A reviewer asked for the page number of a quote. I opened the link. The paragraph was gone. The citation had been correct when I wrote it, and the source no longer existed.
Search "how to cite a website" and you get formatting rules. Punctuation, italics, where the year goes. The rules matter, and they are not where most citations fail.
Citations fail three months later, when someone asks you to verify one. By then the tab is long closed and all you kept was a URL. That URL may now point at a different article, or at nothing.
Formatting is just arranging five facts in a fixed order. If the facts are incomplete, a perfect format produces a broken reference, and you cannot reconstruct what you never recorded.
What to capture before you close the tab
The author. A person, an organization, or a site's editorial team. Look at the top and the bottom of the page; publishers put bylines in both places and they do not always agree. When an organization is the author, APA wants the full name in the author slot, not an abbreviation.
The date. Pages carry more than one. Publication date, last-updated date, and the timestamps in the comments all look similar and sit close together. The one you want usually appears next to the byline as "Published" or "Updated". If the only thing you can find is "6 min read", there is probably no date on the page at all.
The title. Use the page's own headline, not the browser tab. Tab titles get truncated for width, and on many sites they are rewritten to carry the brand.
The site name. The publication or platform the page belongs to. Blog posts often omit it because the domain makes it obvious to a reader. A citation is read out of context, so it cannot rely on the domain.
The URL. Copy from the address bar, not from a search result. Search links carry tracking parameters, and some of them resolve to a different page than the one you read.
The two most commonly skipped are the date and the site name. The date is easy to miss because of where it sits. The site name feels too obvious to write down. Six months later, neither is obvious.
Four styles, one set of facts
Same fictional page throughout: Jane Doe, published March 12, 2025, "How to Read a Report", on Field Notes, at example.org/report.
| Style | What the reference looks like |
|---|---|
| APA 7 | Doe, J. (2025, March 12). *How to read a report*. Field Notes. https://example.org/report |
| MLA 9 | Doe, Jane. "How to Read a Report." *Field Notes*, 12 Mar. 2025, example.org/report. |
| Chicago (note) | 1. Jane Doe, "How to Read a Report," Field Notes, March 12, 2025, https://example.org/report. |
| GB/T 7714 | DOE J. How to read a report[EB/OL]. (2025-03-12)[2026-11-14]. https://example.org/report. |
Four differences account for almost every error when converting between them: the order of the author's names, how the date is written, whether the title takes italics or quotation marks, and whether an access date appears at all.
APA 7: the date in the parentheses has two sources
The author slot takes surname first, then initials. The date goes into parentheses as year, month, day, with the year leading. The page title is italicized and takes sentence case, so only the first word and proper nouns are capitalized. The site name follows in plain text.
When the site name and the author are the same entity, APA drops the site name. Plenty of blogs belong to one person, so this comes up more often than the manual's brevity suggests.
APA's access date is an exception, not a default. Add "Retrieved March 12, 2025, from" only when the content is designed to change and no archived version exists. A wiki entry qualifies. An ordinary article does not.
MLA 9: with no author, the title moves to the front
MLA runs author, title, container, date, URL, separated by periods. Dates are inverted: day first, month abbreviated. The page title takes quotation marks, the site name takes italics.
When there is no author, MLA starts with the title instead of entering a placeholder. APA handles the same absence differently, so this is where most manual conversions break.
MLA 9 also allows dropping the "https://" from the URL. People frequently do the opposite, adding it back because it looks incomplete.
Chicago: notes and bibliography are two different formats
Chicago runs two systems side by side, and the notes-bibliography system changes the shape of every entry between the two places it appears. In a note, the name is in reading order: Jane Doe. In the bibliography it inverts: Doe, Jane.
People copy the note into the bibliography and end up with the wrong name order, the wrong punctuation, and the wrong capitalization. Do the bibliography entry separately.
On access dates, Chicago is stricter than APA and more specific than MLA. Undated online sources require "accessed" followed by the date, placed before the URL.
