Gemini Workspace Export — Google Vault, Admin Console, and Enterprise Data Portability (2026)

Personal vs Workspace: which export path applies to you

The account type determines the entire export path:

The clearest signal for which path applies: if you sign in to takeout.google.com and see "Gemini Apps Activity" as a selectable service, Takeout is enabled for your account and the personal Gemini export guide applies. If Takeout is missing that service, or if takeout.google.com is blocked entirely ("Takeout is not available for your organization"), you are in a Workspace org that has restricted Takeout and the Vault path below is the relevant one.

The Workspace admin controls — before any export is possible

Two admin-level controls determine whether Gemini conversation data exists to export at all:

Control 1: Gemini Apps Activity (history retention)

Admin Console path: Apps → Google Workspace → Gemini Apps → Settings → Gemini Apps Activity

This toggle controls whether Gemini conversation history is saved. When disabled at the org level:

• Gemini sessions are not saved after the session ends.
• Users see no history in their Gemini interface.
• There is nothing to export via Vault or Takeout — the data was never retained.
• Users cannot re-enable history; it's a domain-level control.

When enabled (the default for most orgs that haven't explicitly reviewed Gemini settings), conversations are retained subject to the data retention policy below.

The toggle can be set per Organizational Unit (OU), which means engineering teams can have History on while other departments have it off. Check the OU setting for the specific user you're trying to export before running a Vault search — if the user is in an OU with History off, Vault will return zero results.

Control 2: Vault data retention policy for Gemini

Admin Console path: Google Vault → Retention → Retention rules → Gemini for Google Workspace

Even when Gemini Apps Activity is enabled, data is only retained for as long as the Vault retention rule specifies. Default behavior (no explicit Vault retention rule): data is kept indefinitely or until the user deletes it. With a retention rule: data is purged automatically after the retention period expires. If your org has a 90-day retention rule for Gemini and you're trying to export a conversation from 120 days ago, the data is gone — Vault will return nothing for that time range.

Exporting via Google Vault — the step-by-step path

Google Vault is the admin-only tool for e-discovery and compliance exports across Workspace services. The Gemini for Workspace data type was added in 2024. This path requires Workspace Business Standard or higher.

Step 1: Create a Vault matter

Navigate to vault.google.com and create a new Matter. The matter type is a container for the search and export — it's designed for legal holds and e-discovery, but it also works for straightforward data exports. Name the matter something descriptive ("Gemini export — [user name] — [date]").

Step 2: Run a search scoped to Gemini

Inside the matter, select Search. Set the following:

• Service: Gemini for Google Workspace
• Source: All data — or specify the user account(s) to scope the export
• Date range: Set start and end date — leaving this open exports all retained history
• Query operators: Vault supports keyword search within Gemini conversations (same query syntax as Gmail); leave blank to retrieve all conversations

Run the search and verify the result count. A count of zero with no error message typically means either History was disabled, the retention period expired, or the user account scope is wrong.

Step 3: Create an export

From the search results, click Export. Vault will process the export job in the background — completion time ranges from a few minutes for small accounts to several hours for large ones. You'll receive an email when the export is ready to download.

Step 4: Download and parse the MBOX

The export archive contains an MBOX file for each user. The MBOX format is an email archive format — each message in the MBOX corresponds to one Gemini conversation, with the conversation content in the message body. The content is typically formatted as plain text with alternating user/model turn markers.

The MBOX format — why it exists and how to parse it

Google Vault exports Gemini conversations in MBOX format because Vault was built as an email-archiving and e-discovery tool. When Google added Gemini to Vault's supported data types, the engineering choice was to use the existing Vault export pipeline, which outputs MBOX. This is the same format Vault uses for Gmail exports and Google Chat exports — a design decision that optimizes for infrastructure reuse, not for AI conversation portability.

The result: Vault Gemini exports are paradoxically harder to parse programmatically than personal Gemini Takeout exports, which produce HTML files that at least render directly in a browser. The MBOX format requires explicit parsing to access the conversation content.

