education

Best AI Citation Generators

The best AI citation generators in 2026 for APA, MLA, Chicago and Harvard — Scribbr, Zotero, Grammarly Citation Finder, QuillBot, Paperpile and more, ranked on accuracy, style coverage and price, with the fabricated-reference research behind the ranking.

Updated September 2026

Quick answer: The best citation generator for most students in September 2026 is Scribbr, because it is free, ad-free, covers APA, MLA, Chicago, Harvard, AMA, IEEE and ACS, and builds each reference from source metadata rather than from a language model. Once you pass roughly twenty sources, switch to Zotero — free, open-source, over 10,000 styles, and it stores the sources as well as formatting them. The best AI-agent option is Grammarly’s Citation Finder, which reads your draft, flags claims that need support, finds sources and formats them, and is included in Grammarly Pro at $12 a month billed annually. The distinction that matters more than any ranking on this page: a citation generator looks up a real record from a DOI, ISBN or URL and formats it, while a general chatbot asked for references invents them — a 2023 study by Walters and Wilder across 42 topics found 18 per cent of GPT-4’s citations entirely fabricated (Scientific Reports), and fabricated references now appear in roughly 1 in 277 PubMed-indexed papers. Never ask a chat assistant for a reference list.

This page ranks tools that produce a correctly formatted reference from a source you already have, and the reference managers that store those sources over a degree or a career. That is a narrower job than the pages around it:

• Best AI for research covers finding and synthesising sources — Perplexity, Deep Research agents, Elicit, Consensus. It deliberately leaves referencing to this page.
• Best AI for essays covers drafting the essay itself, including the academic writing platforms that bundle citations with prose.
• Best AI study tools and best AI for college students cover the wider student stack.
• This page covers formatting a reference correctly, managing a bibliography, citing generative AI itself, and checking that a reference is real.

The one distinction that decides everything: lookup versus generation

Two completely different technologies are sold under the phrase “AI citation generator”, and confusing them is the single most expensive mistake a student can make in 2026.

Lookup tools take an identifier — a DOI, an ISBN, a URL, a paper title — query a bibliographic database such as CrossRef, OpenAlex, PubMed or WorldCat, retrieve the real metadata, and apply a style template to it. Scribbr, Zotero, ZoteroBib, MyBib, BibGuru, QuillBot, Paperpile, EndNote and Mendeley all work this way. They can be wrong, but they are wrong in bounded ways: a scraped author name, a journal title in capitals, a missing edition. The reference points at something that exists.

Generative tools ask a language model to produce a reference. The model has no database; it produces the most probable-looking string of author, title, journal, year and DOI. Research on the failure mode found hallucinated citations are not random — they are patterned recombinations of real authors, real journals and real dates, with duplication in nearly 30 per cent of cases. That is precisely why they survive a quick read: everything about them looks familiar.

The practical rule: use a language model to find candidate sources, then paste the DOI or title into a lookup tool to format them. Never copy a reference list straight out of a chat window. Every tool below is scored on which side of this line it sits.

The state of AI citation tools: September 2026

Five things define the category as of today.

Fabricated references became the dominant integrity failure. An analysis of PubMed-indexed papers found roughly 1 in 277 papers published in the first seven weeks of 2026 cited a paper that does not exist, up from 1 in 458 across 2025 and 1 in 2,828 in 2023 — a sixfold rise in two years (Retraction Watch, reported by STAT). This is the context every tool on this page now operates in.

Peer review is not catching it. GPTZero’s January 2026 scan of 4,841 papers accepted to NeurIPS 2025 found at least 100 confirmed hallucinated citations across 53 papers — roughly 1 per cent of accepted work, all of it past review. A separate GPTZero scan of 300 papers under review for ICLR 2026 flagged 90 and, after human verification, confirmed 50 — about 17 per cent — contained at least one hallucinated citation, each already reviewed by three to five experts (GPTZero). If reviewers at the top machine-learning conferences are missing fabricated references, a marker with 90 essays and a weekend will too — which cuts both ways, because the checking tools described below are now being run over submissions at scale.

The AI agents arrived, and they sit on the right side of the line. Grammarly shipped a suite of specialised agents including Citation Finder, which reads a draft, identifies claims that need support, retrieves evidence and formats the citation in your chosen style (Grammarly). It is retrieval-based rather than generative, which is the design that matters. Grammarly has been part of the Superhuman product family since late 2025, alongside Superhuman Mail and Coda.