GB/T 7714: the one Chinese journals will ask for
If you are submitting to a Chinese journal, the national standard applies: Author. Title[EB/OL]. (date of update)[date cited]. URL.
The bracketed EB/OL marks an electronic resource obtained online, and both letters are required. Leave it out and the entry is incomplete on its face.
Chinese names are written as they are. English names go surname in full capitals, given name abbreviated, surname first. When Chinese and English sources share one list, this is where consistency breaks first.
The date cited inside the square brackets is mandatory. That is the opposite of APA's default, and it is the single most common mismatch when a reference list gets translated.
When the author or the date is missing
Missing author: use the organization or the site name, which APA and GB/T both accept. MLA omits the author and starts with the title. Write "Anon." only if the page is actually signed that way.
Missing date: APA writes (n.d.), MLA writes "n.d.", and GB/T drops the update segment and keeps only the date cited. The three placeholders are not interchangeable.
Both cases are more common than the manuals imply, because a large share of the web has no editorial metadata at all. Record what exists and note what you substituted, so future-you knows the entry was assembled rather than copied.
When the page itself changes
This is the case no style guide solves. Your reference was accurate when you wrote it. The reviewer checks it and the passage is not there.
Two options. Find an archived copy and cite the archived URL, stating the archive date in the text. Or keep a snapshot of the page the first time you read it, and verify against your own copy.
The second is more reliable because it does not depend on someone else having run a crawler on the right day. It costs an extra step and some disk space, and it is the only version of the page you control.
Access dates: four different answers
| Style | Required? |
|---|---|
| APA 7 | No, unless the content changes over time and has no archived version |
| MLA 9 | Optional. Recommended when there is no publication date or the content is unstable |
| Chicago | Yes for undated online sources |
| GB/T 7714 | Yes, always |
This row is wrong more often than any other, because people fill it from memory and the memory comes from a different style.
Which date to write causes a second round of confusion. It is the day you actually read the page. Not the day you wrote the draft, not the day the page was updated. If you read it across several sittings, use the first one.
What I do now
In my own drafts, formatting errors are rarely formatting. The author slot is empty, the date is an access date, the site name has been folded into the title. The reference was assembled from incomplete input.
The input goes missing at capture time. I read, I highlight the sentence I want, I move on. Nothing about the source gets recorded, because at that moment the source does not feel like something I will need.
Inkory handles that step: highlighting a passage also keeps the author, date, title, site name and a snapshot of the page. When you write, the passage arrives with its source attached, formatted for APA or GB/T 7714, and checking it later compares against the snapshot you kept rather than the live page.
Highlighting, search, export and full backup are free and need no account. The writing desk and the AI review are the paid part, with a 14-day full trial from install.
Where this breaks
Scanned PDFs need OCR first. The recognizer reads text line by line, so it flattens anything stacked vertically. Basic inline formulas come through. Fractions, integrals and matrices do not, because the recognizer cannot tell that the stacked blob is one object.
Page numbers will not match. The page order your PDF reader shows and the number printed on the page are frequently different once front matter is counted. If a citation needs a page number, read the one printed at the bottom, not the one in the reader's toolbar.
Paywalled pages. A snapshot can only be taken while you have access. Someone else following your citation will hit a wall, and that is a property of the source rather than of the tool.
No sync. Move to another computer and the library does not follow. Export and import is the workaround, and it is the most annoying part of the current design.
Tools can do more of this than most people expect: capturing, ordering, generating four styles, matching each quote to its source. What they cannot do is judge. Whether this version is the one you mean to cite, whether that date is the publication date, whether the author is a person or an organization signing off. That part stays with you, which is why it goes on your name.
On accuracy: the rules above follow APA 7th edition, MLA 9th edition, Chicago 17th edition (notes-bibliography) and GB/T 7714-2015. Examples use a fictional page. Style manuals get revised; check the current edition before you submit.
Related: GB/T 7714 and APA: where auto-generation should stop · Why your highlights never reach your draft