Guide · Verifying

AI search and fact-checking

AI search answers a question in prose and cites pages it retrieved while answering. It fails in a specific way: the sentence can be wrong while the citation next to it is real, because the model wrote the sentence and attached the source afterwards. Checking means opening the source and finding the claim in it.

Last reviewed Who this is for: Anyone who publishes, advises or decides on the strength of something an AI told them, which is now most people who write for a living.

How long does it take to learn?

The method is an afternoon. Building the reflex of not believing a fluent paragraph takes a few weeks of catching yourself. It is less a body of knowledge than a habit, and habits are slower than facts.

The fastest path, in order

  1. 01

    Learn the difference between generating and retrieving

    A model with no search attached is recalling a compressed impression of its training data. A model with search reads pages first. The two fail differently, and knowing which one you are talking to changes what you check.

  2. 02

    Understand why sources get invented

    A fabricated citation is not a bug in a lookup, it is a plausible-looking string produced by the same process that produced the sentence. That is why fabricated references look so right: plausibility is exactly what the machine optimises for.

  3. 03

    Separate the claim from the citation

    Read the linked page and find the specific sentence supporting the specific claim. Most verification failures are not fake links, they are real links that do not say what the summary said they said.

  4. 04

    Triangulate on anything that matters

    Two independent sources that do not cite each other, or you do not have a fact yet. Aggregators quoting one another look like consensus and are one source wearing four hats.

  5. 05

    Date everything

    Ask when the source was published and when the model last knew anything. Most confidently wrong AI answers are correct statements about a world that has moved on, which is the hardest error to spot because nothing about it looks wrong.

  6. 06

    Write down what you could not verify

    An explicit unverified line is worth more than a confident sentence you are quietly unsure about. It is also what makes work defensible when someone eventually checks.

Tools worth your time

ToolWhat it is for
PerplexitySearch-first assistant that shows its sources inline, built for this workflow.
ChatGPT searchRetrieval attached to a strong model, with links you should still open.
Claude with web searchLong-context reading of the pages it finds, good for comparing several sources at once.
Google ScholarWhere you confirm an academic citation actually exists, in about ten seconds.
Semantic ScholarPaper metadata and citation graphs, useful for checking whether a study is contested.
The original pageNot a tool. Still the step almost everyone skips, and the only one that settles anything.

No detector or checker settles this for you. Tools that claim to verify AI output are themselves models, and running a model over a model to see whether the first one lied is not a chain of custody. The load-bearing step is a human opening the source.

Mistakes that cost people weeks

The track that teaches it

Why this is the fastest way to actually get there

Every guide tells you to verify. Almost none of them tell you what verifying is, which is why people who have read three of them still ship fabricated references. A method you can run in ninety seconds is what gets used, and what gets used is what works.

  • Taught as a repeatable procedure with a stopping rule, not as an instruction to be careful.
  • Built on how retrieval and generation actually differ, so the checks are aimed at the failures that exist rather than the ones people imagine.
  • Exercises use real AI output containing real errors, so you practise catching them rather than reading about catching them.

Questions people ask

Because it generates text rather than looking things up. A citation is a sequence of characters that fits the pattern of a citation, and a plausible one is exactly what the model is built to produce. Retrieval reduces this. It does not end it.

It is more transparent, which is not the same thing. It shows sources inline so mistakes are faster to catch. The summary sentence can still misstate what the source said, so the checking step does not go away.

For finding directions, framing questions and reading fast, it is genuinely excellent. As the final authority on any specific fact, no. Used as a first pass with a verification step attached, it is the most useful research tool in years.

Open the source, find the sentence that supports the claim, check the date, and look for one independent source that is not quoting the first. Under two minutes for anything that matters.

How AI search works underneath, why sources get fabricated, a verification method short enough to actually run, and how to use these tools for research you are willing to put your name on.

Vincoria opens soon.

One email when the doors open. Nothing else.

Founding members hear first. No spam.

Used only to tell you when Vincoria opens. One click to leave, from any email. Privacy policy

Next stepStart Lesson 01 · freeDevelop with AI · 9 min