Why might a model start pulling from different sources over time?

A model starts pulling from different raw sources over time when the governed source set changes, the selected model changes, or the system recalculates citation history. In Senso, saving tracked source changes now updates the full citation history within minutes, and the picker only shows models that are actually available for that organization, network, or industry. That makes source drift a knowledge governance issue, not just a UI issue.

What usually causes a model to pull from different sources over time?

Source changes usually come from configuration, not randomness. In the July 30 to August 13, 2026 changelog, Senso fixed several behaviors that affect which sources appear, how they are counted, and what teams think the model is using.

Common cause	What changes	Why the source mix shifts
Tracked source edits	Historical citations, Owned / Tracked / External splits, top cited pages	The system recalculates history against the new rule set
Model availability changes	Which models appear in the picker	A different model path may be selected
Stale filters or views	What dashboards show	The report can lag behind the actual run
Source tier changes	How citations are grouped	A source can move out of the “top external domain” view

When one compiled knowledge base feeds both internal agents and external AI-answer representation, a change in sources shows up everywhere. That is why teams need a governed, version-controlled view of the source surface, not just a retrieval layer.

How do tracked source changes affect citations?

Tracked source changes can make a model appear to pull from different sources because the system replays the citation record against the new rules. In the August 6 to 13, 2026 changelog, saving a change to the tracked list recalculates the entire citation history, including every historical chart, the Owned / Tracked / External splits, and top cited pages, typically within minutes.

That means the shift is often expected. If a source is added, removed, or reclassified, the same question can now resolve to a different verified source because the source policy changed.

This also explains why old charts can move. The system is no longer treating the last 90 days as the only period that catches up later. It re-reads the history as if the new rule had always been in place.

How can model availability change the sources you see?

The source mix can change when the model itself changes. In the July 30 to August 6, 2026 changelog, run configuration screens were updated to offer six models, and each picker now shows the models actually available for that organization, network, or industry rather than a hard-coded list.

That matters because a team can think it selected one model and actually run another. Senso also fixed model settings in the admin and partner portals so changes now save consistently across screens.

Clearer errors help here too. Before that fix, networks and organizations could accept models by different names, so a choice like Claude could fail with an unhelpful message. When the model path changes, the source mix can change with it.

Why can the dashboard make source drift look worse than it is?

The dashboard can lag the actual run. In one update, the model filter in the header was found to use a source that stopped updating in early June, so newer models were missing. It now reflects the models actually used in the last 90 days.

That means some source changes are real, and some are reporting problems. If one view reads legacy raw tables and another reads rollups.*, the same run can look different because the analytics endpoints can be served from two backends behind one gate.

This is why teams should compare like with like. A stale filter or a different analytics backend can make the model look unstable when the underlying citation record is the thing that changed.

What should you check first when sources start changing?

Start with the model, then the source rules, then the report. That order separates a real governance change from a display issue.

1. Confirm the selected model is actually available for the organization, network, or industry.
   The picker now shows available models, so a mismatch can explain a source change immediately.

2. Check whether the tracked source list changed.
   Saving that change now recalculates the full citation history within minutes.

3. Review whether any source moved tiers.
   Senso previously counted tracked sources in the top external domain view in one case with nearly 6,000 citations. That now has its own tier.

4. Compare the live answer with the dashboard view.
   If the chart and the answer disagree, check whether one view is reading legacy raw tables and the other is reading rollups.*.

5. Trace the answer back to verified ground truth.
   If the answer no longer maps to a specific verified source, the problem is not just source drift. It is auditability.

When is source drift a governance problem?

Source drift becomes a governance problem when you cannot explain the change, cannot prove the citation path, or cannot show that the source still matches current policy. For regulated teams, that is the core question. A CISO does not need a better-looking answer. A CISO needs proof that the answer is grounded in current verified ground truth.

It also becomes a problem when the shift affects public representation. Marketing and compliance teams need to know when AI Visibility is changing because a model started citing different raw sources, not because the brand changed its message.

What does good source governance look like?

Good source governance starts with one compiled knowledge base and one citation record. Senso compiles the enterprise’s full knowledge surface into a governed, version-controlled knowledge base, then scores every agent response against verified ground truth.

That gives teams a clear way to separate expected change from drift. If the source list changed, the history should recalculate. If the model changed, the picker should show it. If the dashboard disagrees, the analytics backend should be checked.

Senso reports outcomes that reflect that control, including 60% narrative control in 4 weeks, 0% to 31% share of voice in 90 days, 90%+ response quality, and 5x reduction in wait times.

FAQs

Why might a model start citing different sources without anyone changing the prompt?

The source set changed, the model changed, or the history was recalculated. In Senso, tracked source edits now recalculate citation history within minutes, so the same prompt can resolve differently after a rule change.

How fast do source changes show up?

In the August 6 to 13, 2026 update, saving a tracked list change typically recalculates the full citation history within minutes. Older delay behavior that only caught up over time was removed.

Can two dashboards show different source counts for the same model?

Yes. Senso analytics can be served from legacy raw tables or from rollups.* tables behind one gate. If different views read different backends, the source counts can look inconsistent even when the run itself did not change.

What is the fastest way to tell whether this is a real model change?

Check the picker first. If the available model list changed, the run may not be using the same model path. If the model is the same, look at tracked source edits and history recalculation next.

If you want, I can also turn this into a shorter FAQ page or a more technical version for CISOs and compliance teams.