What do you need before connecting AI search to Drive?
Start with Google Workspace accounts. Google's Gemini Enterprise documentation says personal Google accounts ending in @gmail.com are not supported for Drive data stores, because they have no Workspace customer ID. It asks that documents be made accessible by placing them in a shared drive owned by the domain or assigning ownership to a user in the domain. It also notes that with data federation, the Drive connector only searches documents owned by the searching user's domain. Files owned by a supplier or a former contractor's personal account may be invisible to search.
Why do shared drives matter?
In My Drive, the person who created a file owns it. In a shared drive, the organisation owns the files, so Google says they remain when someone leaves and an administrator deletes their account. For AI search, that ownership difference decides whether your procedures survive staff turnover, and whether connectors that only search domain-owned files can see them. Move approved policies, procedures and manuals into a small number of shared drives with clear membership before you connect anything, and keep sensitive material in separate drives with narrow membership.
How do Drive sharing settings affect what people see?
Drive has four permission types (user, group, domain and anyone) and roles from reader up to owner. Google's developer documentation explains that child files and folders inherit permissions from their parent folder, that an inherited permission cannot be reduced on a child item unless the folder uses limited access, and that a child item can grant more access than its parent. A folder shared with the whole domain years ago therefore opens everything inside it. Review domain-wide and link-shared folders before launch, and ask each vendor how their connector treats link-only sharing.
How much of each file can be searched?
Google lists these as search index limitations of Google Drive in its Gemini Enterprise documentation. Drive extracts only 1 MB of text and formatting data from a file to make it searchable, and keywords beyond that point are not searchable. Most file types cannot exceed 10 MB, with exceptions of 20 MB for XLSX, 30 MB for PDF and 100 MB for text files. Optical character recognition in PDF files is limited to 80 pages. A long scanned manual can therefore be only partly searchable. Split large manuals by chapter and give scanned documents a text layer before relying on them.
Where is the data held?
Google says the Gemini Enterprise Google Drive data store is supported only in its global, US and EU locations, and that Gemini Enterprise's data residency commitments cover data in Google Cloud. For data held in Drive itself, Google points to Workspace data regions. If a client contract requires Australian processing, confirm these terms with Google or your reseller in writing. Ask the same question of any other connector or custom build, including where the model that writes the answer runs.
Can you limit which folders AI search reads?
Gemini Enterprise offers folder filters and shared drive filters, each set to include only the listed locations or to exclude them, and filters apply to nested content. Google states that file access is always governed by the user's Drive permissions and that filters never grant additional access. It also warns that filters only restrict search results, that users can still reach out-of-scope files through other enabled actions if they have permission, and that filters should not be treated as a security boundary. Fix the sharing on sensitive folders instead of relying on a filter.