Perplexity's 45-Point Freshness Premium
Key findings
- 45pp citation advantage for content under 30 days old (ziptie.dev)
- Content older than 18 months is effectively invisible to Perplexity
- Freshness effect is engine-specific — Gemini shows lowest freshness bias
The freshness signal
Perplexity's search architecture uses real-time web retrieval with a strong recency signal. Unlike ChatGPT (which blends training data with web search), Perplexity heavily weights recent content in its citation decisions.
Measurement
Using the ziptie.dev citation dataset, cited pages were binned by their last-modified date with citation probability calculated by age bucket:
| Content age | Citation probability | Relative to baseline |
|---|---|---|
| Under 30 days | 62% | +45pp |
| 30-90 days | 41% | +24pp |
| 90-180 days | 28% | +11pp |
| 180 days - 12 months | 19% | +2pp |
| 12-18 months | 12% | -5pp |
| Over 18 months | 3% | -14pp |
Baseline citation probability across all content ages: 17%.
Engine comparison
The freshness premium varies dramatically by engine:
- Perplexity: 45pp advantage for content under 30 days
- ChatGPT: 18pp advantage for content under 30 days
- Claude: 12pp advantage for content under 30 days
- Gemini: 8pp advantage for content under 30 days
This engine-specific variation reinforces the cross-platform overlap finding: each engine has distinct citation preferences, and optimisation must be engine-specific.
Implications
-
Content refresh cadence matters most on Perplexity. For brands where Perplexity citation share is the target metric, a monthly content refresh schedule produces measurably different outcomes from a quarterly schedule.
-
The 18-month cliff is absolute. Content older than 18 months has near-zero citation probability on Perplexity regardless of its quality, backlink profile, or topical authority.
-
Freshness is not equally important everywhere. Gemini shows the lowest freshness bias, suggesting that content quality and topical authority matter more for Google AI Mode than recency alone.
Where the AnswerGraph engine's own findings will live
The AnswerGraph panel tracks content freshness as a page feature in its inference model. Once enough observations accumulate to produce its own freshness-effect estimates with confidence intervals, those findings will be published here. Until then, the ziptie.dev dataset provides the best public evidence for engine-specific freshness effects.