FAQ

No. Biblibuddy is designed for ad hoc checking and does not store your academic content. Only individual bibliographic references are saved as separate entities, and these are simply metadata facts.

Storage is expensive, I'm running this on a shoestring budget and your document text isn’t required beyond internal consistency checks, and the paper itself is processed in memory only to identify in-text references and the bibliography in order to then work on those. In fact, for online validations or bibliography exports to Excel, you make my life a lot easier and cheaper by not uploading a full doc and only pasting in the bibliography in the first place!

Minimal operational metadata (e.g., your email, when you logged in, when and which run/s happened, basic counters, and basic logs) for analytics and debugging, along with individual bibliography references in online validations for cache storage.

Your actual scholarly content could not be less relevant to me or this app (no offence!).

Biblibuddy begins with a series of deterministic checks to identify the bibliography section - headings, density of 'bibliography-like' lists that have distinct patterns vs normal sentences., and then uses a tiny custom-trained machine learning model that runs in miliseconds in memory to identify where the bibliography starts and ends (the latter is important to avoid poisoning with appendices).

Only then is that identified section extracted for subsequent processing. Some competitors use a cloud AI for this process - not only does this break trust, an LLM having to see an entire doc becomes expensive and slow.

This is something you need to check against your own institutional guidelines.

While the online validation tool processes the bibliography alone, I strongly recommend being transparent if you intend to use the app for checking others' bibliographies: obtain student consent and follow your university's / department’s policies before uploading any work that's not your own to a third-party tool.

Note that unlike some similar tools, Biblibuddy never claims a reference is 'fake' or 'hallucinated', simply that it couldn't be found via standard scholarly databases and web search. there are many valid reasons why a reference can't be found - the original URL or title might have changed, the source itself could be private, or the entry just had a typo that breaks the search. Biblibuddy is designed to aid and improve referencing rigour by providing reports for review, not to police or accuse.

It compares your Harvard-style in-text citations against your bibliography and flags mismatches: missing bibliography entries, uncited references, and near-miss typos.

It tries to verify that each bibliography entry is findable using scholarly indexes and web searches, then marks entries as verified, possible or 'not found'.

Some items are genuinely hard to validate automatically. Since original scholarly publishers' sources cannot be accessed directly, Biblbibuddy uses metadata clues in scholarly citation databases and web search snippets to find the best match it can. Sometimes a snippet won't contain the year (or reference a later year's publication, or have a truncated title or be missing author names). Corporate and organisational sources also often don't have all the usual bibliographic metadata that academic works do; you will see these show up in amber with a lower confidence score even when the match looks like a slam dunk to you.

In 99% of cases the report will link to the matching source, with authoritative publishers, universities and major organisations ranked higher. In some cases there is no direct link available in any search results, and the best match is a second-hand citation in another work. This is normally enough to validate the reference, but if you need to be 100% sure you may need to track down the original source manually.

No. Biblibuddy produces a clear report so you can make the corrections yourself. I am a 'human in the loop fundamentalist' when it comes to AI. Beware any AI service that edits your work directly no matter how convenient it looks!

This is my personal passion project funded by my own extremely limited income. AI let alone hosting, paymentfees and taxes cost money. When an AI-based startup gives you free access, it means they have plenty of generous VC backing to cover their costs while they build market share and figure out how to monetise you later.

Biblibuddy enforces multiple layers of protection including SSRF prevention, DNS validation, redirect chain verification, IP allowlisting, and content validation, to protect both the app and users. IT departments are welcome to request specific technical details by emailing lee@automager.co.uk

Hello. My name's Lee Mager. I've worked in higher education since 2004, including 6 years with a second job in academic publishing / peer review. I did my MSc and PhD at LSE and I am 'old school' when it comes to referencing quality because I believe it's the most important foundation for building scholarly knowledge through a verifiable, historical audit trail. I currently work on AI and automaton at LSE across research, teaching and operations. My LinkedIn tagline is "My mission is to seek and destroy faff wherever it exists, and inspire others to do the same." Connect with me there, I make lots of posts about the good, bad and ugly of AI in academia, as well as occasional rants about sloppy referencing.