Profound vs Citation Junction: which handles citations better?

Everyone's buzzing about AI that can slap a bibliography together. But let's be honest—most of them are just fancy parrots. They'll generate a plausible-looking citation format while quietly hallucinating a journal volume number or inventing an author wholecloth.

I decided to run a little stress test on two tools that claim to excel here: **Profound** (the new "academic-focused" assistant) and **Citation Junction** (the established browser plugin). The prompt was designed to be *just* tricky enough:

> "Generate APA 7 citations for the following three sources: 1) A pre-print from arXiv with the ID `arXiv:2403.12345`, titled 'Large Language Models are In-Context Reasoners,' uploaded March 2024. 2) A chapter titled 'The Ethics of Synthesis' from an edited volume 'AI & Society,' editors P. Chen and M. Okada, published by MIT Press in 2022. The chapter author is Jordan Lee. 3) A website article from 'The Gradient' with no publication date visible, titled 'Post-Training Modifications,' author Sam Rivera, accessed today."

Here's what I got back.

**Profound's Output:**
```bibtex
Lee, J. (2022). The Ethics of Synthesis. In P. Chen & M. Okada (Eds.), AI & Society (pp. 145-167). MIT Press.

Anonymous. (2024, March). Large Language Models are In-Context Reasoners. arXiv preprint arXiv:2403.12345.

Rivera, S. (n.d.). Post-Training Modifications. The Gradient. Retrieved May 22, 2024, from https://thegradient.pub/post-training-modifications
```

**Citation Junction's Output:**
```bibtex
Author(s). (2024). Large Language Models are In-Context Reasoners. arXiv. https://arxiv.org/abs/2403.12345

Lee, J. (2022). The Ethics of Synthesis. In Chen, P., & Okada, M. (Eds.), AI & Society. MIT Press.

Rivera, S. (2023, August 15). Post-Training Modifications. The Gradient. Retrieved May 22, 2024, from https://thegradient.pub/post-training-modifications
```

Now for the autopsy. Profound gets points for using "Anonymous" for the arXiv preprint, which is correct APA 7 for when no author is listed, and for correctly handling the `(n.d.)` for the undated website. However, it failed to include the arXiv URI, which is a critical omission. It also formatted the book editors' initials incorrectly (should be Chen, P., & Okada, M. Y. — but we don't have M.'s full initial, so Profound's guess is as bad as anyone's).

Citation Junction, meanwhile, completely whiffed on the arXiv spec. It invented an author list where none was provided, which is a cardinal sin. It also *invented a specific date* for the website article (August 15, 2023) out of thin air, which is arguably worse than just saying `n.d.`. Its handling of the book chapter was slightly more standard, but it missed the page numbers.

The takeaway? Profound understands the *rules* better but fails on the *completeness* test. Citation Junction is lazy, filling gaps with confident fabrications. For a quick draft, maybe Profound. For anything you actually need to submit, you're still going to be elbow-deep in the APA manual. They've automated the easy 80%, but the remaining 20% will get your paper flagged.