Reference managers added AI — the incumbents first, the open-source one not at all. EndNote 2025, introduced by Clarivate in April 2025, was the first EndNote with AI features built in — a Key Takeaway tool that extracts insights from a paper and a Find a Journal tool — with a generative research assistant for chatting with full-text documents promised for a later 2025 update. Zotero still ships no native AI — summarisation and related-work suggestion come from third-party plugins, which many researchers consider a feature rather than a gap.

A new category appeared: the citation checker. Tools including Citely, SwanRef and the open-source RefCheckAI do the reverse job — take a finished reference list and verify each entry resolves against CrossRef, PubMed, arXiv, OpenAlex or Google Scholar. They also target “chimera” references, where a valid DOI is paired with fabricated metadata. Vendor accuracy claims in this category are unaudited; treat them as marketing until independent evaluation exists.

Best AI citation generators ranked (September 2026)

Prices are USD and were read on 5 September 2026 from vendor pricing pages and, where a vendor page could not be reached, from independent summaries that agreed with each other. Confirm any figure before subscribing.

Not on this table on purpose: ChatGPT, Claude, Gemini and any other general chat assistant. They will produce something that looks like a reference. See the accuracy section for what that costs.

Free citation generators, reviewed

1. Scribbr — best free citation generator overall

Price: Free Builds from: Title, URL, DOI or ISBN, with an autocite flow that searches for the source details Styles: APA, MLA, Chicago, Harvard, AMA, IEEE, ACS Owner: Learneo, the parent company of QuillBot, LanguageTool, Course Hero, CliffsNotes and Symbolab

Scribbr’s Citation Generator is the tool we would put in front of an undergraduate with no other context. It is free with no advertising, it covers the four styles that account for the overwhelming majority of university coursework plus three specialist ones, and its autocite flow will find a source from a partial title rather than demanding a clean DOI. It also generates in-text citations alongside the reference-list entry, which is where most students actually lose marks.

One caution on accuracy. No peer-reviewed evaluation of consumer citation generators exists, and the percentage accuracy figures that circulate for Scribbr come from competitor blogs whose methodology cannot be checked, so we do not repeat them. The pattern that user reports agree on is the useful part: strong on journal articles and books with clean metadata, weak on multi-author works, translations, archival and legal material.

Why it wins: Free, ad-free, correct on the common cases, and it produces the in-text citation as well as the reference.

Limitations: Weak on archival material, legal sources, translations and anything without clean metadata. No library — it formats citations, it does not store them, so it does not scale past a single essay.

Best for: Undergraduates and anyone writing a single piece with under twenty sources.

2. QuillBot Citation Generator — best style breadth for free

Price: Free for the citation generator; Premium about $19.95 a month, verified students about $6.25 a month Builds from: Title, URL, DOI or ISBN Styles: Over 1,000

QuillBot’s Citation Generator is free and supports over 1,000 styles, which matters if your department mandates something outside the big four — OSCOLA, Vancouver, a house style. QuillBot, LanguageTool and Scribbr were consolidated into a single business unit under Learneo, so the underlying citation engine is largely shared with Scribbr; the difference is style coverage and the surrounding product.

We rank it below Scribbr for one reason: the free citation generator sits inside a product whose main business is a paraphraser marketed partly at rewriting text to evade AI detection. That is an academic-integrity problem in its own right, covered on our best AI humanizer and best AI detector pages. The citation generator is not implicated in that, but it is worth knowing what account you are creating.

Why it wins: Over 1,000 styles at no cost is the widest free coverage available.

Limitations: Same metadata-scraping weaknesses as Scribbr. Heavy upsell into Premium. The wider product is contentious in academic settings.

Best for: Anyone whose required style is not APA, MLA, Chicago or Harvard.

3. BibGuru — cleanest ad-free free tool

Price: Free; Plus $4.99 a month Builds from: Title, URL, DOI or ISBN Styles: APA, MLA, Chicago, Harvard, IEEE, AMA and more

BibGuru is built by the Paperpile team and aimed at students who want a bibliography without an account, an advert or an upsell. It keeps projects, it is fast, and it is the tool most likely to be recommended by a librarian who has watched students get trapped in a Citation Machine subscription.

