Read the video transcript
Read along, pause, or return to an idea. This transcript follows the selected lesson video.
I'm Carla, an AI avatar.
AI can summarize a news article in seconds.
But a summary can quietly change the story.
Today I'll show you how to catch it.
This matters.
In a study published October 21,
2025, journalists from 22 public service media
organizations reviewed more than 3,000 AI answers about the news.
They found that "Almost half of all AI answers had at
least one significant issue."
The same study found that a third of responses had serious
sourcing problems, and a fifth had major accuracy issues,
such as hallucinated or outdated information.
So I treat every AI summary as a draft to check.
I start with an article my teacher shared about the
Stanford study of AI and early-career jobs.
Before I open any AI, I read it myself and underline every
number, date, name, and hedge word,
like may, could, and some.
Why read first?
I want my own picture of the article before the AI hands
me its picture.
Otherwise I might not notice what is missing.
Now Dr. Q's Five Steps.
Role: "You are a careful news summarizer.
Do not do anything yet."
Objective: "I need an accurate five-bullet summary for my
AI News Log."
Context: the article, its outlet,
and its date.
Task: "Summarize it in five bullets.
Keep every number, date, and name exactly as written.
Keep words like may, could, and some.
Then list anything important you left out."
Here is the summary.
Five neat bullets.
It reads well.
Now step five, the audit.
I put the article and the summary side by side and fill in
a table: the detail in the article,
the detail in the summary, and the result.
Same, changed, dropped, or added.
I check four things in every bullet:
numbers, dates, names, and hedge words.
Hedge words matter.
"May cause" and "causes" are very different claims.
Bullet one.
The article says "workers ages 22 to 25."
The summary says "young people."
That is changed, and less exact.
The article says the researchers found "no evidence of
widespread, economy-wide job displacement."
That line is missing from the summary,
even though I asked for anything important it left out.
Dropped.
Bullet four has a quote from a researcher.
I search the article for it.
It is not there.
The AI added a quote the article never contained.
That one worries me most.
A made-up quote in a summary looks exactly like a real one.
The only way to catch it is to search the original for
every quote.
Then I verify the main claim by reading laterally.
I open the study's own page at the Stanford Digital
Economy Lab and one more news report.
I check that the numbers match the original.
Then I log the story in my AI News Log:
date, headline, outlet, label, what was measured and by
whom, two other sources, my verdict,
and any hype words.
One story a week, all semester.
Last, I record my AI use: the tool,
what I asked, and what I kept.
I kept nothing I had not checked.
Count them up: one changed, one dropped, one added.
None of them jumped out until I compared line by line.
Three problems in five bullets.
The summary sounded perfect.
The audit showed me what changed.
Your turn: find what your AI changed,
dropped, or added.
