1Phase 1 · The Setup
Module 2 · Prompt Like a Pro
Integrity Focus

Hallucinations, Outdated Info & Generic Slop: How Claude Actually Fails

Lesson 2.4 5 screens · the verify habit

Claude doesn't know that it doesn't know.

Claude will produce a confidently-worded, beautifully-formatted, completely false answer and have no idea it just did. It does not feel embarrassment, it does not flag uncertainty unless you ask, and it cannot tell the difference between a fact it learned from a textbook and a plausible-sounding sentence it made up to complete the pattern.

This isn't a flaw to fear. It's a quirk to plan around. Claude is genuinely useful when you treat its output as a confident first draft from a smart roommate who reads a lot but sometimes makes things up. Useful, fast, often right: but never the last word on anything you'll be graded on or held to.

The three pitfalls every student needs to know

  • Hallucinations: Claude makes up facts, quotes, citations, statistics, names, and dates. Confidently.
  • Outdated info: Claude's training has a cutoff. Anything that happened after that date, it can't really know: but it will guess with information that sounds current.
  • Generic slop: Claude's default voice is the average of everything it read. Without specifics, you get the same paragraph every other student got.

Pitfall 1: Hallucinations: confidently invented facts.

Most likely places Claude hallucinates on student work: specific quotes from books or articles, citations and page numbers, statistics, the existence of obscure papers, historical dates of less-famous events, scientific values that "sound right," and anything where the right answer is one specific number or name.

🚨

Five red flags: assume Claude might be making it up

(1) A specific date, statistic, or page number that wasn't in your source material. (2) A quote from a book, attributed by chapter or page. (3) A citation in any format (Claude will produce real-looking fake citations). (4) An obscure historical figure or event. (5) An exact number used to support an argument.

Five safer zones: usually fine

(1) Conceptual explanations of well-known topics. (2) Outlining and structure. (3) Brainstorming. (4) Editing your own writing. (5) Simplifying jargon. Even these need a quick gut-check, but the failure rate is much lower.

The "show me the receipts" follow-up: paste right after any factual answer
For the answer you just gave me, label every claim with a confidence tag: [CONFIDENT]:well-known and you'd bet on it. [GENERAL]:broadly true but I should verify the specifics myself. [INFERRED]:you're reasoning from related knowledge but didn't directly read this. [UNCERTAIN]:you're not sure; this could be wrong. If anything was a specific date, statistic, page number, quote, or citation: flag it as [VERIFY]. Don't tell me a citation is real unless you'd stake the answer on it. If you can't tell, say so. Do not rewrite the answer. Just relabel.

Why this works better than "is that true?"

Asking "are you sure?" makes Claude people-pleasing: it'll often double down or vaguely agree it might be wrong without telling you which part. Asking it to tag claims forces a real assessment. You'll often see [INFERRED] tags on things Claude initially stated like fact. That's gold: you now know exactly which sentences to verify before you use them.

Pitfall 2: Outdated info: the cutoff problem.

Claude was trained on data up to a specific date: its "knowledge cutoff." After that date, Claude can't really know what happened: but unless you ask, it'll often answer in present-tense voice that sounds current. This is how students end up writing about "the current Speaker of the House" with the wrong name, or "recent" events from two years ago, or a startup that already shut down.

The cutoff check: paste at the top of any "current events" or "recent" question
Before you answer: what's your training data cutoff date? Today is [today's date]. If my question depends on anything that may have changed between your cutoff and today (recent news, current officeholders, new policies, recently released products, current statistics, who's CEO of X), do TWO things: 1. Answer based on what you knew as of your cutoff, and clearly label it: "As of [cutoff], …" 2. Tell me what specifically I should re-verify against a current source before using your answer. If my question doesn't depend on recent info, ignore this and answer normally.

Topics where the cutoff trips students up

  • Anything in current events: politics, recent legislation, court rulings, who holds what office.
  • Application info: "What's [school's] current acceptance rate?" / "What's the SAT requirement now?":verify on the school's actual site.
  • Tech and startups: companies pivot, get acquired, or fail; "current CEO of X" is volatile.
  • Statistics: "the latest data shows" is almost always not the latest data.
  • Pop culture: what's #1 right now, who won X, new releases.

Pitfall 3: Generic slop: the smoothed-out-average voice.

Generic slop isn't a hallucination. It's not technically wrong. It's just an answer that could have been written for any student, in any class, at any school. The kind of cover letter a recruiter has read forty versions of by lunch. Generic slop is the most common Claude failure: and the most fixable.

