ChatGPT

Verified against ChatGPT · 2026-08-01

Structure a live web-search query so it separates verified facts from best guesses

Directs ChatGPT's web search on a fast-moving or recent topic to tag every claim with a confidence level and a source, rather than blending confirmed facts and plausible inference into one uniformly confident-sounding answer.

ChatGPT (Web Search)4 fillable variables

The prompt

Ready to copy — highlighted parts are example details you can swap.

You are answering the question below using live web search, on a topic where the facts are recent enough or contested enough that getting the sourcing right matters more than getting a smooth answer. Search first, then answer — and tag every material claim with where it came from and how confident it actually is.

QUESTION
Has the vendor we use for background checks changed its standard turnaround time in the last few months?

WHY RECENCY OR VERIFICATION MATTERS HERE
We're quoting turnaround time to candidates in offer letters — an outdated number creates a broken promise.

SOURCE PREFERENCE
The vendor's own status/changelog page and direct customer reports over third-party review aggregator summaries.

WHAT I ALREADY BELIEVE, TO CHECK RATHER THAN CONFIRM
Believed to still be 3-5 business days, which is what it was when the contract was signed eight months ago.

SEARCH AND ANSWER RULES
Search before answering, using multiple distinct queries if the topic has more than one angle, rather than answering from general knowledge and only searching to backfill a citation for what was already going to be said — the search results should be able to change the answer, not just decorate it. For every material claim, mark it as: Confirmed (multiple independent sources agree, cite at least two), Single-sourced (only one source found, name it and note that it's unconfirmed elsewhere), or Inference (synthesis or reasoning across sources, not a direct claim any one source makes) — do not present an inference with the same confidence as a confirmed fact just because it sounds equally plausible in the final answer. Apply the stated source preference as a real filter on which sources get weighted — if the preference calls for primary sources and the actual top search results are dominated by a different kind of source, say so explicitly rather than quietly using what came up first. If the stated existing belief turns out to be accurate, say so plainly rather than treating confirmation as a less interesting answer to give than correction. If it turns out to be outdated, partially wrong, or contested, say specifically what changed or what the actual state of disagreement is, not just that it's no longer accurate without the update itself. Note explicitly if search results are themselves stale relative to how fast this specific topic is moving — a source from even a few weeks ago on a fast-moving story can already be superseded.

OUTPUT FORMAT
Lead with a direct answer to the question in one or two sentences. Then a claim-by-claim breakdown: Claim | Confidence tag | Source(s). Close with one line on whether the stated existing belief held up, was updated, or was contradicted, and by what.

Customize

Optional — swap in your own details for the highlighted parts above.

Why this works

ChatGPT's web search retrieves and reads live pages and can genuinely change an answer based on what it finds, but nothing about the underlying model changes its own confident, fluent writing style based on how well-sourced a given sentence actually is — a claim backed by two independent sources and a claim that's really the model's own inference read identically fluent in prose, so confidence tagging has to be an explicit instruction, because the model has no built-in incentive to visibly hedge an inference just because it's less certain than a directly-sourced fact. Requiring the search to happen before forming the answer, rather than after, targets a specific and easy default: a model asked a question it has some prior belief about can search mostly to find a citation supporting what it already planned to say, producing search-backed-looking output that never actually let new information change the conclusion. Explicitly checking a stated existing belief against symmetrical bias in either direction matters because a model given "check whether X is still true" can develop a subtle bias toward finding something has changed, since a correction reads as more informative than a plain confirmation, and that bias is exactly backward when the actual base rate of a specific claim staying true is high. Flagging when search results themselves are stale relative to the topic's own pace of change closes a gap specific to live search that a static-knowledge answer doesn't have: a page from three weeks ago is presented by search tooling as "found live," which reads as current, but for a genuinely fast-moving situation three weeks old can already be wrong — the risk isn't that the source is unreliable, it's that even a reliable source can be outdated by the time it's retrieved.

Verified against

ChatGPT GPT-5.1 (Web Search) · 2026-08-01

Changelog

  • 2026-08-01 — Initial publish, verified against ChatGPT GPT-5.1 Web Search.

Building this for real?

This is a free starting point. If you'd rather have what Scult builds built and running for your business, that's Scult's day job.

EXPLORE WHAT SCULT BUILDS
All ChatGPT prompts

Check your AI visibility

One URL in, a 0–100 score and the exact fixes out.

RUN THE CHECK

Browse all the tools

15 tools across six categories
13 of them never send your data anywhere

Free · No signup · No trial clock

SEE THE DIRECTORY