History Essays and AI Detection: Your Footnotes Are Not the Part Being Scored

Chicago notes-bibliography style pulls every source detail out of your sentences and into footnotes, which leaves your body text as continuous narrative prose with no parenthetical interruptions in it. Turnitin says its AI writing report only measures prose sentences, and separately that false positives can turn up in text with little structural variation. A history essay hands it a large, clean block of exactly that. Meanwhile the shortened notes at the bottom of every page, the part that repeats most visibly, are short non-sentence structures, and the vendor publishes nothing at all about how it treats them.

Before going further: none of this comes from Turnitin talking about Chicago, because it does not. We read the AI writing detection FAQ end to end in a browser on 27 August 2026, saved 30,987 characters of rendered article text, and searched it. Every term a history student would look for returns nothing.

• `Chicago` 0 · `Turabian` 0 · `footnote` 0 · `endnote` 0 · `bibliograph` 0
• `citation` 0 · `quotation` 0 · `quoted` 0 · `style guide` 0 · `discipline` 0
• `primary source` 0 · `archive` 0 · `historian` 0. The single hit for `history` is in a sentence about the history and structure of society, not about the subject you study.

The search was working. On the same saved text, `qualifying text` returns 4, `word probability` 2, `structural variation` 1, `false positive` 13. So the connection drawn below is ours, assembled out of two bodies of documentation that do not acknowledge each other's existence. `curl` returns 403 on that host, which is why the read had to go through a browser: a negative produced with `curl` there would be manufactured.

Open any page of a notes-bibliography paper and you are looking at three kinds of writing that happen to share a sheet of paper.

The first is your argument. In Chicago style it runs without a single `(Smith 2019)` in it, because the citation has been moved out. The Turabian tip sheet for footnotes states the mechanism in one line: "For note numbers in the text, use superscript." A raised digit, and nothing else. The sentence closes normally.

The second is the material you transcribed: block quotations from letters, memoirs, depositions, pamphlets. Somebody else's sentences, reproduced.

The third is the apparatus. The tip sheet again: "Each footnote should appear at the bottom of the page that includes its numbered in-text reference."

Now hold that against the vendor's definition of what its AI writing report reads at all:

"This qualifying text includes only prose sentences, meaning that we only analyze blocks of text that are written in standard grammatical sentences and do not include other types of writing such as lists, bullet points (short non-sentence structures), or other non-sentence structures."

Read the line that follows it as well, because it changes what the number means: "This percentage is not necessarily the percentage of the entire submission."

Your argument prose is standard grammatical sentences with nothing else mixed in. It is about as unbroken a run of qualifying text as an undergraduate ever hands over. The apparatus is not sentences, or mostly is not, and by the definition above it looks like it falls outside. That last step is our reading of the published definition, not something the vendor demonstrates anywhere, and the honest state of it is in the counts above: `footnote` 0, `endnote` 0. There is no published position on footnotes to appeal to.

One neighbouring fact does have a source. A 2023 release note reports a fix for highlighting that had been landing on references listed in a bibliography, and states: "Bibliographies are now excluded when processing the AI writing report." The same entry adds that a paper has to be resubmitted for the change to apply to it. A Chicago bibliography is a bibliography, so those alphabetised entries are out of the measurement. For the general rule see What Is "Qualifying Text" in Turnitin AI Detection?, for the wider list see What Content Does Turnitin AI Skip?, and for what that exclusion does and does not cover, Why is Turnitin flagging my references and citations?.

Chicago's free citation guide opens the notes-bibliography examples like this:

"Sample notes show full citations followed by shortened citations for the same sources."

Turabian's version of the same page says it the same way: "Sample notes show full citations followed by shortened forms that would be used after the first citation." The ninth edition gives the rule its own section, headed "Short Forms for Notes." And the Chicago Style Q&A, answering on the eighteenth edition, states it as flatly as it can be stated: "all citations after the first can be in shortened form."