Why it wins: No advertising, a real free tier, and the engineering pedigree of a paid reference manager behind it.

Limitations: Smaller style list than QuillBot. Light on storage and organisation — it is a generator with projects, not a manager.

Best for: Secondary and undergraduate students who want a clean tool and no account.

4. MyBib — best zero-friction option

Price: Free, with no advertising and no privacy trackers Builds from: URL, title or ISBN Styles: APA, MLA, Chicago, Harvard, Vancouver and more

MyBib does one thing: paste a URL, get a citation. No sign-up, no advertising, no tracking. Its Chicago generator is one of the free tools that has been updated for the 18th edition, which is not universal — see the styles section below.

Why it wins: The lowest-friction free tool here, and the privacy position is unusually good for a free student product.

Limitations: No library, no in-text citation help, and web-page scraping is the weakest input type across every tool on this page.

Best for: Citing web pages and news articles quickly.

5. Citation Machine, EasyBib and BibMe — the famous ones, and why they rank tenth

Price: Free tier heavily upsold into a paid subscription Owner: Chegg, which acquired Imagine Easy Solutions — the company behind all three — in 2016 Styles: APA, MLA, Chicago and more

These are three brands running on one platform, and for a decade they were the default answer to “how do I cite this”. They still work and they remain live as of September 2026. They rank tenth because the free experience is now advertising-dense and routes persistently towards a paid plan, because third-party reviews report DOI and ISBN imports that do not clean up the data — journal titles returned in capitals is the recurring complaint — and because free alternatives above are better on every axis except brand recognition.

Why anyone still uses them: Coverage of unusual source types is genuinely broad, and school libraries have linked to them for fifteen years.

Limitations: Advertising, subscription pressure, and reported accuracy decline on APA 7th edition formatting.

Best for: Nothing that Scribbr, BibGuru or MyBib does not do better and more cheaply.

Reference managers, reviewed

Past roughly twenty sources, a generator stops being the right tool. You need something that stores the source, the PDF, your notes and the citation together, and writes the bibliography into your document.

6. Zotero — best reference manager for almost everyone

Price: Free and open source; 300MB of online storage free, paid tiers above that Version: 10.0.1, released 24 August 2026 (Zotero 10 shipped on 17 August) (changelog) Styles: Over 9,000 CSL styles, per Zotero Integrations: Word, LibreOffice, Google Docs, browser connector for Chrome, Firefox, Edge and Safari

Zotero is the default recommendation for anyone doing sustained research, and it has been for years. The browser connector captures a paper, a book record, a news article or a court filing from the page you are on, with the PDF attached, in one click. The word-processor plugins insert citations and rebuild the bibliography as you write. It runs offline, it runs on Linux, and because it is open source your library is not hostage to a subscription.

It has no native AI — Zotero’s own feature list mentions none — and this is the honest trade. Summarisation, related-work suggestion and semantic search come from third-party plugins rather than from Zotero itself. Given what the fabrication research says about generative citation features, a reference manager that refuses to invent anything is a defensible design.

ZoteroBib (zbib.org) is Zotero’s free web generator for people who do not want the application: paste a URL or DOI, get a bibliography, no account required. It is a genuine alternative to Scribbr for one-off work.

Why it wins: Free, open, offline-capable, the widest style support of any tool here, and a capture workflow nothing else matches.

Limitations: No native AI. Free storage is 300MB, which fills quickly with PDFs. Group libraries and syncing need configuring.

Best for: Dissertations, theses, literature reviews and anyone who will still be citing in three years.

7. Paperpile — best inside Google Docs

Price: Regular $4.15 a month and Expert $5.75 a month with the academic discount ($8.30 and $11.50 without), billed annually only, with a 30-day free trial (Paperpile) Integrations: Google Docs, Google Drive, Word, browser extension

Paperpile is the answer if your writing lives in Google Docs. Zotero’s Google Docs integration works but is clearly a port; Paperpile was built for it. Sources sync to Google Drive, the citation picker is native, and the whole thing is fast. At $4.15 a month for students and academics it is the cheapest paid reference manager on this page, though it bills annually only.

Why it wins: The best Google Docs citation experience available, at the lowest paid price of any reference manager here.

