AI prompts

After you have chosen the topic, before you go to the library. This is also the core of the Library Skills Task.

My research topic: [write your topic in one sentence]
The current state of the question: [write the question you have in mind, rough is fine]
The type and length of the text I will write: [e.g. ENG 102 research paper]

Do not fix this question for me. Make me do the narrowing:
1. Break my question into its parts: which concepts are in it, what exactly does each concept cover, where is it vague?
2. How many different questions is my question asking at once right now? Separate them one by one.
3. List the axes that narrow the scope: time, place, group, criterion, context. For each axis, ask me one question I have to answer ("which period", "for whom", "compared to what", and so on).
4. Would this question be any use in a library search? If not, why: too general, too narrow, or is its answer a preference rather than a fact?
5. For each concept, write 3-4 English synonyms that appear in the literature. Then build two Boolean search strings out of them: one narrow, one broad.

Do not suggest a source, an author, an article title or a citation, I only want the terms and the questions. Do not write the answer to my question either, do not write in my place, criticize what I wrote. If there is a field you do not know, say "I do not know this field", do not invent.

Doldur: your topic (one sentence), the current state of your question, the type of text you will write

Whether the Boolean string is right is settled by the search, not by the model: paste the string into Quick Search. Zero results means the term was invented, tens of thousands means it is too general, and both are visible in a minute. That is also the reason for not asking the model for sources: invented citations appear precisely when sources are requested.

When your thesis has taken its first shape, before the in class writing day and before the outline tutorial.

My thesis is this: [your thesis]
The assignment question is this: [copy the question from the task sheet word for word]
The texts I will use: [names of the texts]

I do not want you to fix this thesis. Ask me 5 questions that weaken it. Questions of this kind:
- Which word of it carries more than one meaning, so that I have not actually said what I mean?
- Which part of it is beyond dispute, so that nobody would object and it is not worth defending?
- Which claim of it answers a side issue rather than the question the assignment asks?
- Which piece of it needs evidence in my hands, and where in the texts should that evidence be?
- Someone who accepts this thesis is forced to accept which conclusion, and do I accept that conclusion?

After the questions, say one more thing: is this thesis balanced, or is it a one sided verdict? Give your reason in one sentence.

Do not give me a corrected thesis, do not propose an alternative thesis, do not say "it would be better if you wrote it this way"; do not write in my place, criticize what I wrote. Do not invent sources, quotations or citations. You did not read the texts, I only gave you their names, so do not give information about them, and say "I do not know" where you are not sure.

Doldur: your thesis, the assignment question (word for word from the task sheet), the names of the texts you will use

Not all five questions have to be good ones; if one of them finds the real weak point of your thesis, it did its job. The model did not read the assignment texts, you only gave it their names: if it produces a sentence like "this text says the following", that sentence is invented, so open the text yourself. If your syllabus does not allow GenAI in graded work, this use falls under that too, and the university guideline says do not use it when it is not permitted, ask your instructor if you are not sure, and keep a record if you did use it.

After the thesis has settled, before you write the counterargument paragraph.

My thesis is this: [your thesis]

Build the three strongest objections that could be raised against this thesis. I do not want weak or easily refuted objections; I want the one that will really give me trouble. For each objection write three things:
1. The objection itself, one paragraph, in the voice of someone who holds it.
2. What assumption does the objection rest on? Write the opposing assumption my thesis rests on as well.
3. What KIND of evidence would support this objection: what sort of study, measuring what, on which sample, or which kind of passage from which text.

Then say: which of the three does not merely limit my thesis but falsifies it?

Do NOT give a source name, an author name, a study name, a year or a citation, the answer to "what kind of evidence" is enough; invented citations appear precisely when sources are requested. Do not answer the objections for me, I will write the answers; do not write in my place, criticize what I wrote. Say "I do not know" for any fact you are not sure of.

Doldur: your thesis

Take the "what kind of evidence" answer in item three seriously, but then go after it yourself: search for that study in the library. If you find it, the objection is real and it goes into your paper; if you cannot, the objection is fictional and it gets dropped. The source you find is a gain too, it becomes the basis of your counterargument paragraph.

When a source lands in your hands, before you decide to use it.

I have a source in hand; I pasted its citation and first pages below. My research question is this: [write your question in one sentence].

I will decide whether to use this source. What I want from you is the questions I should ask while deciding:
1. What kind of source is this, a book, a peer reviewed article, an edited collection, a newspaper piece, a report, a web page? Say which part of the citation told you. Then: in THIS kind of source, what determines reliability?
2. What does the source's own claim appear to rest on, data, other sources, the author's interpretation? Write which sentence told you.
3. What can be understood about the author's position and the place of publication from the text I pasted? If it is not in the text, write "not in the text".
4. Given my question, where is this source useful and where is it not?
5. If I am going to use this source, what are 5 things I have to check in the source itself?