What that produces looks like this, straight out of the published examples in the two quick guides (rows one and three are Chicago's, row two is Turabian's):

A chapter that works one manuscript collection hard will carry a long run of those, each differing from the last by two or three digits. Chicago's own footnote tip sheet, which happens to use a history paper about German migration to Texas as its worked example, prints note 24 in full as: `24. Rosa Kleberg, 297–98.` Name, page range, done.

Set that beside the one sentence in the vendor FAQ that names what its own mistakes look like:

"Sometimes false positives (incorrectly flagging human-written text as AI-generated), can include content without a lot of structural variation, text that literally repeats itself, or text that has been paraphrased without developing new ideas."

The line that comes next is the one to write down before any meeting: "If our indicator shows a higher amount of AI writing in such text, we advise you to take that into consideration when looking at the percentage indicated."

Here is where the intuition goes wrong, though. "My paper repeats itself" is true of a Chicago essay, and the repetition is almost entirely in the notes, which are the part with the weakest claim to being measured. When an essay of this kind comes back highlighted, the highlighting is on the narrative. Shortening your notes further, or reverting them to full citations, changes the apparatus and leaves the measured text exactly as it was. It also breaks the style.

Turabian's ninth edition puts quotations in chapter 25 and gives section 25.1 the title "Quoting Accurately and Avoiding Plagiarism." Modification gets its own section, 25.3. The layout rule sits in the tip sheets: "Quotations of five or more lines should be indented as a block," single spaced, indented half an inch, with a blank line before and after.

Five lines. That is a low bar for a deposition or a committee minute, which is why history essays carry more block-quoted text than most disciplines.

Chicago's own sample page shows what accurate reproduction looks like in practice. The block quotation in the footnotes tip sheet reproduces Robert Kleberg's 1876 recollection with its period capitalisation intact: "I wished to live under a republican form of Government, with unbounded personal, religious and political liberty, free from the petty tyrannies and the many disadvantages and evils of the old countries." Capital G on Government, because that is what the source says.

The discipline's own standards statement is blunter about why. The American Historical Association's Statement on Standards of Professional Conduct says that "Any changes to a primary source or published secondary work, whether digital or print, should be noted," and that "A basic rule of good note-taking requires every researcher to distinguish scrupulously between exact quotation and paraphrase."

So a substantial fraction of the prose sentences in a history essay are prose sentences you are not allowed to alter. And here is the gap worth knowing about before anyone advises you to "vary" them.

Turnitin does document an exclude-quoted-text filter. It belongs to the Similarity Report, not to the AI writing report, and the two numbers are calculated over different things entirely. We went through the Similarity side in Why your quoted and cited text still shows up in the Similarity Report, and set the two reports next to each other in Turnitin AI Writing Report vs Similarity Report. On the AI writing FAQ page, across all 30,987 characters, `quotation` and `quoted` both return zero. There is no published statement that quoted material is removed from the AI writing measurement.

That article also carries a fact worth reading against your own footnotes: the Similarity Report's citation-recognition model was trained on four named formats, and Turabian is one of them, while Chicago author-date is not.

This is the part that changes what you do next.

Chicago and Turabian specify the apparatus, the layout and the treatment of sources. They do not specify how you argue. There is no tense table in Turabian the way there is in some other manuals. There is no required content list telling you which facts about your subject must appear in which order. Nothing in the manual makes one paragraph of historical analysis read like another.

The closest either manual comes to reaching into a sentence is advice rather than a rule, and it is worth knowing it exists: Turabian's chapter on revising sentences opens with a section headed "Focus on the First Seven or Eight Words of a Sentence." That is a book telling you where in a sentence to put your subject. It is still not telling you what to say.

Which means the uniformity people notice in their own history writing is mostly habit, not compliance. Chronological narration pulls every paragraph into the same shape: date, actor, action, consequence, note number. Model monographs get imitated. Both are worth breaking on their own merits, and neither costs you anything in the manual's eyes.

The parts where variation is completely free are the ones that carry your argument: what the source does not say, why the gap in the existing literature matters, what a piece of evidence will not bear. If those paragraphs read as interchangeable with the ones around them, that is a writing problem, and it is the one place in the file where fixing it costs nothing.

