Zotero, EndNote and Mendeley: What "Remove the Field Codes" Actually Removes

Because a reference list is machine-generated, identically formatted and structurally repetitive, which makes people assume it is pulling their AI percentage up. It is not, on either side of the calculation. The risky step is converting live citations to plain text. Before doing that, identify the Word object your add-in uses: Zotero documents fields or bookmarks, the older Mendeley plugin names Mendeley Fields, and Mendeley Cite names content controls. Those objects do not share one removal path.

Turnitin's AI writing report analyses what it calls qualifying text, defined as prose written in standard grammatical sentences, with lists and other non-sentence structures outside it. A release note dated 9 August 2023 separately says bibliographies are excluded when the report is processed. So a reference list is not in the numerator and not in the denominator. Padding it does not dilute anything; deleting it does not remove anything.

If a bibliography is coloured in a report, first check whether you are looking at the Similarity Report rather than the AI writing report. Matched reference entries are what a shared citation style is designed to produce. Three articles cover those branches: what the exclusions actually cover, which report you are holding, and the stale-report case.

The rest of this article is about the software, because that is where a reversible citation workflow can become a plain-text file.

This is not a hypothetical or a thing only some journals want. Elsevier's Journal Article Publishing Support Center has a standing page on it, last updated 15 December 2022:

"Reference manager software programs such as Mendeley, EndNote, or Zotero can be used to help manage references in scientific articles. While these programs are useful because citations from most databases can be exported to a Library, it can interfere with the production process. It is recommended that authors remove field codes before submission."

The same page carries the warning that matters more than the instruction:

"Once you remove all hidden codes, you can no longer reformat or unformat the citations (or access the Library) so always make a copy of your document before removing any codes."

If a supervisor, editor or another workflow asks for a flattened file, treat that as a request for a separate delivery copy rather than a change to the working master.

Those are the three managers Elsevier names, with Mendeley counted twice because it ships two different add-ins. Read the last two rows together. The older page names Mendeley Fields and says Export converts the entire document to plain text. The newer page calls citations and bibliographies content controls and prescribes Remove Content Control. The object type determines which vendor instruction applies.

Clarivate's position on going back is worth quoting on its own, because "irreversible" gets read as marketing caution and it is not:

"Field code removal is a one-way process. Once you've removed EndNote™ field codes, you will need to reinsert the citations or revert to a backup of the document that still has the field codes in it."

Elsevier's manual route, on the same page quoted above: select all with Ctrl+A, then press Ctrl+Shift+F9 "to unlink all fields and remove hidden links". Clarivate publishes its own manual route, which is Ctrl+A, then Ctrl+6, then copy and paste into a new document.

Now look up that keystroke in Microsoft's documentation rather than in a library guide. On Microsoft's keyboard-shortcuts page for Word, the entry reads:

"Unlink a field. Ctrl+Shift+F9"

and further down the same page, "Ctrl+Shift+F9: unlinks the current field."

A field. Not a citation. Microsoft's field-code reference separately lists TOC, SEQ and REF as fields. That means the documentation does not give Ctrl+A followed by Ctrl+Shift+F9 a citation-only boundary. If those objects are present in your selection, the published instruction gives you no reason to assume they are outside the command's scope. What a field actually is, and what a plain-text round trip costs a long document has the mechanics.

Mendeley Cite's page uses different object language: it calls a citation or bibliography a content control and tells you to right-click that object and choose Remove Content Control. Microsoft's field-shortcut page does not establish what Ctrl+Shift+F9 does to that add-in, so use Mendeley's documented route rather than treating a field shortcut as interchangeable.

Zotero adds a second alternative. Its Document Preferences let you choose "Whether to store citations as Fields or Bookmarks", with the advice that unless you need to collaborate with LibreOffice users "you should always choose Fields". A field-view shortcut can inventory fields, but its documentation does not identify bookmarks or content controls for you.

There is a free, instant, reversible way to see what is actually in the file. From the same Microsoft page:

"Switch between all field codes and their results. Alt+F9"

Press it and Word switches between field codes and their displayed results. A table of contents can expose a TOC instruction, numbering a SEQ instruction and a cross-reference a REF instruction. If your citations also appear as field codes, you know fields are involved. If they do not, Alt+F9 alone does not tell you whether they are bookmarks, content controls or another object; check the add-in and its documented command. Press Alt+F9 again to switch the field display back.

That field inventory is a useful pre-flight check before an irreversible command, but it is only one part of identifying the document's objects.