Do not decide whether the source is good or bad, do not try to summarize it for me, do not suggest other sources. Do not invent information I did not give you: if you do not know the author, the citation count, the journal's standing or its impact factor, say "I do not know". Do not write in my place, criticize what I wrote; I will judge the source, you produce the questions I should ask.

CITATION AND FIRST PAGES:
[paste the source's citation and its first few pages]

Doldur: your research question (one sentence), the source's citation + its first pages

The model does not know the source itself; it only sees what you pasted. Do not trust sentences like "this journal is respected" or "this author is well known in the field". Whether something is peer reviewed is told by Quick Search's "Scholarly (Peer Reviewed) Journals" filter, not by the model; and what got discussed after an article is told by the "Cited by" list in Scopus or Web of Science.

After you have collected the sources, before you move to the outline.

Below is the draft of my annotated bibliography: the citation for each source and the annotation I wrote.

Do not summarize the sources for me, do not write new annotations. Check these:
1. Does every annotation do three jobs at once, saying what the source does, evaluating it, and saying how it connects to my project? Name the one that is missing.
2. Which annotation has turned into summary only, with no evaluation? Which sentence looks like evaluation but is actually praise (like "a very important study")?
3. Is there any inconsistency of format in the citations, are they all in the same style? Count the citations with a missing field: author, year, journal, volume, page range, DOI, access date.
4. Is this set of sources made of the parts of a single argument, or are they writings independent of each other? Which two contradict each other; a contradiction is good news, so say which one it is.
5. What TYPE is missing from the set? (Are they all reviews, is there no primary study; are they all from the same country, the same decade?) Say it as a type, do not suggest a source name.

Do not complete a missing citation field, do not fill in a year or a page number as "it is probably this", write it as missing if it is missing. Do not suggest sources, do not invent citations; say "I do not know" where you are not sure. You did not read the sources, so you cannot know whether my annotation is correct; look only at the internal consistency of what I wrote. Do not write in my place, criticize what I wrote.

MY DRAFT:
[the citations + your own annotations]

Doldur: MY DRAFT, the citations and the annotations you wrote

The model did not read the sources: it cannot say "this annotation describes the source correctly", it can only see whether the annotation is incomplete in itself. Never have the model complete a missing citation field; the publisher, year, volume, page range and DOI are copied from the source's own citation page. The answer to the format question is in the library's own Annotated Bibliography guide.

After you have written the outline or the paragraph plan, before you move to prose.

Below is the plan of a paper: my thesis, what each paragraph will do, and which evidence I will use under each claim.

I do not want you to approve or fix the plan. Answer these four questions:
1. Which claim has no evidence under it? Count them one by one.
2. Which piece of evidence does not actually support the claim it sits under, and is merely related to the topic? Say why it does not support it in one sentence.
3. Which part of my thesis is not covered by any paragraph, and which paragraph has nothing to do with the thesis?
4. Is the order of the paragraphs logical for a reader? If not, which two should swap places, and why?

Then ask me 3 questions I need to answer: questions that make me see the weakest point of the plan myself.

Do not write a new plan, do not fix my thesis, do not fill in the missing evidence yourself, do not suggest sources. Do not invent a source, an author, a study name or a citation; say "I do not know" for anything you are not sure of. Do not write in my place, criticize what I wrote, I will fix the plan myself.

MY PLAN:
[your thesis + your paragraph plan + the evidence under each claim]

Doldur: MY PLAN, your thesis, your paragraph plan, the evidence per claim

The model saying "this evidence does not support it" is not the last word: sometimes the connection is in your head but not written in the plan, and then the problem is not the evidence, it is that the plan is underwritten. Also this prompt cannot know whether the evidence really exists, it only looks at what you wrote in the plan; the page number is yours to verify too.

When the first draft is finished, before you move on to fixing sentences.

Below is a text I wrote myself. I do not want you to fix it, I want you to map it.

For each paragraph in order:
1. What does this paragraph say in one sentence? If you have to use "and" in your summary, note that separately, it means the paragraph is saying two things.
2. Which is the paragraph's main sentence? Does the rest serve that sentence, or does the subject drift halfway through?
3. How does it connect to the previous paragraph: continuation, contrast, example, or no connection at all?
4. If I deleted this paragraph entirely, what would the text lose? If the answer is "nothing", write that.

At the end say three things: how many separate subjects are in the text; which two paragraphs would improve the text if they swapped places; which paragraph I need to split and at exactly which sentence.

Do not write new paragraphs, do not rewrite my paragraphs, do not give a sample text as "this would be better", do not write in my place, criticize what I wrote. Do not assume anything that is not in the text; if there is a sentence you do not understand, say "I did not understand this sentence". Do not invent sources, quotations or names, and say "I do not know" where you are not sure.