Symptoms of slop

Sentences that start with "It is important to note that…" or "In today's society…" or "Throughout history…" Three-item lists for everything. Words and phrases like "leveraging," "delve," "navigate the complexities of," "robust framework," "multifaceted approach." Conclusions that begin "In conclusion." Confident-sounding sentences with zero specifics.

If you read the paragraph aloud to a friend and they could not guess what class it was for, it's slop.

The fix

Slop is almost always caused by under-briefed prompts. The five upgrades from 2.2 fix it: more context, examples (good and bad), tone (named specifically), length, and format. The follow-up #9 from 2.3 ("you're being vague: quote your three most generic sentences and replace each one with something specific") does the cleanup pass.

Module 4.6 has the deeper AI-pattern workshop: the surgical phrase-level scrub. Today, just learn to spot slop and reflexively follow up.

The de-slop follow-up: paste when an answer feels too generic
This answer is too generic. Read it back to yourself and identify: 1. Every sentence that could have been written for any student in any class: quote them. 2. Every "default essay phrase" you used (e.g., "throughout history," "in today's society," "delve into," "leveraging," "multifaceted"). 3. Every claim that has zero specific evidence behind it. Then rewrite ONLY those sentences with specifics that match my actual situation. Keep the sentences that were already specific. Don't expand the answer; just sharpen it.

Five categories where AI is worse than the textbook.

The other half of "how Claude fails" isn't a pitfall in Claude:it's a category of work where reaching for AI at all is the wrong move. Five places where the textbook, your prof, or your own slow handwriting beats Claude every time.

When to put Claude down

  • 1 · When you need the exact wording. AI paraphrases. The textbook is the official source. Your prof's exam may use the textbook's exact phrasing of a definition. Look it up there; use Claude to quiz you on it.
  • 2 · When the field has known facts and AI hallucinates them. Specific dates, named scholars, exact quotes, court cases, study results, drug dosages. AI confidently makes these up. The textbook, JSTOR, your prof's handout: those are the sources. Module 5 (Research) is the deep dive.
  • 3 · When your prof is the source of truth, not the field. Upper-level seminars where the prof has a particular framework. AI knows the field, not your prof. Trust the lecture, the slides, and office hours.
  • 4 · When the skill IS the slow human process. Sentence diagramming. Working through a proof step by step. Translating word by word the first time. Do the slow human version first; bring AI in after to check.
  • 5 · When the test is whether you can do it without AI. In-class essay, blue-book exam, no-tech lab practical, interview, whiteboard problem. If the assessment is closed-book, your prep should be closed-book too: at least some of it.

The quick decision before you open Claude

You don't need to memorize the five categories. Just one mental check: "Is the goal of this study session to understand the material, or to build the skill of producing this work without help?" If it's understanding: Claude is excellent. If it's the skill of producing under pressure: close Claude. Set a timer. Do a mock attempt with no help. Bring Claude back AFTER, to grade and find what to drill.

The verify habit: what 2 minutes of checking saves you.

Run the 2-minute verify pass on anything Claude gave you that you're going to submit, send, or rely on.

The 2-Minute Verify: run this before you submit anything

  • 1 · Hallucination check. Run the "show me the receipts" prompt. Investigate any [VERIFY] flags. Google the citations or stats.
  • 2 · Cutoff check. Anything time-sensitive? Open a tab, Google the specific current claim.
  • 3 · Slop check. Read one paragraph aloud. Could it have been any other student's? If yes, run the de-slop follow-up.
  • 4 · The "would I be fine if a prof read this and asked me about it" check. Could you defend every line? Do you understand the argument?
  • 5 · The save. Save the chat. Rename it to something you'll find later (Module 1.2). The transcript is your process trail.

Try this: run a hallucination test on your hardest subject

Pick the subject where you'd be most reliant on Claude getting facts right (history, lit, biology, anything specific).

1. Ask Claude a hard, specific question in that subject: one that requires a date, a name, a statistic, or a quote. ("What was the casualty count at [specific battle]?" "What's the exact line from [specific scene of the play]?")

2. Get the answer.

3. Run the "show me the receipts" prompt.

4. Google the answer.

You might find that Claude's confident answer was wrong.

You finished Phase 1. Up next: Module 3: The Study System.

Setup, prompts, follow-ups, verify habit: Phase 1 is done. Phase 2 is "The Wins":the four flagship modules where Claude actually saves you serious time on real schoolwork.

Continue to Module 3 →