Limitations: Subscription, and it is most valuable inside the Google ecosystem. Less useful if your institution runs on Word and OneDrive.

Best for: Postgraduates and academics who draft in Google Docs.

8. EndNote 2025 — best if your institution mandates it

Price: About $274.95 full version, about $149.99 student edition, per reseller listings Launched: April 2025 (Clarivate) Owner: Clarivate Styles: Over 7,000 output styles

EndNote is the institutional standard in medicine, the hard sciences and publishing, and EndNote 2025 is the first release to ship AI-powered features inside the product — Key Takeaway and Find a Journal at launch, with a generative research assistant promised in a later update (Clarivate). Its strengths are real: deep integration with Web of Science, precise control over output styles, and the ability to handle libraries of tens of thousands of references without falling over.

Its weakness is the price. Zotero does most of what most researchers need for nothing. The rational reasons to buy EndNote are that your institution provides a site licence, your supervisor’s shared library is in EndNote, or your target journal supplies an EndNote style file and nothing else.

Why it wins: Scale, Web of Science integration, output-style control, and it is what the people marking your work probably use.

Limitations: The most expensive tool here by an order of magnitude. Prices quoted are reseller listings; confirm at endnote.com/buy.

Best for: Medical, health-science and lab researchers at institutions with a site licence.

9. Mendeley Reference Manager — free, with an Elsevier attached

Price: Free; paid storage tiers Owner: Elsevier Integrations: Word, browser importer, Mendeley Cite

Mendeley is free, works well, and has moved cloud-first with AI-assisted reading features and tighter integration into Elsevier’s publishing infrastructure. That last clause is the whole debate: some researchers consider deep integration with a major publisher a convenience, others consider it a reason to use Zotero instead. Both positions are reasonable and we are not going to adjudicate a politics question with a benchmark.

Why it wins: Free, solid Word integration, and genuinely convenient if you already work inside ScienceDirect and Scopus.

Limitations: Elsevier ownership, a web-first direction that suits some workflows and not others, and less style flexibility than Zotero.

Best for: Researchers already embedded in the Elsevier ecosystem.

AI tools that find sources as well as cite them

These sit between best AI for research and this page. They do not just format a source you have — they go and find one for a claim you have written.

10. Grammarly Citation Finder — best AI agent for evidence and citation

Price: Included with Grammarly Pro, $12 per member per month billed annually ($144 a year) or $30 a month billed monthly Styles: APA, MLA, Chicago, selected from a dropdown in the Citation Finder panel

Citation Finder is one of the specialised agents Grammarly shipped in 2026. It reads your draft, flags statements that need support, retrieves evidence that supports or disputes them, and formats the citation in your chosen style. The architecture is retrieval-first, which places it on the safe side of the lookup-versus-generation line.

The honest limitation, and reviewers make this point consistently, is that finding a source for a sentence is not the same as your reference set being complete, current or strong enough for the argument. Citation Finder will not tell you that you have cited three papers from one lab, or that your key claim rests on a 2011 study that has since failed to replicate. It closes gaps; it does not assess a bibliography.

Why it wins: The only tool here that starts from your writing rather than from a source you already have, inside a product most students already run.

Limitations: Requires Pro. Three styles, not a thousand. It finds support for claims rather than evaluating whether the support is good.

Best for: Essay and report writers who have drafted first and need to evidence claims afterwards.

11. SciSpace — best for reading and citing together

Price: Free Basic with 100 credits a month; Premium $12 a month billed annually with 1,200 credits; Advanced $70 a month billed annually with 10,000 credits and Deep Review

SciSpace pairs a large paper index with AI explanation of dense passages, which makes it useful when the problem is not finding the paper but understanding it well enough to cite it correctly. Credits meter usage, so heavy literature work moves quickly from free to paid.

Best for: Postgraduates working through unfamiliar literature.

12. Sourcely — best for retrofitting sources to a draft

Price: Free for inputs up to 300 characters with no sign-up; Ultra $19 a month with 30 deep searches; Max $39 a month with 1,000 deep searches

Sourcely takes text you have already written, identifies claims lacking support, and searches academic databases for sources that match, with factual summaries per source and export to a citation style.

