ChinaAI / DeepSeek Search source-checking records / 23 September 2026

These files document three first answers from DeepSeek's official consumer web chat. The test protocol was written before each prompt was sent. Each task used a fresh chat and was not regenerated. Search was selected in all three tasks; DeepThink was additionally selected for task 3.

The *-ax.txt files are raw browser accessibility snapshots of the complete first answers. They include interface labels around the answer and are not hand-edited article transcripts. The PNG files are genuine captures of the displayed interface. They do not certify the answer's claims. The brief explains which displayed citations support or fail to support each assertion.

Original prompts and checks: protocol.md
Source-by-source assessment and direct Google search context: brief.md
First answers: task1-first-answer-ax.txt; task2-first-answer-ax.txt; task3-first-answer-ax.txt
Screenshots: both-controls-selected.png; task1-first-answer.png; task1-unsupported-claim.png; task2-first-answer.png; task3-first-answer.png; task3-both-modes-evidence.png

The signed-in chat did not display a serving model or plan. No API keys, private chats or credentials are included. These observations are not a benchmark or a claim about every account.
