A team drafts client advice with ChatGPT. Their rule is: always enable web search, and publish any answer that comes back with citations attached. Where does this rule fall short?

Hardchatgptaccuracyverification

Last reviewed 11 Aug 2026

Citations should be removed before publishing, since showing sources implies a guarantee the team cannot give
The rule is sound, since a citation is what turns an answer into a verified claim
Search should be left off, because retrieved pages are less reliable than what the model already knows
A citation makes a claim checkable, but nobody opens it, and client advice is where checking mattersCorrect answer

Show answer & explanation

Correct answer: D. A citation makes a claim checkable, but nobody opens it, and client advice is where checking matters

Explanation

The principle — a citation changes what is possible, not what has happened. It makes a claim checkable by someone, and a workflow only benefits if someone actually does the checking.

Why the key is correct — the team has built half of a good process. Enabling search is right, because ChatGPT can search the web and cite sources when it needs current information, and deep research returns a report with citations or source links specifically so you can verify the information. The word in OpenAI's own description is verify, which presupposes a person doing it. Against that, OpenAI's guidance is explicit that ChatGPT can make mistakes and that you should check important information, particularly anything date-, time- or location-sensitive. Client advice is the highest-stakes case in the scenario, so it is precisely where the missing step costs the most. The rule as written treats the appearance of a link as the end of the process rather than the beginning of it.

Why the others are wrong — turning search off would make things worse, since retrieval is what makes a current claim checkable in the first place. Calling the rule sound repeats the exact confusion under examination, treating a citation as evidence that checking occurred. And removing citations before publishing is the worst of the four: it makes the claim harder for anyone downstream to verify while doing nothing about whether it is correct.

Remember this — the useful test for any AI workflow is to ask which human step converts a plausible answer into a checked one, and whether anyone actually performs it. Calibrate that step to the stakes: brainstorming tolerates errors easily, while advice someone will act on deserves a source that has been opened and read. A process that produces citations nobody reads has bought the appearance of rigour rather than rigour.

Sources — OpenAI's ChatGPT Voice, What is ChatGPT FAQ and Deep research articles.

Sources

Primary sourceChatGPT Voice — OpenAI · accessed 12 Jul 2026
“ChatGPT can make mistakes. Check important information, especially for date-sensitive, time-sensitive, or location-sensitive questions.”
Primary sourceDeep research in ChatGPT — OpenAI · accessed 29 Mar 2026
“You get a structured report with citations or source links so you can verify the information.”

Practise 10 questions on this topic

Take ChatGPT Basics — Timed Test 1 (10 questions) — scored instantly, explanation for every question, no login.

Start the test

A team drafts client advice with ChatGPT. Their rule is: always enable web search, and publish any answer that comes back with citations attached. Where does this rule fall short?

Related questions

A team drafts client advice with ChatGPT. Their rule is: always enable web search, and publish any answer that comes back with citations attached. Where does this rule fall short?

Related questions