1. Split the file into the three texts before touching a word. Your prose, the transcribed sources, the apparatus. Do that with the manual open, not while revising, because the second and third categories are the ones you must not edit for style.
2. Count the analysed prose, not the page count. A 1,500-word essay with forty notes has less measured prose in it than the word count suggests, and the vendor says that in short documents "the prediction will be mostly 'all or nothing' because we're predicting on a single segment without the opportunity to overlap." See Short Documents and the All-or-Nothing Problem in Turnitin.
3. Do not paraphrase a primary source to make it look less repetitive. That is the one edit that damages the paper as history, and the AHA line above is the reason a supervisor will say so.
4. Bring the vendor's own position. Its FAQ says the percentage "should not be used as the sole basis for action or a definitive grading measure by instructors." What your instructor is told to do when your AI percentage is high collects that guidance, and Is 20% AI too high? covers why no number works as a pass mark.
5. Remember whose rules win. Chicago prints it on its own tip sheets: "Directions from your teacher, instructor, or dissertation office overrule these guidelines." That applies to formatting arguments, and it applies to whoever is telling you what your score means.

Suppose you decide the argument prose genuinely does need reworking. The thing at risk in a Chicago essay is not the prose. It is the superscript numbers attached to it, and the transcribed material sitting inside the block quotations. Merge two sentences and a note number ends up sitting after a statement its source does not support. Reflow a block quote and a nineteenth-century spelling quietly becomes a modern one.

HumanPen is ours, so the next few lines are the makers describing their own thing. It works on the uploaded file, not on pasted text. Scope is the whole point of it. You can mark the passages by hand, or upload the Turnitin or iThenticate report you already have and let the flagged ones be located out of it. The engine will not rewrite less than a whole paragraph, so a partial selection grows until it covers one, and that grown boundary goes on screen for your approval before anything runs. Nothing on the far side of that boundary is modified. You get back an editable document that has kept its structure, terminology, citations, layout and styles. A passage that qualifies and still comes back flagged can be put through again at no extra charge. The report-driven route is set out in Rewriting only the paragraphs a Turnitin report flagged.

Two things nobody can promise you. One is where the number lands afterwards. The other is that any tool has been measured on the layer that matters most to you here. Inside a Word file, a note marker and the note text sit in separate parts of the package, joined only by a number, and a proofread of the prose will never show you that the join came apart. So open the returned file, click three note numbers at random, and check they land where they did before. It takes a minute, and it is easy to forget. Word fields, tables of contents and cross-references is about the same layer.

Does writing in Chicago style raise an AI score by itself? Nothing the vendor publishes says so, and on the FAQ page described above `Chicago` and `Turabian` both return zero. What the vendor does say is that its false positives can involve text with little structural variation, and that only prose sentences are analysed. A Chicago essay concentrates its prose into one continuous narrative and puts everything else in the notes.

Are my footnotes counted in the percentage? There is no published answer. The definition of qualifying text excludes non-sentence structures, and a shortened note is not a sentence, but the vendor never uses the words footnote or endnote on that page. Bibliographies are a different matter: a 2023 release note says they are excluded when the AI writing report is processed, with resubmission needed for that to apply to an older report.

Should I convert my shortened notes back to full citations? No. That changes the apparatus, not the prose the score is computed from, and it puts you outside the style your department asked for. If the highlighting is on your narrative paragraphs, the notes were never the issue.

A block quote from a 1640 pamphlet came back highlighted. Can I clean up the spelling? Not for that reason. Turabian gives quoting accuracy its own section, and the AHA statement says any change to a primary source should be noted. If a transcription is genuinely wrong, fix it because it is wrong; do not modernise a source to change how a paragraph reads to software.

My whole methodology and historiography section is one continuous highlight. That pattern has a mechanical explanation before it has an interpretive one, and it is worth understanding first: Turnitin flagged my whole methodology section. Take the report, your notes and your drafts into the conversation rather than a rewrite.