search-rateShare of a day's questions, per mode, for which ChatGPT queried the web (`routing.searched`). Measured 22 Aug 2026: 18 of 26 in standard mode (69 %), 5 of 7 in thinking mode. The metric is 0 or 1 per question; the time series shows which questions search consistently, which consistently do not, and which flicker. After two measurement days three of 26 flickered (K8-01, K11-02, K9-02).
| Prompt | 20 Aug 2026 | 21 Aug 2026 | 22 Aug 2026 |
|---|---|---|---|
| K1-01 instant | 1 | 1 | 1 |
| K1-01 thinking | 1 | 1 | 1 |
| K1-02 instant | – | 1 | 1 |
| K10-01 instant | – | 1 | 1 |
| K11-01 instant | – | 1 | 1 |
| K11-02 instant | – | 0 | 1 |
| K12-01 instant | – | 0 | 0 |
| K12-02 instant | – | 0 | 0 |
| K13-01 instant | – | 1 | 1 |
| K14-01 instant | – | 0 | 0 |
| K14-01 thinking | – | 0 | 0 |
| K14-02 instant | – | 0 | 0 |
| K14-02 thinking | – | 0 | 0 |
| K2-01 instant | 1 | 1 | 1 |
| K2-01 thinking | 1 | 1 | 1 |
| K2-02 instant | – | 1 | 1 |
| K2-03 instant | – | 1 | 1 |
| K3-01 instant | – | 1 | 1 |
| K3-02 instant | – | 1 | 1 |
| K4-01 instant | 0 | 0 | 0 |
| K4-01 thinking | 0 | 0 | 0 |
| K4-02 instant | – | 0 | 0 |
| K5-01 instant | 1 | 1 | 1 |
| K5-01 thinking | 1 | 1 | 1 |
| K5-02 instant | – | 1 | 1 |
| K6-01 instant | 1 | 1 | 1 |
| K6-01 thinking | 1 | 1 | 1 |
| K6-02 instant | – | 1 | 1 |
| K7-01 instant | – | 1 | 1 |
| K7-02 instant | – | 1 | 1 |
| K8-01 instant | – | 0 | 1 |
| K9-01 instant | – | 0 | 0 |
| K9-02 instant | – | 1 | 0 |