platforms and architectures.

Available in EUDC: Yes

When: Available now — latest version is 1.0.1

Resources:

• Docs: https://amplitude.com/docs/sdks/session-replay/session-replay-react-native-standalone-sdk
• NPM: https://www.npmjs.com/package/@amplitude/session-replay-react-native

The Guides and Surveys settings drawer now has a "Default user exclusions" section that lets you add user segments that should never see guides or surveys.

WHAT IT IS

Description:

The Guides and Surveys settings drawer now has a "Default user exclusions" section that lets you add user segments that should never see guides or surveys. This way you don't have to set it for every guide individually.

When:

Now

Documentation:

https://amplitude.com/docs/guides-and-surveys/setup-and-target#exclude-a-group-of-users

The Web Experiment script now supports cookie consent gating. Customers can defer all device storage and third-party requests until a visitor consents — while still running experiments flicker-free while consent is pending.

WHAT IT IS

Description: The Web Experiment script now supports cookie consent gating. Customers can defer all device storage and third-party requests until a visitor consents — while still running experiments flicker-free while consent is pending.

Why: EU customers (and others with CMP banners) need to run web experiments without writing cookies, localStorage, or sessionStorage — or firing impressions to third-party origins — before consent. Previously the only safe option was blocking the script entirely, which meant no experiments and visible flicker when consent arrived. This gives a compliant default that still delivers the test experience.

What’s new:

• consentOptions on window.experimentConfig — opt in with consentRequired: true
• window.webExperiment.setConsentStatus() — wire your CMP callback ('granted' / 'denied')
• Script behavior
  • While pending: experiments evaluate and variants apply with no flicker; cookie/storage writes, impression events, and the RTBT relay iframe stay in memory only
  • On grant: buffered identity, storage, and impressions flush to the device; relay loads
  • On deny: client doesn’t start (or erases prior-session data on revocation); nothing persists from the pending buffer

Available in EUDC: Yes

When: Now!

Docs: amplitude.com/docs/web-experiment/cookies-and-consent-management

Implementation snippet:

<script>
window.experimentConfig = {
  consentOptions: {
    consentRequired: true
  },
};
</script>

<!-- Web Experiment script tag -->
// CMP callback
window.webExperiment?.setConsentStatus('granted'); // or 'denied'

Notes:

• This gates Web Experiment only. Analytics consent is separate — pair with Browser SDK cookies & consent if they use both.
• Works with the anti-flicker snippet and cross-subdomain identity/RTBT.

Observe what users ask your AI agent, evaluate where it fails and what those cost your business. Agent Analytics is now available to all customers.

WHAT IT IS

Description

Traditional analytics tells you a session happened. LLM observability can show you the trace. Neither immediately answers the question you care about: Did the agent give the user a good answer, and did that make them convert or churn?

Amplitude launched our specialized AI agents in 2025 and we faced the same issue. We built Agent Analytics for ourselves to close this gap. Each trace is scored for task success, failure type, and tool-call errors. Every conversation becomes agent behavior events you can query in the same funnels, cohorts, and retention charts you already use.

What it does

• Score every conversation. Inspect agent traces, take advantage of online evaluations such as task success and helpfulness, and filter quickly by behavioral cohorts.
• Rank failing topics. Conversations are clustered by intent so you can see which topics fail most and fix the expensive ones first.
• Connect agents to outcomes. View journeys and session replays inclusive of agent usage, compare different model providers on conversion, and find failure modes that cost you users.

Why

Most teams verify agents with offline evals, ship, and then have no way to see what the agent actually did with a real customer. Offline evals measure the model in a lab; Agent Analytics measures the behavioral trace downstream of the agent's decision, which is exactly the layer that's missing. We know how important it is to measure AI products so every Amplitude plan, including yours, gets a default allocation to use Agent Analytics. Go beyond observability. Get started today.

Where

Agent Analytics

Available in EUDC

Yes

When

Now!

Documentation

https://amplitude.com/agent-analytics

Show a guide or survey in more than one situation without duplicating it, by adding several trigger conditions to a single one.

WHAT IT IS

What it is: Add more than one trigger condition to a single guide or survey. The conditions combine as an OR list, so the experience launches when any one of them is met. This replaces the previous workaround of duplicating a guide and giving the copy a second trigger, which meant maintaining two of everything. One behavior change worth knowing: session properties moved from the trigger level to the nudge level, so a session-property condition now gates every trigger in the set rather than a single one.

Why: Customers may want to specify multiple different conditions that should trigger their guide or survey and up until now our best recommendation has been to duplicate the guide or survey and to specify the new one with your additional trigger condition.

Where: Available on the Setup tab when configuring a web guide or survey. Mobile not yet available.

Available in EUDC: Yes

When: Now!

Ask questions about your product data from a Discord channel and get an answer back with charts, without switching over to Amplitude.

WHAT IT IS

Run '/amplitude' with a prompt in a Discord server channel to ask a question about your product data. Amplitude answers in the channel with chart images and suggested follow-up buttons, running the query as you and respecting your existing project permissions. Setup has two parts: each person connects their own Discord identity from Personal Settings, and a Discord server admin installs the Amplitude bot into the server. The agent currently uses your last-used project, or your only project if you have just one.

Where: Any Discord server channel, via /amplitude. Connect your account at Settings > Personal Settings > Connected Apps > Discord.

Who: Organizations with Global Agent enabled.

Available in EUDC: Yes

When: GA now

WHO IT'S FOR

Free

Plus

Growth

Enterprise

Must have Global Agent enabled.

DOCUMENTATION

Integrate Discord https://amplitude.com/docs/analytics/integrate-discord

RELEASE DETAILS

RELEASE ID

amplitude-global-agent-in-discord

SHIPPED

Aug 18, 2026

AREA

AI

Spend less of your agent's context on tool definitions and more on the actual question, with a smaller and more task-shaped set of MCP tools.

WHAT IT IS

The Amplitude MCP server exposes a smaller set of outcome-based tools rather than one tool per operation. An agent picks a goal, like working with dashboards or cohorts, and passes an action, instead of loading separate tools for each CRUD step.

In internal measurement of the customer-facing tool surface, the upfront tool schema payload dropped by roughly 42%. Workflows that previously took several turns also collapse: dashboards and notebooks accept temporary chart-edit IDs and promote them server-side, so query, save, and insert happen in one call. When an event or property name doesn't resolve, the error returns the closest matching real names instead of asking the agent to go look them up.

Why: Not every MCP client supports dynamic tool search and the excessive context bloat caused due to our server required more end user tokens.

Where: Nothing to configure — the tools change automatically in any MCP client connected to Amplitude. Legacy tool names stop being advertised.

When: GA now

WHO IT'S FOR

Free

Plus

Growth

Enterprise

Available on the external MCP server.