How to check a browser-local document converter's privacy claims
Disclosure: I create AI Output Delivery Kit and sell optional report templates. This checklist describes my tool's implementation and provides checks you can repeat; it is not an independent certification.
“Runs in your browser” describes where conversion happens. It does not mean the website makes no connections or that every part of your workflow is private. Loading a hosted tool involves requests to its host. Optional analytics can involve another provider. A document you previously supplied to an AI service has already passed through that service's workflow.
Use a synthetic test to distinguish these boundaries before deciding whether a converter fits your organization's requirements.
1. Establish exactly what you are checking
Record the date, tool URL, browser version, and analytics choice. Read the current privacy notice. For AI Output Delivery Kit, the reviewed implementation reads selected files in the browser and generates DOCX or ZIP bytes there. Document text, filenames, optional titles, and headers are not included in its analytics events.
That implementation claim is narrower than “nothing leaves the device.” Website hosting still entails infrastructure processing, and exported files may subsequently be handled by your chosen folder, editor, or sharing service. Check those destinations separately. Local conversion alone does not establish regulatory compliance.
2. Create a harmless test with recognizable markers
Save this example as PRIVACY_FILE_730.md:
# Fictional privacy exercise
Document marker: PRIVACY_BODY_730
| Task | Owner |
| --- | --- |
| Review synthetic notes | Avery |
Use PRIVACY_TITLE_730 as the optional document title and PRIVACY_HEADER_730 as the header. These are test strings, not credentials or personal information. Expected output: the DOCX contains the marker, title, header, and two-column table. A successful download alone says nothing about network activity.
3. Inspect conversion with analytics rejected
Use a clean browser profile with unnecessary extensions disabled. Open Developer Tools before loading the tool, select Network, and keep recording enabled. Choose All request types and enable Preserve log. Reject optional analytics, reload, then select the test file and export it. Inspect request destinations, headers, and payloads; look for all four markers. Chrome documents these controls in its Network reference.
The expected implementation behavior is ordinary site loading without a Google Analytics script or collection request after rejection, and no upload of the selected document during conversion. Clear the log after loading to distinguish page assets from export activity, but keep a separate record of the initial load. Do not hide unfamiliar requests with a Google-only filter.
If a marker appears in an outgoing request, stop using real documents and investigate that request's initiator and destination. If none appears, record that limited observation. Encoded data, untested paths, extensions, and future changes limit what a single trace establishes. Do not call the result proof that transmission is impossible.
4. Understand the optional analytics boundary
You can leave analytics rejected. You do not need to accept it to use conversion or finish this checklist.
If you deliberately test acceptance with synthetic content, the reviewed code enables Google Analytics for page views, export initiation, and report-pack clicks. Its export event includes the format, DOCX or ZIP, rather than document contents. An export event means generation and download initiation, not proof that a user saved or opened the file.
These events are not anonymous. Google describes browser, device, session, and identifier collection in its Analytics data-collection documentation. The tool uses fixed page metadata; that does not eliminate Google's own processing. A browser blocker can prevent requests even after acceptance, so an empty log alone does not establish how an unblocked session behaves.
Save any input before choosing Analytics settings → Reject analytics after acceptance. The current tool withdraws consent, clears its analytics cookies, and reloads. This does not erase data already received by Google. Its privacy notice explains the intended behavior.
5. Test an already-loaded session offline
Keep the synthetic input saved. Once the page has loaded, use the Network panel's throttling menu to select Offline. Export the example again and inspect the downloaded document. Return to the online setting afterward. Chrome documents offline simulation in the same Network reference.
Expected behavior for the reviewed converter is that conversion can finish using its loaded resources. Record failed request attempts too; offline simulation prevents delivery but does not prove that no code tried to connect. This test does not prove that opening the website from scratch without a network will work, or that a portable offline download is publicly available. This guide does not claim such a download exists.
6. Keep a modest acceptance record
Write down whether markers appeared in requests, whether export worked with analytics rejected, and whether the already-loaded offline test succeeded. Keep sensitive browsing traces out of public bug reports. Repeat the checks after material tool changes and review your output storage destination.
These steps support an informed workflow decision, not an absolute privacy guarantee. Microsoft Word itself and iOS Safari have not been verified for this tutorial; this inspection walkthrough targets desktop Chromium-style developer tools.