Python script to parse Vault Gemini MBOX

import mailbox
import sys

def parse_gemini_mbox(mbox_path):
    mbox = mailbox.mbox(mbox_path)
    conversations = []
    for msg in mbox:
        subject = msg.get("Subject", "")
        date = msg.get("Date", "")
        body = ""
        if msg.is_multipart():
            for part in msg.walk():
                if part.get_content_type() == "text/plain":
                    body = part.get_payload(decode=True).decode("utf-8", errors="replace")
                    break
        else:
            body = msg.get_payload(decode=True).decode("utf-8", errors="replace")
        conversations.append({"subject": subject, "date": date, "body": body})
    return conversations

if __name__ == "__main__":
    convs = parse_gemini_mbox(sys.argv[1])
    for c in convs:
        print(f"--- {c['date']} | {c['subject']} ---")
        print(c['body'][:500])
        print()

The script uses Python's standard library mailbox module (no additional dependencies). Each MBOX message is one Gemini conversation; the body contains the full conversation text. The Subject header is typically the auto-generated conversation title; the Date header is the conversation creation timestamp. For downstream processing by decision-extraction tools, the body's conversation text is the relevant content.

What Vault captures — and the four gaps

When History is enabled and the retention period hasn't expired, Vault captures:

• Full conversation threads — all user turns and Gemini responses, in sequence.
• The model version — which Gemini model handled each response (logged in Vault metadata).
• Timestamps — conversation start and last-modified times, which the personal Takeout HTML export often omits or buries.
• User identity — the Workspace account that ran the conversation, relevant for multi-user export scopes.

Four things Vault does not capture:

1. Gem (custom Gemini agent) configurations — if a user interacted with a custom Gem, the conversation content is captured but the Gem's system prompt, instructions, and name are not included in the export. There is no Vault export path for Gem configurations; those are in the user's Gemini interface and have no separate export mechanism.
2. "Help me write" inline suggestions from Docs/Gmail — Gemini's in-product suggestions within Google Docs ("Help me write") and Gmail ("Help me write" smart replies) are not captured as conversations in Vault. They are treated as drafting assistance, not conversation history, and are not retained under the Gemini Apps Activity history policy.
3. Image generation outputs — Gemini image generation sessions are partially captured: the text prompts and Gemini's responses are in the conversation body, but generated images are stored as links to Google infrastructure, not as embedded files. If the image hosting expires, the link in the MBOX export becomes a dead reference.
4. Incognito mode sessions — Gemini sessions in Incognito mode are not saved and cannot be recovered, exactly as with personal accounts. Vault only captures what was stored; Incognito sessions store nothing.

Tier differences: what your subscription determines

The Workspace subscription tier controls both the export path and what is available at all:

For Business Starter orgs: the only export path is Takeout (if not disabled by the admin), which produces the same HTML format as the personal Gemini export. For Business Standard and above: both Vault (admin-run MBOX export) and Takeout (user-run HTML export, if enabled) are available.

When Takeout is disabled in your org

Enterprise and some Business Workspace admins disable Google Takeout domain-wide for data governance reasons — preventing employees from bulk-exporting org data to personal storage. When Takeout is disabled:

• Users attempting to access takeout.google.com see a message that Takeout is not available for their organization.
• The self-service export path is gone — the user cannot export their own Gemini history without an admin-run Vault export or a formal DSAR.
• A DSAR (data subject access request) submitted to Google via the Workspace privacy settings page triggers a separate process where Google produces an export on the user's behalf — but this is a legal-process path, not a quick self-service export. Response time is up to 30 days.
• The Vault path for admins remains available and is unaffected by the Takeout setting.

For users in a Workspace org who need their own Gemini conversation history: the most reliable approach is to ask your IT/admin team to run a targeted Vault export scoped to your account and a specific date range. This is a reasonable ask for legal, compliance, or onboarding-handoff purposes — it doesn't require a full e-discovery investigation.

Comparing Workspace Gemini export to other enterprise AI platforms

Gemini and Microsoft Copilot both route enterprise export through their existing e-discovery infrastructure (Vault and Purview respectively), producing email-archive formats (MBOX/PST) rather than purpose-built AI export formats. ChatGPT and Claude use purpose-built export pipelines that produce structured JSON — more useful for downstream processing but requiring separate infrastructure investment. For organizations that need AI conversation content in a machine-readable format, the Vault MBOX output requires an extra parsing step that ChatGPT and Claude exports don't.