A word on the workflow it enables. Finding evidence for an argument you have already made is not the same as forming the argument from the evidence, and at its worst this is citation-shaped decoration. Used well — you know the field, you are recovering a reference you could not relocate — it is a genuine time-saver.

Best for: Writers reconstructing a reference list, with the caveat above understood.

How accurate are AI-generated citations? The evidence

This is the section that should decide your tool choice.

A general chat model fabricates a large fraction of the references it produces. Walters and Wilder’s 2023 study in Scientific Reports, which prompted GPT-3.5 and GPT-4 for literature reviews across 42 multidisciplinary topics, found 55 per cent of GPT-3.5’s citations and 18 per cent of GPT-4’s citations were entirely fabricated. Among the citations that were not fabricated, 43 per cent (GPT-3.5) and 24 per cent (GPT-4) contained substantive errors — wrong volume, wrong issue, wrong page range. Newer frontier models have improved on this, and retrieval-grounded modes improve it further, but no published figure puts any general assistant at zero.

The fabrications are designed, structurally, to pass a glance. Research published in Publications found hallucinated references are patterned recombinations of real authors, journals, dates and keywords rather than random strings, with duplication across outputs in nearly 30 per cent of cases. A fake reference to a plausible author in a plausible journal in a plausible year is exactly what an experienced reader expects to see.

It is reaching published work at an accelerating rate.

Peer review is not a sufficient filter. GPTZero’s January 2026 scan of 4,841 papers accepted to NeurIPS 2025 identified at least 100 confirmed hallucinated citations across 53 papers, about 1 per cent of accepted work. In a 300-paper sample of ICLR 2026 submissions, GPTZero confirmed 50 papers — about 17 per cent — carried at least one hallucinated citation that three to five reviewers had not flagged; one cited example.com (GPTZero).

What this means for the tool you pick. Every failure above originates in generation, not lookup. A DOI-based generator cannot fabricate a paper, because it is reading a record that exists. The entire risk in your bibliography enters at the moment you accept a reference that no database produced.

Which citation style, and where the generators lag

The current editions, as of 5 September 2026:

The Chicago 18th-edition trap is the one to know. Two changes bite immediately. From the 18th edition, the city of publication is no longer given for books — publisher name only. And for sources with more than six authors, only the first three appear in the bibliography, with no more than two named in notes (Chicago Manual of Style, and see Purdue OWL). If your generator returns “New York: Norton, 2023”, it is on the 17th edition. MyBib and Grafiati are among the free tools confirmed updated for the 18th; check any other tool with a book citation before trusting it.

Harvard has no governing body, which is why generators disagree about it and why your department’s own guide overrides any tool on this page.

How to cite ChatGPT and other generative AI

This is a different question from the rest of the page — not “how do I cite a source with an AI tool” but “how do I cite the AI tool itself” — and it comes up constantly, so here is the current guidance.

APA 7th edition treats a generative AI tool like software or an algorithm. The reference names the company as author, the tool and version as the title, and a bracketed description (APA Style):

OpenAI. (2026). ChatGPT (5 September version) [Large language model]. https://chat.openai.com

APA does not put the prompt in the reference. Describe the prompt in the body of your paper, and for anything longer than a sentence or two, put the full exchange in an appendix and cite it in text as (OpenAI, 2026; see Appendix A for the full transcript).

MLA 9th edition puts the prompt first, as the title of the source, and does not treat the AI as an author (MLA Style Center):

“Identify the main themes in The Lord of the Rings trilogy” prompt. ChatGPT, 24 May version, OpenAI, 12 June 2026, https://chat.openai.com/share/a80e77ac.

MLA asks you to cite the tool whenever you quote, paraphrase or incorporate its output, and to acknowledge other functional uses — proofreading, translation, idea generation — in a note or in the text.

Chicago 18th edition treats AI output as personal, unrecoverable communication: cite it in the text or in a note, and generally do not include it in the bibliography, since a reader cannot retrieve your specific exchange. Check Chicago’s citation quick guide before submitting, as the guidance here has moved more than APA’s or MLA’s.

Before you do any of this, check whether you are allowed to. Style guidance tells you the format; your institution’s academic-integrity policy tells you whether the use is permitted at all. A correctly formatted citation of a tool you were not allowed to use is still a breach. Our best AI for essays page covers the disclosure question in detail.

