Genome Computer
Back to tutorials

Your first walkthrough 路 Lesson 3 of 3

Check the evidence in an AI answer

Trace each claim to a file or source, and spot unsupported conclusions.

6 min read 路 allow extra time to follow alongAny AI tool

Before you start

  • An AI answer to review, from your first conversation or another session.

AI can move quickly from a gene to a pathway to a possible interpretation. That speed is useful, but it can also make weak answers sound stronger than they are. Reading critically means slowing the answer down and asking what each claim is attached to.

Keep the original question and answer beside you. For each claim, identify the file record or checked source that supports it. If you used the synthetic exercise, the source placeholder is the first thing to flag.

Five evidence categories

  • Direct file evidence: something found in your .genome bundle.
  • Annotation context: an interpretation or label included in the bundle.
  • Source-backed research: a claim tied to a named paper, database, or source.
  • Reasoned inference: a connection the model is making from multiple pieces of evidence.
  • Speculation: a claim that sounds plausible but is not grounded enough to rely on.

Critical reading workflow

  1. Let the model answer your original question.
  2. Ask it to classify every important claim into the five evidence categories.
  3. Ask which claims came directly from your .genome.
  4. Ask which claims depend on external research.
  5. Ask what source, gene, variant, or file supports each claim.
  6. Ask it to remove or soften unsupported claims.
  7. Save the cleaned answer as the version worth keeping.

Audit the answer

Review your previous answer. Break it into five sections: direct evidence from my .genome, annotation context from the bundle, source-backed research, reasoned inference, and speculation. For each important claim, name the gene, variant, annotation, file, source, or missing evidence that supports it.

Rewrite with evidence

Rewrite the answer as an evidence-grounded summary. Keep claims that are connected to my .genome or a named source. Soften claims that are uncertain. Remove claims that are not tied to a file, annotation, variant, pathway, or citation.

Signs of a weak answer

  • It talks about a trait without naming genes, variants, annotations, or sources.
  • It says your genome shows something but does not say where in the bundle it found it.
  • It treats a research association as a personal conclusion.
  • It uses confident language while admitting it did not inspect the relevant file.
  • It does not tell you what would make the answer stronger.

A strong answer usually becomes more modest after review. That is a good thing. The point of .genome exploration is not to force certainty. It is to make your genome easier to question, inspect, and understand.

Before you continue

Check each item when you can do it.

Walkthrough complete 路 Put it into practice

Create an evidence-backed report

Save a focused investigation with file references and a reusable prompt.