GrantWise
FREE · NO AI INVOLVED · RESULT ON THE PAGE

Check your AI’s hallucinations before an evaluator does

Generative AI writes fluent references that do not exist: a DOI nobody registered, a real DOI attached to the wrong paper, an author who never wrote on the topic, a link that answers 404. Horizon Europe makes you responsible for every one of them. Paste your reference list and find out in under a minute.

CrossRef · OpenAlexDead linksReddit, blogs & social media flaggedHorizon · EIC · Eurostars · Norway

GrantWise › Free tools › AI hallucination check

What is checked

  1. 1

    Does the work exist?

    Every DOI is resolved in CrossRef and OpenAlex. A reference without a DOI is searched by title, author and year in both registries.

  2. 2

    Is it the work you cite?

    The registry record’s title, first author and year are compared with your reference. A DOI that exists but belongs to another paper is the classic AI tell.

  3. 3

    Does the page still answer?

    Each cited web address is requested from our server. Pages that are gone, and host names that do not resolve at all, are reported.

  4. 4

    Will an evaluator accept it?

    Each web host is classified by evidence weight. Forums, social media, blogs, video, press-release wires and AI chat transcripts are flagged as weak even when the page exists.

The invented-reference patterns it catches

DOI does not existPlausible-looking, correctly formatted, registered nowhere. The single most reliable sign of an AI-written reference.
DOI points to a different workA real identifier borrowed from another paper — the title you cite and the title the DOI resolves to do not agree.
No matching record anywhereAuthor, title and year that no registry can find. Sometimes grey literature; often invented. Marked for you to check, never called fake.
Dead or non-existent addressA link that answers 404 or a domain that does not resolve — AI invents URLs on real-looking domains.
Year in the futureA publication date later than next year.
Weak sourceReddit, Quora, LinkedIn, Medium, YouTube, a press release or a market-report press page cited as evidence for a figure.

Nothing here is a model’s opinion. It is your text compared with registry records, which is why the check can be repeated and why it never says more than it knows.

References are the part an evaluator can verify in a minute. The rest takes a review.

Whether the claims hold, whether the evidence sits where the criteria need it, whether the figures agree across sections — a GrantWise review reads the whole proposal the way an evaluator will and names every weakness that remains, with its location in your text and the fix. Start with the free preview.

Start a review

Why this matters for Horizon Europe and the EIC

The Horizon Europe standard application form permits generative AI in proposal preparation and makes applicants fully responsible for what it produces, including references. It also asks applicants to be transparent about the tools used and to be able to list the sources behind generated text. An invented reference is therefore your problem, not the tool’s — and it is the kind of problem an expert evaluator can confirm with one search.

GrantWise is independent and is not affiliated with, endorsed by, or acting on behalf of the European Commission, the EIC, Eureka, Innovasjon Norge, Forskningsrådet or any other funding authority. This check compares references with public registries; it does not read the cited works and cannot say whether a source supports the claim it is cited for.