How to check a citation is real, in under a minute

Four steps, in order of speed:

1. Resolve the DOI. Paste it after https://doi.org/. A real DOI resolves to the paper. A fabricated one 404s. This catches most invented references immediately.
2. Search the exact title in quotes on Google Scholar or Semantic Scholar. No exact-title match is a strong signal the paper does not exist.
3. Check the metadata matches the paper, not just that the paper exists. The “chimera” failure — real DOI, wrong authors or wrong year attached to it — survives step one. Open the record and compare the author list, year, volume and pages.
4. Run the whole list through a checker if you have more than about thirty references. Citely, SwanRef and the open-source RefCheckAI cross-reference against CrossRef, PubMed, arXiv, OpenAlex and Google Scholar. Vendor accuracy claims in this category are unaudited, so use them to shortlist suspects for manual checking, not as a clearance certificate.

If a reference was produced by a lookup tool from a DOI or ISBN, steps one to three are near-instant and usually unnecessary. If it came out of a chat window, do all four.

Best citation generator for each situation

Best free citation generator: Scribbr. Free, ad-free, seven styles, in-text citations included, and correct on the common source types.

Best for a dissertation or thesis: Zotero. Free, over 9,000 styles, stores sources and PDFs alongside the citations, and will still be readable in five years because the format is open.

Best AI-powered option: Grammarly Citation Finder. Retrieval-based rather than generative, works from your draft, included in Pro at $12 a month billed annually.

Best for Google Docs: Paperpile, from $4.15 a month with the academic discount. The only tool here built for Docs rather than ported to it.

Best for an unusual style: QuillBot. Over 1,000 styles free, which covers OSCOLA, Vancouver and most departmental house styles.

Best if your lab runs on it: EndNote 2025. Buy it when the institution provides it or the shared library demands it, not otherwise.

Best for citing a web page fast: MyBib. No account, no advertising, no trackers, updated for Chicago 18th.

Worst option, unambiguously: asking a general chat assistant for a reference list. See the accuracy section.

Frequently asked questions

What is the best AI citation generator in 2026?

Scribbr’s Citation Generator is the best free option for most people in September 2026, because it is ad-free, covers APA, MLA, Chicago, Harvard, AMA, IEEE and ACS, builds each reference from real source metadata rather than from a language model, and generates the in-text citation alongside the reference-list entry. Once your project passes roughly twenty sources, Zotero is the better tool, because it stores the sources and PDFs as well as formatting them and supports over 9,000 styles for free. If you want an AI agent that works from your draft rather than from a source you already have, Grammarly’s Citation Finder is the strongest option and is included in Grammarly Pro at $12 a month billed annually.

Is there a free AI citation generator?

Yes, and the free tools are genuinely competitive with the paid ones for the formatting job. Scribbr, QuillBot, MyBib, BibGuru and ZoteroBib are all free citation generators covering the major styles, and Zotero and Mendeley are free reference managers that format citations and store your library. The paid tier only becomes worth it for specific needs: Paperpile from $4.15 a month for Google Docs, Grammarly Pro at $12 a month for the Citation Finder agent, or EndNote at institutional scale. Nobody writing an undergraduate essay needs to pay for citations.

Can ChatGPT generate accurate citations?

Not reliably, and this is the most important thing on this page. Walters and Wilder’s 2023 study in Scientific Reports, across 42 multidisciplinary topics, found 55 per cent of GPT-3.5’s citations and 18 per cent of GPT-4’s citations were entirely fabricated, and among the citations that were real, 43 per cent and 24 per cent respectively contained substantive errors in volume, issue or page numbers. Newer models and retrieval-grounded modes reduce the rate but no published figure reaches zero. The safe workflow is to use a chat assistant to identify candidate sources, verify each one exists, then paste the DOI or title into a lookup-based generator such as Scribbr or Zotero to produce the formatted reference.

What is the difference between a citation generator and a reference manager?

A citation generator takes one source and returns one correctly formatted reference; you copy it into your document and the tool forgets it. A reference manager stores the source, the PDF, your notes and the citation together in a library, syncs across devices, and inserts citations into Word, Google Docs or LibreOffice while rebuilding the bibliography automatically as you write. The practical threshold is around twenty sources or any project lasting more than a few weeks. Below that, Scribbr or MyBib is faster; above it, Zotero saves hours and prevents the reference-list drift that happens when a document is edited over months.

