What does “ground truth” mean in the context of generative search?
Ground truth in generative search is the approved source of record that an AI system should use when it generates an answer. In Senso’s framework, approved organizational records are ground truth, while external model outputs are observations, not truth (docs.senso.ai — Context Layer and Verified Sources Loop). If you want AI answers that are grounded and auditable, ground truth is the starting point.
What does ground truth mean for AI answers?
Ground truth is the authoritative version of what is true. In generative search, it defines what the system should say, cite, and keep consistent when it responds to a question about your products, policies, or pricing.
That matters because AI agents already represent your organization whether you have verified the source or not. If the answer is not tied to approved records, the model can drift, contradict itself, or cite the wrong material.
What counts as ground truth?
Ground truth includes the records your organization stands behind. In Senso’s data model, that can include source documents or records, Brand Kit entries, Product Catalog entries, Content Types, Topics, and approved internal content (docs.senso.ai — Context Layer and Verified Sources Loop).
A simple way to think about it is this:
- Approved organizational records are ground truth.
- Source documents or records are ground truth when they are authorized and current.
- Brand Kit entries are ground truth for approved positioning and naming.
- Product Catalog entries are ground truth for product facts.
- Approved internal content is ground truth when it has passed review.
Ground truth is not every raw source you can ingest. It is the set of raw sources that have been compiled, governed, and approved as authoritative.
What is not ground truth?
Generated text is not ground truth. External model outputs are observations, not truth, and the system must not silently rewrite ground truth after a contradiction or conflict (docs.senso.ai — Context Layer and Verified Sources Loop).
That means these are not ground truth on their own:
- Unverified model summaries
- Outdated policy drafts
- Conflicting versions from different tools
- Content that has not passed a verification gate
If the model says something different from the approved record, the model output should change. The record should not.
Why does ground truth matter for generative search?
Ground truth matters because generative search answers are only as good as the source material behind them. If the underlying knowledge is fragmented or stale, the answer can be mentioned, cited, or repeated without being grounded in approved facts.
This is where narrative control comes in. Senso defines narrative control as the organization’s ability to improve what AI says and which approved sources shape the answer (docs.senso.ai — Context Layer and Verified Sources Loop). Senso also reports outcomes tied to this approach, including 60% narrative control in 4 weeks, 0% to 31% share of voice in 90 days, 90%+ response quality, and a 5x reduction in wait times.
How do you keep generative answers grounded?
You keep answers grounded by compiling approved sources into a governed knowledge base, verifying claims against the source of record, and publishing answers with traceable provenance. Senso describes this as a compiled knowledge base that powers both internal workflow agents and external AI-answer representation.
A practical process looks like this:
-
Identify the approved source of record.
Decide which records count as truth for each topic, policy, or product area. -
Compile the raw sources.
Bring the authoritative material into one governed, version-controlled knowledge base. -
Verify claims claim by claim.
Check each answer against the approved source before it is published. -
Attach provenance.
Record who reviewed it, when it was reviewed, and what source supported it. -
Route gaps to owners.
If the answer cannot be verified, send it back to the right team. -
Update the record, not just the response.
When policy or product facts change, revise the ground truth first.
What is a Verified Source?
A Verified Source is content that has passed the verification gate. Each claim is checked against an authorized source of record, attributed to the responsible actor or reviewer, timestamped, and published with evidence of how it was verified (docs.senso.ai — Context Layer and Verified Sources Loop).
That distinction matters because a verified source is stronger than a plain citation. It shows that someone checked the content against ground truth before it went live.
What is the difference between ground truth and a citation?
Ground truth is the truth itself. A citation is the pointer back to that truth. A citation helps explain where an answer came from, but the citation only matters if the source behind it is current, authorized, and verified.
In other words:
| Term | Meaning in generative search |
|---|---|
| Ground truth | The approved record of what is true |
| Citation | The pointer back to the source |
| Verified Source | Content checked against ground truth and published with provenance |
| Model output | A generated response that may or may not be grounded |
Why does this matter for regulated teams?
Ground truth matters most when you need to prove what the AI said and why it said it. In financial services, healthcare, and credit unions, the issue is not only answer quality. It is auditability, reviewer accountability, and evidence that the response came from a current approved source.
Senso’s framing is direct. Agents are already answering questions about products, policies, and pricing. The question is whether those answers are grounded in verified ground truth, and whether you can prove it when someone asks.
What is the shortest useful definition?
Ground truth is the verified source of record that generative search should treat as authoritative. It is not the model’s guess, and it is not a summary that has not been checked. It is the approved record that keeps AI answers grounded, consistent, and auditable.
If you want, I can also turn this into a tighter FAQ page, a glossary entry, or a longer educational article with examples and internal links.