Units
1. Artificial Intelligence in Writing and Editing: Roles, Boundaries, Author Voice, and Verification 2. From Idea to Draft: Structure, Content Plan, and the First Draft in Nonfiction 3. Plot Outline: Premise, Genre, Scene, and Narrative Voice 4. Character Development: Depth, Consistency, and Character Arc 5. Plot and Structure: Act Structure, Suspense, Pace and Stage Cards 6. Proofreading and Proofreading: Grammar, Consistency, and Copy Editing 7. Style and Author's Voice: Preservation of Tone, Style, Rhythm and Voice 8. Summary, Synopsis and Promotional Texts: Blurb, Logline and Presentation 9. Translation and Localization Support: Meaning, Tone and Cultural Adaptation 10. Research and Fact-Verification: Risk of Sources, Quotes, and Fabricated Sources 11. Originality, Copyright and Ethics: Avoiding Plagiarism, AI Copyright, Transparency and Over-Dependence
Unit 10 / 11

Research and Fact-Verification: Risk of Sources, Quotes, and Fabricated Sources

Gains:

  • Ability to use artificial intelligence as a starting map in the research process and connect each fact and quote to the primary source
  • Ability to recognize the risk of fabricated sources, fake quotes and false dates (hallucinations) and apply systematic verification
  • Ability to edit research notes, bibliography and fact checklist with artificial intelligence support and maintain the reliability of the article

The invisible backbone of good writing is research. A date, a statistic, a quote, a scientific finding—these carry the credibility of the text. A single false fact can destroy the credibility of an entire work. Artificial intelligence (AI) seems seductively useful in the research process: it provides an immediate, fluid, confident answer to every question. But herein lies authorship's most dangerous trap: AI can produce sources, quotes, dates, and findings that do not actually exist, as if they were true. We can summarize this unit in one sentence: AI can be the starting map for research, but it can never be the final resource.

Hallucination: the most insidious risk

Hallucination is when AI presents non-existent information in a safe and fluent language. According to research, the most dangerous forms of this are:

Fabricated source: AI can fabricate a book, article, or report that does not actually exist; Moreover, with realistic author name, year and journal information. "Yılmaz, A. (2019). Digital Literacy. Journal of Communication, 14(2)." — sounds real, but might not be.

Fake quote: AI can attribute a quote to a real person that they never said. Gives the name of a famous writer and a sentence that that writer has never uttered.

Incorrect date/number: May give incorrect date of an event, a statistic, a percentage; all with the same precise tone.

Confused fact: May confuse two true facts and produce a false combination (such as attributing one person's statement to another).

What these risks have in common: they all look fluid and convincing. Fluency is not accuracy. The only protection is systematic verification linking to the primary source.

Caution: Ask the AI ​​"is this source real?" Asking is NOT a method of verification. AI can also confirm the source it made up as "yes, it's real". Verification is done outside the AI, at the primary source: opening the journal, the book, the database itself and finding it.

Legitimate research role of AI

Despite all these caveats, AI is not worthless in research; only its role is limited. Legitimate uses:

Starting map: Determining where to start with a topic, what subheadings there are, and what concepts need to be researched. These are lists of things to look for, not answers.

Concept clarification: Understanding an unfamiliar term firsthand (to then verify it from the source).

Question generation: "What questions should I ask about this, what might I have missed?" AI is good at pointing out points of blindness.

Editing: Summarizing, categorizing, and formatting the bibliography of your collected (verified) notes.

The table below summarizes what AI is and isn't good for in research:

Quest

Is AI suitable?

note

Introduction map to the topic

Yes

Just direction, not answer

Generating questions to search

Yes

Shows blindness points

Provide source/citation

no

It can be made up; verify from primary

Export history/statistics

no

It may be wrong; buy from source

Edit verified notes

Yes

You collect, AI organizes

Systematic verification flow

One rule for every fact: go to the primary source. A primary source is where information is first published: original research article, official statistical agency, one's own book/interview, original document. What the AI ​​(and even secondary summaries) say is a "claim" until verified, not a "fact".

Verification flow:

  1. Mark: As you write, mark each fact, date, quote, and number with the [VERIFY] tag.
  2. Find primary source: Open the source yourself for each tagged item; Is there any, does its content match what is said?
  3. Cross-check: Verify critical facts from at least two independent reliable sources.
  4. Keep records: Note the source of each verified fact; In publication, the bibliography comes from this note.