Do citation generators support Chicago 18th edition?

Not all of them, and this is currently the most common source of wrong output. The Chicago Manual of Style’s 18th edition was published in September 2024, and its two most visible changes are that the city of publication is no longer given for books and that sources with more than six authors list only the first three in the bibliography. MyBib and Grafiati are among the free generators confirmed updated for the 18th edition. Test any tool before trusting it by generating a book citation: if the output reads “New York: Norton, 2023” with a city, the tool is still producing 17th-edition form.

Is APA 7th edition still current in 2026?

Yes. The seventh edition of the Publication Manual of the American Psychological Association, published in 2020, remains the current edition as of September 2026, and no eighth edition has been announced. Any tool, template or guide advertising “APA 8th edition” is either mistaken or describing an unofficial product. The same applies to MLA: the ninth edition of the MLA Handbook, published in 2021, is current, and listings for an “MLA 10th edition” almost always refer to third-party textbooks about MLA rather than to the handbook itself.

How do I cite ChatGPT in APA format?

APA 7th edition treats a generative AI tool as software, so the company is the author and the tool and version are the title, with a bracketed description: OpenAI. (2026). ChatGPT (5 September version) [Large language model]. https://chat.openai.com. The prompt does not go in the reference. Describe the prompt in the body of your paper, and if the output you are discussing runs longer than a sentence or two, place the full exchange in an appendix and cite it in text as (OpenAI, 2026; see Appendix A for the full transcript). Check your institution’s policy on whether the use itself is permitted before worrying about the format.

How do I cite ChatGPT in MLA format?

MLA 9th edition puts the prompt in quotation marks as the title of the source, followed by the word prompt, then the tool in italics, the version, the publisher, the date you generated the content and the URL — for example: “Identify the main themes in The Lord of the Rings trilogy” prompt. ChatGPT, 24 May version, OpenAI, 12 June 2026, https://chat.openai.com/share/a80e77ac. MLA does not treat the AI as an author. It also asks you to acknowledge functional uses that do not produce quoted text, such as translation or proofreading, in a note or in your prose.

Will my university know if I used a citation generator?

Using a citation generator to format a reference is not an integrity issue at any university we are aware of — it is the same category as using a spell checker, and many libraries link to Zotero and Scribbr directly from their guides. The thing universities do detect, and increasingly do look for, is fabricated references. Markers now paste suspicious titles into Google Scholar, and institutions have begun running submissions through citation checkers that cross-reference every entry against CrossRef and PubMed. A formatting error looks like carelessness; a citation to a paper that does not exist looks like fabrication, and is treated as such.

What is a hallucinated citation?

A hallucinated citation is a reference produced by a language model to a source that does not exist. It typically combines real authors, a real journal and a plausible year into a paper nobody wrote, sometimes with a DOI-shaped string attached that resolves to nothing. Research published in Publications found these are patterned recombinations rather than random inventions, with duplication across outputs in nearly 30 per cent of cases, which is why they read as familiar and pass a quick review. A related failure is the “chimera” reference, where a real DOI is attached to fabricated metadata, so the DOI resolves but to a different paper than the one cited.

Can a tool check whether my citations are real?

Yes. A citation-checker category emerged in 2026 doing the reverse of a generator: it takes a finished reference list and verifies each entry resolves against bibliographic databases including CrossRef, PubMed, arXiv, OpenAlex and Google Scholar. Citely, SwanRef and the open-source RefCheckAI all work this way, and they also target chimera references where a valid DOI carries wrong metadata. Their published accuracy figures are vendor claims with no independent audit behind them, so the sensible use is to shortlist suspicious entries for manual checking rather than to treat a clean report as proof. For a list under about thirty references, resolving each DOI by hand is faster and definitive.

Which citation generator is best for Harvard referencing?

Scribbr and QuillBot both support Harvard, but the honest answer is that no tool can be fully correct for Harvard, because Harvard is not a single style with a governing body. It is a family of author-date conventions that differ by institution and sometimes by department, which is why two generators produce two different “Harvard” outputs from the same source and both can be defended. Generate the reference with Scribbr, then check it against your department’s own referencing guide and correct the differences. If your institution publishes a CSL style file, load it into Zotero and the problem disappears permanently.