How WhyChose fits in

WhyChose's primary input paths are ChatGPT JSON exports and Claude ZIP archives — the platforms with purpose-built machine-readable export formats. For Gemini Workspace exports via Vault, the MBOX parsing script above produces conversation body text that can be pasted as plain text input to the open-source extractor. The extractor identifies decision-shaped content — threads where alternatives were named, trade-offs were discussed, and a conclusion was reached — regardless of the original export format.

For engineering and product teams that use Gemini for Workspace as part of their AI-assisted decision workflow alongside ChatGPT or Claude, the complete decision record spans multiple platforms: Gemini for research and context (MBOX → parse → paste), ChatGPT or Claude for synthesis and drafting (JSON/ZIP → extractor). WhyChose handles the exportable layer automatically; the Gemini Workspace MBOX requires the manual parse step described here before the content is in a form the extractor can process.

Get early access

Related questions

Further reading

• Gemini conversation export — personal Takeout path, HTML format, and parsing recipes — the companion page for personal Google account users: how to request a Takeout export, what the HTML archive actually contains, the 30-line parsing script that normalizes Gemini's HTML export to a JSON shape, and what's missing from Takeout that Vault captures (timestamps, model version). If you're not in a Workspace org, this is the page that applies to you.
• Claude Team workspace export — admin DSAR path, member scope, and Project isolation — the symmetric Claude-side enterprise export guide: how Anthropic's admin DSAR path works for Claude Team orgs, what the org-scoped export includes vs what individual member exports cover, and how Project isolation affects conversation scoping. Shows the contrast with Vault's MBOX format — Claude Team DSAR produces per-conversation JSON.
• ChatGPT Team export — differences from Plus, workspace admin flow, and the Compliance API — the ChatGPT-side enterprise export guide: three ways Team differs from Plus (admin compliance export, workspace audit log, per-user export scope), the Compliance API for Enterprise orgs, and the sample audit-log.jsonl entries. Useful context for comparing how ChatGPT handles enterprise export vs Vault's MBOX approach.
• How to export your ChatGPT history (2026 guide) — the personal ChatGPT export path for comparison: Settings → Data Controls → Export data produces a ZIP with conversations.json, memory.json, and user.json. The JSON format contrast with Gemini's MBOX is the clearest illustration of how different platforms approach data portability.
• How to export your Claude conversations — the personal Claude export path: Settings → Privacy → Export conversations produces a ZIP of per-conversation JSON files. Another JSON-format contrast with Vault MBOX — Claude's export was designed with downstream processing in mind, Vault's was designed for e-discovery.
• Perplexity conversation export — no native export, GDPR path, and the manual workarounds — the fourth major AI platform, which has no export path at all (no Takeout, no Vault equivalent, no JSON). Useful context for the spectrum of AI data portability: ChatGPT/Claude (JSON, self-service) → Gemini personal (HTML, self-service) → Gemini Workspace (MBOX, admin-only) → Perplexity (manual copy, per-thread).
• How to extract decisions from your ChatGPT chats — the downstream workflow once conversations are exported: identifying decision-shaped threads where alternatives were named, trade-offs were discussed, and conclusions were reached. The same extraction logic applies to Gemini Workspace MBOX content once parsed — paste the conversation body text as plain text input to the extractor.
• The open-source extractor — processes ChatGPT and Claude exports automatically; accepts plain-text paste for Gemini Workspace MBOX content once parsed by the Python script above. Identifies decision-shaped conversations for structured ADR output regardless of the original export format.
• Microsoft Copilot export — how to retrieve conversation history (2026) — the other major enterprise AI platform in the admin-only export category: Microsoft 365 Copilot (embedded in Teams, Word, Outlook) exports via Microsoft Purview eDiscovery in PST format, similar to Vault's MBOX format for Gemini Workspace. Both are comprehensive for enterprise compliance; neither is self-service for individual users. Includes the platform comparison table across all five major AI platforms.
• Gemini Deep Research export — reports, citations, and what appears in Google Takeout — the specific guide for Gemini Advanced's Deep Research feature (available within Google Workspace via the Gemini Advanced add-on or Google One AI Premium). Multi-page research reports appear in the personal Takeout export as long model messages; they are also accessible via Google Vault for Workspace admins as part of the standard MBOX export. The page covers what citation data the export preserves, the intermediate search queries it does not preserve, and the decision-capture workflow for using Deep Research reports as ADR Context material.