Knowledge base and source traceability
The knowledge base is where GDDAO retains your organization's information. Understanding how it works explains why AI-generated content can be traced back to its sources.
Supported data types
- Documents: PDFs, Word documents, spreadsheets, and text files;
- Images: Photos and related image information;
- Audio and video: Recordings, videos, transcripts, and meeting excerpts;
- Webpages: URLs and captured webpage content;
- Folders: Folders for organizing documents, with configurable privacy and sharing scopes.
How information is retained
Depending on the entry and selected processing, GDDAO can perform these steps. Check each processing status to confirm completion:
- Parses text, images, or media;
- Produces summaries, transcripts, and searchable segments;
- Creates an index that AI can query and cite later.
Recordings can be transcribed again and revised. Web content can be saved and analyzed.
Finding the source of a figure later
Because all information is organized within the same knowledge base context, AI can cite sources directly when producing content. When you or a reviewer needs to verify a figure or statement, you can return to the relevant knowledge item and inspect the original document, photo, or transcript.
This is the key difference between GDDAO and a basic AI chat tool. The question is not only “What did AI say?” but “Which source supports this statement?”
Permissions and sharing: who owns and sees the data?
Access is restricted according to:
- Owner type: individual, project, or workspace;
- Workspace or organization scope;
- Project scope;
- Folder-level privacy and sharing settings.
Knowing a data item's ID does not let AI read it. Every access request is evaluated against these authorization scopes. Review project, folder, and individual sharing settings when inviting an external collaborator.
What a figure needs as support
A report stating “120 service attendances” needs a period, counting rule, original records, and a method for repeat participation. Attendances and unique people are different measures. Finding a list provides reference material; it does not establish that the list supports the statement.
Originals, processed content, and citations
The original preserves the material itself. Parsing or transcription makes content searchable. A citation connects a conclusion to the source used. These are related but distinct states. Inspect the source contents rather than relying only on filenames or summaries.
When support is missing
Check processing and permissions, then confirm that the AI task used the intended material. Mark unsupported statements for review and ask an owner to supply evidence. Preserving an explicit gap makes handoffs more reliable than inserting a plausible figure.
Return to the original content
Original, summary, citation, and evidence
| Term | Meaning | Check |
|---|---|---|
| Original | Received or recorded material | It may contain errors or omissions |
| Summary | A condensed interpretation | Conditions may have been omitted |
| Citation | A pointer to a used source | A link may not support the whole claim |
| Evidence | Material and reasoning supporting a question | Match population, period, and scope |
Meeting notes may support “accessibility was selected as a venue criterion.” They do not establish “the venue meets accessibility requirements” without a venue check. Small wording changes can require different evidence.
Help the next person find the right source
Use topic and date in filenames and explain scope in folder descriptions. Record why material changed and which version was adopted. Calling every file “latest” does not explain revisions.
Check for existing sources before importing again. Similar files from different periods can confuse selection; explicitly identify the version for the current request rather than expecting AI to guess.
Four source checks
- Does the material concern the same population and period?
- Is the statement an observation, a participant's account, or an inference?
- Did the summary omit a condition or exception?
- Can the intended reader access the original, or is an appropriate public explanation needed?
Practice: correct an overstatement
If the original says a participant wanted more discussion time, a draft claiming substantially improved outcomes is unsupported. Retain the participant's feedback and list the follow-up questions separately. Source retention helps readers determine the actual reach of each statement.
Next step
To understand how AI helps organize this information, read How AI collaborates.