MY TEXT:
[paste your text here]

Doldur: MY TEXT, your whole draft or the section you want checked

This prompt does not look at content: whether your thesis is strong or your evidence is enough is not answered here, only the structure is. And the model saying "this paragraph is unnecessary" is a warning, not a decision, because the paragraph that looks unnecessary is sometimes the most personal part of the piece. The decision to delete is yours.

On the mechanical pass before submitting, or when there is a mark in your instructor's feedback you do not understand.

Below are English sentences I wrote myself. For each sentence, do this:
1. Is there an error in this sentence? If there is, NAME the error (subject-verb agreement, dangling modifier, comma splice, run-on, article, preposition, tense shift, unclear pronoun reference, wordiness, and so on).
2. Explain in plain English why it is an error, in one sentence.
3. Write the rule behind the error in one sentence, stated independently of my sentence.
4. Do NOT give me the corrected sentence. Instead, show me which word or which ending I need to touch, and leave the decision to me.

For a sentence with no error write "no error", do not strain to find one. Separate a style preference from a rule violation and say which one it is. At the end add this: is there a repeating error pattern in these sentences? If there is, name it, because that is what stops me making the same error again.

Do not write in my place, do not rebuild the paragraph, do not give an alternative text as "the academically appropriate version"; criticize what I wrote. Do not invent rules or sources; if you are not sure about a usage, say "I do not know, check it in a dictionary".

MY SENTENCES:
[paste your sentences here]

Doldur: MY SENTENCES, your own English sentences

There is a practical reason for not asking for the corrected sentence: in ENG 101 all of Essay 2 and the three hour final are written in class, without a dictionary. The only thing with you there is knowing the rule. If your syllabus says "GenAI cannot be used in any graded work", this pass falls under that too; make the decision knowingly, and ask your instructor if you are not sure.

When the draft is finished, before you submit. If there is even one phrase in quotation marks in your text, do not skip this pass.

Below is the draft I wrote, and under it the relevant parts of the source text I quoted from.

Your only job is this: take EVERY phrase inside quotation marks in my draft, one by one, and search for it word for word in the source text. For each one, write:
- the quotation (as it stands in my draft)
- word for word in the source / similar but different / could not find it
- whatever the difference is, say it letter by letter: has a word changed, has a singular slipped to a plural, is the punctuation different, has the order changed, is it missing entirely
- the full text of the version you found in the source

Then two more lists:
1. Which sentences in my draft read like somebody else's idea even though I gave no source?
2. Where did I paraphrase without quotation marks, and does the source actually say that?

Do not write that something is in the source text when it is not, do not fill gaps from memory; if you cannot find it, write "could not find it in the source", do not guess. Do not invent a source, an author, a citation or a page number, and say "I do not know" if you are not sure. Do not fix the quotations, do not rebuild my draft, do not write in my place, criticize what I wrote: show me where it does not match the source, that is all.

SOURCE TEXT:
[paste the full text of the pages you quoted from here]

MY DRAFT:
[paste your draft here]

Doldur: SOURCE TEXT, the full text of the pages you quoted from, MY DRAFT, your own draft, with its quotation marks

This prompt only works if YOU paste the source text. Ask "is this quotation right" without giving the source and the model invents from memory. Exactly that was tried on an HCIV assignment: all three quotations the model gave failed to match the source word for word. In one a single letter had changed (favorable / favourable), in one a singular had slipped to a plural (society / societies in Marx), and one was a number with no source behind it at all. None of them looked wrong while reading; the only reason they got caught was that every quoted phrase was searched in the source text. The model saying "it is there word for word" is not evidence either, so Ctrl+F the place it says it found yourself.

Before the oral presentation, the proposal talk or the end of term interview.

Below is a summary of my own research: my question, my thesis, the sources I used and my main findings.

Do not write a presentation script for me. Be the instructor sitting across from me and ask questions:
1. Assuming you did not understand my question and my thesis, ask 5 clarifying questions.
2. Ask 5 questions that test whether I really read my sources, of the kind "who did that study measure", "which part of your claim does its finding support".
3. Ask 3 hard questions that probe the method and the chain of evidence.
4. When a question comes that I do not know the answer to, what should I say? Tell me the right form of that.
5. Is the thesis you hear from this summary the same as the thesis I think I wrote? If it is different, write the difference.

Do not write the answers to the questions, I will answer them. I do not want a presentation draft, a speech script or a slide plan either. Do not invent information about my sources that I did not give you, you did not read them; do not add a quotation, a citation or a finding, and say "I do not know" where you are not sure. Do not write in my place, criticize what I wrote.

MY SUMMARY:
[your research question, your thesis, your sources, your findings]

Doldur: MY SUMMARY, your research question, your thesis, your sources, your findings

These questions are not the questions of the real exam, they are a rehearsal. Their value is here: in ENG 102 a significant part of the grade is oral, and that part is not text checking, it measures whether you know the sources you read. Do not write the answers down and read them, answer out loud; the place where you get stuck shows up while speaking, not while writing.