You can use AI in the editing part of this flow:

Context: Below are the notes I have collected and VERIFIED from the PRIMARY source, with the source written next to each note. Task: Group these notes by theme and suggest a title for each group. DO NOT ADD any claims without a source; Facts, dates or quotes that I did not give are FAKE. Mark the missing parts as "the source is missing here".

You can also use AI as a "skeptic":

Task: List ALL factual claims, dates, numbers, and quotations that need to be verified in the text below. For each one, write down "what type of primary source should be verified?" DO NOT answer it yourself; Only the verification list appears.

Primary, secondary and tertiary source

Knowing the source hierarchy in validation makes your job clear. A primary source is where information is produced firsthand: the original research article, the official statistical table, one's own book, the court decision, the original document. A secondary source is the text that interprets or conveys the primary: a news report, a review, an encyclopedia article. Tertiary sources are summaries that compile secondary sources. The output of the AI ​​is like a tertiary digest at best; Most of the time it is even weaker than that, because it cannot show its source and can make it up. Rule: the more critical a fact, the more primary you should go. A secondary source may be sufficient for the atmosphere of a novel; A primary source is required for a health claim, a date, or a statement attributed to a person. Use AI on the "where should I look" layer at the bottom of this pyramid, never on the "this is the truth" layer at the top.

Tip: When placing an unverified fact in your text, leave a visible [VERIFY] tag next to it and do not submit the text with any sentences bearing this tag. Labels are a simple but life-saving habit that makes forgotten facts visible, saying "I'll look at them later".

three mini cases

Case 1 — Fake article. An author wrote that YZ gave "2020, Journal of Media Studies, Demir et al." put the source directly into the bibliography. A reader found that such an article does not exist. The author had to re-verify the entire bibliography from the primary, resulting in two more apocryphal sources. Lesson: no source is included in the article without being found in person.

Case 2 — Correct use. A journalist started a complex topic with AI: extracting subheadings and questions to ask. Then he collected each fact from the official statistics agency and from one-on-one interviews; He used the AI ​​only to edit notes that he verified. The writing came out fast and solid; Not a single fabricated fact was passed.

Case 3 — Fake quote. One speaker included in his presentation an "inspiring" quote attributed to a famous AI scientist. The promise was made up; someone in the audience noticed and the speaker's credibility was damaged. The correct way: was to verify the quote from one's own book/speech.

Weak prompt / Strong prompt

Weak prompt:

Give me 5 academic sources and one quote on this subject.

This prompt invites the AI ​​to fabricate resources; The output looks realistic but is likely unverifiable.

Powerful prompt:

Your role: research direction assistant. Topic: [topic].Task: There are 6 SUB-HEADINGS that I need to research on this subject and 2 questions that I need to ask for each. Also, direct me to "what kind of primary sources (official statistics, peer-reviewed journals, one-on-one interviews) can I search?" Just show me where to look.

Difference: AI is asked for direction, not answers; The door of fabrication is closed, verification is with the author.

Common mistakes

  • Using the resource provided by AI without verifying it. The fabricated source becomes a lie published with your signature.
  • Asking the AI ​​"is this true" and trusting its answer. AI can also confirm its own invention.
  • Not verifying quotes from one's own source. Fake quotes destroy credibility.
  • Being satisfied with a single source. Critical facts require cross-checking from at least two independent sources.
  • Thinking of AI as the "last resource" rather than the "starting map". AI shows direction; The primary source gives the fact.

In summary

In research, AI is a powerful starting map and organizer: extracting subheadings, generating questions, collecting verified notes. But the job of providing sources, quotes, dates and statistics can never be trusted; because he can make it all up convincingly (hallucination). Golden rule: connect every fact to the primary source, cross-check the critical, do not use any source given by the AI ​​without finding it yourself. AI tells where to look; You decide what you find.

Application task

Choose a topic. First, get subheadings and questions from the AI ​​with the “direction” template (NOT the source). Then deliberately ask the AI ​​for a source/quote and try to verify that source as primary: does it really exist, does its content match what is said? Note your findings (how many sources were verified, how many were not found). Finally, make a verification list of your own text using the "skeptic" template.

checklist

  • [ ] Did I use AI to generate direction and questions, not for resources?
  • [ ] Have I marked every fact, date, quote, and number with [VERIFY]?
  • [ ] Have I personally found and verified each source at the primary source?
  • [ ] Have I cross-checked critical facts against at least two independent sources?
  • [ ] Have I verified the quotes with the individual's own source?
  • [ ] Have I explicitly given the "source/citation fabrication" rule to the AI?
Previous Next