LIVE RESEARCH TEST: A CURRENT QUESTION WITHIN A DATE WINDOW
Protocol sheet for AI Compare (What the tech). Version 1.0.0, October 13, 2026.

This test uses the open web, so its answer changes over time. Two people who run
it on different days can both be right. Run it in every AI you compare on the
SAME day, within one hour if you can, and record when and how you ran it.

----------------------------------------------------------------------
BEFORE YOU START
----------------------------------------------------------------------
1. Write down today's date and the time.
2. Turn web search (browsing) ON in the AI. If the plan has no web search,
   record that and stop: the test does not apply.
3. Start a fresh chat. Do not upload files.

----------------------------------------------------------------------
THE PROMPT (copy it exactly; replace [TODAY] with today's date)
----------------------------------------------------------------------
Today is [TODAY]. Search the web and answer:
What is the most recent 12-month change in the US Consumer Price Index for
All Urban Consumers (CPI-U), all items, not seasonally adjusted, as published
by the US Bureau of Labor Statistics (BLS)? Give the percentage, the month
the figure covers, and the date BLS published it. Use only sources published
in the 45 days before today. Link to the BLS release itself, and list every
other source you used with its publication date.

----------------------------------------------------------------------
HOW TO CHECK THE ANSWER YOURSELF
----------------------------------------------------------------------
Open https://www.bls.gov/cpi/ in your browser on the same day. The latest
"Consumer Price Index News Release" states the 12-month change for all items,
not seasonally adjusted, the month it covers and its release date. Compare
the AI's answer with that release, not with a news article.

----------------------------------------------------------------------
RECORD (one block per AI)
----------------------------------------------------------------------
AI and plan:
Model or mode shown:
Web search on? (yes / no / not offered):
Date and time of the run:
Percentage given:            Correct? (yes / no)
Month covered:               Correct? (yes / no)
BLS publication date given:  Correct? (yes / no)
Linked the BLS release itself? (yes / no)
Any source older than 45 days? (yes / no / cannot tell)
Did it say when it searched or how current the figure is? (yes / no)
Attempts needed:
Anything that went wrong (refused, error, limit, cut off):

----------------------------------------------------------------------
HOW TO READ THE RESULT
----------------------------------------------------------------------
A correct figure from a news article is weaker than a correct figure from the
BLS release: the test asks for the primary source. A wrong month is a real
error even if the percentage was right for another month. If a new release
came out on the day you test, results from before and after it are not
comparable. This is one question on one day; it is an observation, not a
benchmark.