Gains:
- Ability to apply the discipline of using artificial intelligence for a conceptual map and not as a resource catalogue.
- Ability to verify each imprint produced by artificial intelligence in an independent catalog and eliminate fake sources
- Being able to represent each view in its strongest and honest form by having the synthesis done with real texts provided by oneself as much as possible.
Every philosophical and ethical study is built on a literature. Literature review (the study of mapping the field by finding and reading previous key works, articles, and discussions on a topic) is the basis of the work of a thesis writer, a referee, or a lecturer preparing a course. An original contribution cannot be made without knowing who says what on an issue, which camps have formed and where the debate is tied up. This job is traditionally a tiring, messy labor that takes weeks. AI can speed up the mapping and synthesis of this scan — but in human domains there is a very special and dangerous pitfall here: fake source and fake attribution.
In this unit we will see how to confidently use AI as a literature assistant, how to eliminate the most devastating mistake of imaginary sources, and how to honestly synthesize different views.
The biggest danger: fake sources
We put this at the top because it is the number one bane of using AI in human fields. When you tell an AI to “list key articles on this topic,” the model often presents a mix of real articles and articles that never existed: a made-up article under the name of a real author, a volume/issue not in a real journal, a fake DOI number that looks right. These look so convincing that even experienced researchers cannot distinguish them at first glance. A single spurious citation in a thesis or article can destroy the credibility of the entire work and lead to an academic investigation.
This is because of how AI works: the model is not a “citation database”; It has learned "what a citation looks like" from text patterns and produces a realistic-looking citation, regardless of whether it is real or not. For this reason, no source provided by YZ can be used without being verified verbatim in an independent catalog (library, peer-reviewed database, publishing house page).
Caution: An AI-generated resource is not real just because it "looks too realistic". The author name, journal, and year may be familiar, but the article itself may never have been written. Never use each source without finding it in a reliable catalog other than YZ and matching its citation exactly.
The right workflow: source first, synthesis later
The order of safe literature study is important.
1. Use AI for the map, not the catalogue. Ask AI “what are the main camps of this debate, which concepts are central, which distinctions are important?” ask. This conceptual map usually works. But take the "give me source list" output as a starting point, not as a definitive list.
2. Verify each resource against the individual catalogue. Search every work recommended by YZ in a reliable database. Remove the resource you cannot find from the list. Also correct the citation (author, title, year, journal, page) of the existing source in the catalogue.
3. Read the sources yourself, then have them synthesized. The safest synthesis is to have the AI summarize and compare the actual texts you provide. Thus, AI cannot add fabrication from outside; It only works with the actual text you provide.
4. Represent views honestly in synthesis. Present the different camps at their strongest; Do not caricature an opinion.
Tip: When you have the AI do a literature summary, instruct it to “only rely on the texts I give you, add authors or articles from your own memory.” This largely cuts out the extraneous hallucination and ties each claim to the text you provide.
three mini cases
Case 1 — Fabricated attribution caught. A graduate student asked AI for the "8 most cited articles" on a topic. The student searched the library database for each one: 3 of the 8 sources did not exist at all — made-up articles under the name of real authors, articles that looked true but did not exist. The student eliminated these 3 and corrected the remaining 5 with their real ID tags. The verification step prevented 3 ghost citations from entering the thesis.
Case 2 — Given the text, the synthesis was safe. A lecturer collected 6 real articles as PDFs himself and had the AI compare only these texts and create a theme table. The output was ready in 30 minutes and contained no dummy sources because the AI only worked with the real texts provided. The lecturer still linked each theme reference back to the corresponding article.
Case 3 — Vision distortion corrected. A researcher had AI summarize two opposing ethical positions. AI presented the minority position in a weak and stereotypical way. When the researcher turned to the texts of the original proponents of that position, he found a much stronger argument and corrected the synthesis. AI's tendency to portray the "popular as strong and the minority as weak" has been balanced by honest representation.
Weak prompt / Strong prompt
Weak prompt:
Give your top 10 sources on this topic and their summaries.
This prompt is a fake source factory: the AI generates realistic but fake identifiers from its memory and you trust the list before you can verify it.
Powerful prompt:
Your role: careful literature assistant. We will do TWO STAGES. Stage 1 (map): Draw out the main camps, central concepts, and key distinctions of this discussion. DO NOT PROVIDE SOURCE CREDITS; just give a conceptual map. Phase 2 (synthesis): We will compare the actual texts that I will upload to you. Just rely on the texts I provide, do not add authors/articles from memory. Link each claim to the relevant text. Topic: [insert topic here]
This request prevents the production of fake imprints; It gets the conceptual map from the AI and the real resources from you.
Four copyable templates
1) Conceptual map (without sources):
Identify the main camps of the [topic] debate, the central claim of each camp, and the main points of distinction between the camps. Source citation/CITATION; Just give a conceptual map. If you are not sure, mark the distinction as "must be verified".
2) Synthesis table of given texts:
Below I give you N real texts. Based solely on these; adding external resources. A table appears: rows = comparison theme, columns = position of each text in that theme. Reference the name of the relevant text in each cell. Texts: [paste texts]
3) Source verification checklist production:
For each citation in the source list below, a checklist appears in the individual catalog of the fields I need to verify: author, full title, year, journal/publishing house, volume/issue, page, DOI. YOU verify the authenticity of the source; List only the areas I will check. List: [paste sources]
4) Honest representation control:
In the synthesis below, check whether each position is presented in its strongest form or in a weak caricature. Has a minority or unpopular opinion been unfairly weakened? Highlight where fairer representation is needed. Synthesis: [paste text]
Source trust level table
Where the source comes from
Trust level
What to do
Imprint produced from AI memory
too low
Verify or discard exactly in the catalog
The text you provide that AI summarizes
high
Link the claim back to the text
Peer-reviewed database/library catalog
high
Use the imprint literally
Publisher / magazine official page
high
direct reliable
Common mistakes
- Relying on the AI's resource list. The most devastating mistake; Each imprint cannot be used without being verified in an independent catalogue.
- "Looks realistic" test. Just because a reference seems convincing does not mean it is true; Made-up imprints look just as believable.
- Accepting DOI without verifying it. AI can produce fake but formally correct-looking DOI; Parse the DOI and see if you've reached the real article.
- Caricature the view. AI tends to present the minority position as weak; You provide honest representation.
- Not setting external resource limits. Asking for a synthesis without saying "just rely on the texts I give you" opens the door to external fabrications.
In summary
Literature review is the foundation of philosophical work, and AI can speed it up — but the most dangerous trap in the humanities is fabricated sources and citations. The safe path is clear: use AI for the conceptual map, not the resource catalogue; verify each imprint verbatim against an independent database; have the synthesis done, preferably with actual texts you provide; and honestly represent each view at its strongest. "Looks realistic" is not a verification. Source integrity is the guarantee of your academic credibility.
Application task
- Extract camps on a topic from AI with a "conceptual map" template (without sources).
- Also have the AI produce 6 resource tags.
- Search for these 6 tags one by one in a reliable catalogue; Determine how many are real and how many are fake.
- Read 2 of the real ones yourself and have AI create a synthesis table.
- Check whether a view is fairly represented in the synthesis.
checklist
- [ ] I used AI for the conceptual map, not as a resource catalogue.
- [ ] I have verified each source citation verbatim in the independent catalogue.
- [ ] I removed the fake sources from the list.
- [ ] I had the synthesis done with real texts that I provided as much as possible.
- [ ] I linked each claim back to the relevant actual text.
- [ ] I represented each view in its strongest and honest form.