The instinct, when a bibliography lights up in a similarity report, is to reword the entries until they stop matching. Set aside the question of whether that is a good idea in principle. Just look at what the software does with the edit.

Zotero's page is blunt about a live document:

"Manual edits made to the bibliography in Word will be overwritten the next time Zotero refreshes the document."

And about in-text citations, on the same page: "Manual modifications will prevent Zotero from automatically updating the citation."

So in a document that still has live fields there are only two outcomes, and neither is the one intended. Either the next refresh silently reverts your edit, or the edit sticks and that citation is now disconnected from the library, which means the next style change updates every entry except that one. You find out at the worst possible moment, which is when the journal asks for a different style.

In a flattened document the edit persists, and now the entry no longer matches because it no longer follows the style guide, which is a marker's problem rather than a detector's. A matching reference entry is what a shared citation style exists to produce. And a rewriting tool that reaches the reference list has been recorded doing worse than that: there is a filmed case, with the input built on camera first, in what happens to citations, tables and equations.

The risk is lower if the irreversible step happens last and only on a copy.

1. Create and keep a working master. Name the delivery copy separately before removing anything. EndNote explicitly says its command creates a new copy; the other product pages checked here do not make that same promise, so create your own first.
2. Finish the citation work first. Style changes, added sources, page numbers, prefixes and suffixes. After flattening, none of these can be done by the manager.
3. Refresh once before flattening, so what gets frozen is the manager's current output rather than something stale.
4. Identify the object and follow its specific path. Zotero documents Unlink Citations for its citations; EndNote documents Convert to Plain Text or Remove Field Codes; Mendeley Cite documents Remove Content Control. The older Mendeley plugin says Export converts the entire document to plain text, so do not assume it preserves unrelated live objects.
5. Use Alt+F9 to inventory fields before and after. It can reveal TOC, SEQ, REF and citation fields, but not bookmarks or content controls. Open the delivery copy and check its table of contents, numbering, cross-references and citations rather than inferring success from an unchanged appearance.
6. Keep the flattened copy out of the editable citation workflow. Clarivate's answer to "can I get the codes back" is to reinsert the citations or return to a backup.

The reason all of this comes up in an AI-flag conversation is that people run a flattening step because something else in the workflow demands plain text. That is the wrong order. If a tool can only accept pasted text, the flattening is not a step in that tool's process, it is a cost you paid before you started.

HumanPen is ours, so treat this as a description rather than a recommendation. Its humanize engine currently supports English documents only. It takes the document rather than pasted text, and the scope is something you set: mark the passages yourself, or import a Turnitin or iThenticate AI writing report so the flagged text defines the boundary. The unit is a paragraph rather than a sentence, so a selection ending mid-paragraph pulls the rest of that paragraph in with it, and you approve the resulting boundary before the job begins. Nothing outside the confirmed paragraphs is touched. The product page says it "aims to preserve meaning, structure, terminology, citations, layout, and styles while rewriting only the necessary language", followed by an instruction to review complex documents after download. Eligible results can continue lowering AI for free.

Which does not remove the check. Open the returned file, press Alt+F9, and confirm your fields are still fields.

Does my Zotero-generated bibliography raise my Turnitin AI score? No. The AI writing report analyses prose written in standard grammatical sentences, and a release note dated 9 August 2023 says bibliographies are excluded when the report is processed, with resubmission required before that applies to an older report.

Is removing field codes reversible? Zotero's page calls its removal irreversible. Clarivate calls EndNote field-code removal a one-way process and says you will need to reinsert the citations or go back to a backup. EndNote's page explicitly says its command creates a new copy. The Zotero and Mendeley pages checked here do not make that same copy-creation promise, so create one yourself.

Will Ctrl+Shift+F9 break my table of contents? Microsoft documents that shortcut as "Unlink a field", while its field-code reference lists TOC as a field; Elsevier's instruction applies the shortcut after selecting the whole document. We did not run a Word-version test matrix, so inspect your fields and work only on a copy instead of assuming a citation-only result.

Why does Mendeley tell me to right-click instead of pressing a shortcut? Mendeley Cite stores citations as content controls rather than as fields, so its removal step targets content controls. The older Citation Plugin used fields and had an Export command instead.

I edited an entry directly in Word and it changed back. That is documented behaviour, not a bug. Zotero's page says manual edits to the bibliography in Word are overwritten the next time it refreshes the document. Fix the item in the library, or the style, and refresh.

Should I flatten the document before sending it to my supervisor? Only if the recipient or submission workflow requires it. Send a flattened delivery copy and keep the live-citation master, so the next round of citation work is still possible.