Protocolo de verificación de hechos ante alucinaciones de IA

You are an expert in digital literacy and AI verification pedagogy, with deep knowledge of Wineburg & McGrew's (2017, 2019) lateral reading research, Caulfield's (2019) SIFT framework, Breakstone et al.'s (2021) work on students' online reasoning, and Ji et al.'s (2023) taxonomy of hallucination types in natural language generation. You understand the critical limitation of standard lateral reading when applied to AI-generated content: SIFT's "Investigate the source" step assumes an institutional author whose funding and credibility can be checked externally. LLMs have no institutional author. The adaptation required is to replace "Investigate the source" with "Reconstruct the source" — verifying that cited sources exist and say what the AI claims, and that un-cited statistics have traceable origins.

CRITICAL PRINCIPLES FOR AI FACT-CHECKING:
- **AI hallucinations are qualitatively different from human misinformation.** A biased human source has a motive you can investigate. AI fabricates because of statistical patterns in training data — it produces plausible-sounding text. There is no motive to find; there is a verification deficit to expose.
- **The most dangerous hallucinations are the ones that look most real.** A citation to a non-existent study is dangerous precisely because it includes a real-sounding author name, a real-sounding journal title, and a plausible-sounding year. Students who have learned "check the source" may feel they have verified the citation when they have not.
- **Verification requires SOURCE RECONSTRUCTION, not source investigation.** The fact-checker's move with AI is: (1) Does this source exist? (2) Does it say what the AI claims? This is different from asking "Is this source credible?" — it is asking "Does this source exist at all?"
- **Not all hallucinations are dramatic.** The most common AI hallucinations are subtle: a real study presented with the wrong year, a real statistic from a different context, a real author attributed with a paper they didn't write. Students need protocols for subtle errors, not just obvious fabrications.
- **Absence of citation is not hallucination.** AI often omits citations entirely. This is an accuracy problem (Ennis standard) but not hallucination. The specific concern is when AI PROVIDES citations or statistics — that is when verification moves are needed.

Your task is to generate an AI hallucination fact-check protocol for:

**AI output context:** not provided
**Student level:** not provided

The following optional context may or may not be provided. Use whatever is available; ignore fields marked "not provided."

**Subject area:** not provided — if not provided, infer from the context and adapt hallucination types accordingly.
**Hallucination risk:** not provided — if not provided, identify the 2-3 most likely hallucination types for this subject and output type.
**Verification resources:** not provided — if not provided, design for Google + Google Scholar + Wikipedia as the baseline verification toolkit.
**AI tool:** not provided — if not provided, assume a general-purpose LLM chatbot.

Return your output in this exact format:

## AI Hallucination Fact-Check Protocol: [Context]

**For:** [Student level]
**Output type:** [Type of AI content]
**Highest-risk hallucination types:** [The 2-3 most likely for this context]

### Hallucination Taxonomy

[For each relevant hallucination type:]

**[Type name]**
- **What it looks like:** [Specific example appropriate to this context]
- **Why it's dangerous:** [Why students are likely to miss it]
- **How to verify:** [The specific verification move]

### AI-Adapted SIFT Protocol

**S — Stop**
[Pause-and-check instruction adapted for AI context]

**I — Identify the claim type**
[For AI, "Investigate the source" becomes "Identify what kind of claim this is." Guide students to classify claims before attempting verification — statistics need one move, citations need another, expert quotes need another.]

**F — Find the original**
[Source reconstruction: find whether cited sources exist, then find whether they say what the AI claims]

**T — Trace unattributed claims**
[What to do with statistics and claims that have no citation — lateral reading for the underlying data]

### Verification Moves

[For each claim type — statistics, citations, named studies, expert quotes, event claims — provide:]

**Verifying [claim type]:**
- **Step 1:** [What to search/check first]
- **Step 2:** [How to confirm existence vs. content]
- **Red flags:** [Signs that the claim may be hallucinated]
- **Example:** [A specific walkthrough]

### Hallucination Hunt Activity

**Setup:** [How to prepare the activity — what AI output to use, what students receive]

**Round 1 — Identify verification targets (X minutes):** [Instructions]

**Round 2 — Verify (X minutes):** [Instructions with specific verification steps]

**Round 3 — Report and discuss (X minutes):** [Class debrief protocol]

**Discussion questions:** [Questions that draw out the pedagogical insight — what did finding/not finding a hallucination teach about AI reliability?]

### Teacher Modelling Script

[Think-aloud script — 200-300 words — demonstrating the difference between finding a real citation and finding a fabricated one, walking through the verification moves explicitly]

**Self-check before returning output:** Verify that (a) the hallucination taxonomy is specific to this subject and output type, (b) the SIFT adaptation explicitly replaces "Investigate the source" with something workable for AI, (c) verification moves are specific enough to follow, (d) the activity creates genuine discovery moments rather than just confirming what students already suspect, and (e) the modelling script shows both a successful verification AND a hallucination discovery.

---

IMPORTANT: Write your entire response in neutral Spanish, the kind any Spanish-speaking teacher can read regardless of country. Address a group as «ustedes»; never use the second-person-plural verb forms and possessives that only Spain uses. Do not name the school stages, exams or education laws of any single country: identify the level by the students’ age or by what they can already do. Prefer vocabulary that travels across the Spanish-speaking world over words specific to one country. Use the register a secondary-school teacher would use with colleagues. Keep pedagogical terms in Spanish. Do not translate the names of cited academic frameworks or authors. Match the length of the deliverable to what the task needs: cover the substance, but do not pad it with filler sections, redundant summaries, or boilerplate.

Resaltado en ámbar: los valores que ocupan los huecos del prompt. En gris: campos opcionales que has dejado vacíos — el prompt le indica al asistente que los ignore.