Verify a Google Drive Knowledge Import
Check the selected files, visible source status, and displayed errors, then test representative questions against the approved document.

Treat import as a sequence to verify
A connected account is not evidence that the assistant has received useful content. In Resonoon, Google Docs knowledge uses Google Drive Picker: authorize the account, browse and select files, then confirm a non-empty selection to start import and save it. Each step answers a different question. Authorization establishes access to the picker. Selection identifies intended files. Confirmation commits that choice. Visible source status and displayed errors help you inspect what happened. Skipping one check can leave a team believing a source is ready when it is not.
Before starting, write down which document should appear and one question it should answer. Confirm that the document is approved and that the account is the one intended for this assistant. In the picker, select the known file and verify the selection before confirming. A non-empty selection is required. Connecting the account alone imports nothing, and closing the picker without confirming should not be described as a completed import.
Inspect evidence after confirmation
After confirming the selection, return to knowledge-source management and inspect the source entry. Review the displayed selection, status, and any errors. Use the words and details the interface actually shows; do not assume fixed status names or a particular processing time. An entry that looks connected may only establish access, so check whether the selected material and import result are visible. If an error appears, read it in context and check the account and selected files before retrying.
A useful verification record includes the date, assistant, source document, visible result, and test question. Keep the record in a shared team document so the next reviewer can repeat the checks and identify the source used. It helps distinguish an initial authorization from later changes. If several similar files exist, record the exact approved version so another reviewer can reproduce the selection. Avoid logging sensitive information from customer conversations when a simple question and policy excerpt are enough.
Imagine an assistant for a repair service. The team selects an approved warranty guide but later sees an error in the source view. It should not announce that warranty answers are ready based solely on the account connection. Confirm that the correct guide remains selected, read the displayed error, and follow only the visible retry or sync operation. Then inspect the result again. If the source still cannot be verified, keep the assistant's response boundary conservative and have staff answer the question from the approved policy.
Test the content, not just the connection
A successful looking import still needs a content check. Ask one question whose answer appears explicitly in the document, one question that tests an important condition, and one question outside the selected source. Compare each response with the exact passage. A relevant answer should preserve conditions such as coverage dates, exclusions, or required evidence. An unsupported question should not invite an invented policy. Record expected behavior before running the test to avoid grading a response by whether it merely sounds plausible.
For a warranty guide, the direct question might ask how long coverage lasts. A boundary question could ask whether accidental damage is included if the guide lists exclusions. The outside question might ask how to transfer coverage to a different owner when the selected source says nothing about transfers. The test should expose whether the material itself is incomplete or the answer goes beyond it. If a gap appears, correct the approved source or define a safe response, then repeat the same prompt.
Do not confuse a source check with a guarantee about every answer. A single successful question only verifies that specific case. Choose examples that reflect real customer concerns and revisit them when the policy changes. Keep the same prompt set long enough to compare results, but update expected answers when the owner approves a new rule.
Monitor changes without promising immediacy
Connected source management includes operations for overview, browse, save, sync, and delete. Use the visible operation appropriate to the source and inspect its displayed result. If the underlying document changes, do not promise that the assistant reflects the edit immediately. Check the visible status and last-synced details before treating the refresh as complete. When a change affects an urgent customer answer, use an approved interim response until the source and relevant tests have been reviewed.
Give someone responsibility for the source and for recognizing policy changes. Set a clear trigger, such as a revised warranty period or new exclusion, and record who approved the new wording. After syncing where appropriate, rerun the direct, boundary, and outside questions. If the answer differs, compare both the source version and the assistant response. The AI chatbot widget page provides public context on the customer-facing assistant.
Verification is practical evidence: the intended account was authorized, the intended file was selected and confirmed, the interface displayed a result, and representative answers were checked against the approved text. Keeping those observations separate makes troubleshooting specific. It also prevents a connected account or an optimistic assumption from standing in for a verified knowledge import.