Model Transitions
Citaition measures your brand's visibility across AI platforms. The analysis models that power those measurements are updated periodically. This page describes how we manage those updates.
Our commitment
We manage model transitions like an index-provider rebalance – announced, dual-run verified, and never silent. Every model change follows the same process:
- Announced in advance. Model changes are communicated before they take effect, not after.
- Dual-run verified. During a transition window, both the current and new analysis models run in parallel on every affected query. Your headline scores continue to use the current model until cutover.
- Uniform cutover. All accounts switch to the new model on the same date. There is no per-account opt-in – staggered switching would put different brands on different instruments, breaking cross-brand comparability.
- Permanently annotated. After cutover, a permanent marker appears on your visibility trend chart at the transition date, so any level shift is attributed to the instrument change, not to a change in your brand's actual visibility.
How managed transitions work
- Candidate validation. A new analysis model is tested against a structured set of queries. We compare mention accuracy, citation quality, competitor-set agreement, and search-invocation behaviour against the current model.
- Transition window opens. The new model is activated alongside the current model. Both produce results for every query. A bridge panel appears on your visibility tab showing side-by-side metrics from both models.
- Fidelity gate. Any brand whose mention rate under the new model drops more than 10 percentage points below the current model is flagged for review before cutover proceeds.
- Cutover. The new model becomes the production instrument. A permanent annotation is added to the visibility trend chart. The bridge panel remains visible for 30 days after cutover.
Current analysis instruments
The table below lists the model currently powering each AI surface measured by Citaition.
| Provider | Model | Since |
|---|---|---|
| ChatGPT | gpt-4o-mini | July 2026 |
| Gemini | gemini-2.5-flash | January 2026 |
| Perplexity | sonar | January 2026 |
| Google AI Overviews | Google's live AI output * | March 2026 |
| Google AI Mode | Google's live AI output * | May 2026 |
| * Google AI Overviews and AI Mode are Google's own features. For these surfaces Citaition measures the AI results Google itself returns, rather than running a separate analysis model of its own. | ||
Search-invocation rate
For each ChatGPT response we record the number of web searches the model invokes before answering. This metric is measured from the July 2026 model onward (the prior instrument did not expose per-response search counts). The search-invocation rate is available in per-brand analytics and contributes to citation source analysis.
For a dated log of all model updates, see the changelog.