Late to the thread but I have been very satisfied with these for my research.

• Scite AI Research Assistant is superior for verifying if claims are actually supported by other authors.
• Scholarcy Premium Library works better for breaking down massive lit reviews into digestible parts. No complaints about either regarding safety and theyre both under the 30 dollar limit.

I was digging through some old research folders earlier and remembered the headache I had with hallucinated DOIs last fall. Honestly, once you see a fake link that looks 100 percent real, you stop trusting the base models for anything serious. I had to pivot my entire workflow because I couldnt risk my reputation on a bunch of made-up sources. I ended up settling on a stack that actually hits the APIs for academic databases:

• Elicit AI Research Assistant Plus is basically a godsend for lit reviews because it extracts data directly from the papers.
• Consensus AI Search Engine Premium works well when I just need a quick answer backed by peer-reviewed studies.
• Zotero Reference Manager 7.0 is what I use to verify every single metadata entry manually. The Elicit seat is around 25 bucks a month so it fits your budget perfectly... definitely better than spending hours chasing ghost citations.

Oh man, I totally feel your pain with those fake citations because it is so scary when you realize half your bibliography is a total lie! I found a few tools that are actually reliable and wont get you in trouble and I love them because they saved my sanity during my last project.

• Perplexity AI Pro Monthly Subscription
• This is $20 a month and it is amazing because it cites everything from the live web and lets you set a custom writing tone.
• Consensus AI Search Engine Premium
• This one is about $10 a month and it only pulls from peer-reviewed journals. Fantastic for staying safe! Honestly, Perplexity AI Pro Monthly Subscription is my absolute favorite for speed and accuracy. Youll be fine, you got this!

Unfortunately, general purpose LLMs are pretty disappointing when it comes to actual academic integrity since they are basically just predicting text strings instead of querying real databases. I had issues with those fake citations too until I looked into how RAG pipelines actually function. You basically need a tool that forces the model to look at a vector database of peer-reviewed papers first. You should look into Consensus Premium Research Assistant because it costs around 20 dollars a month and specifically queries the Semantic Scholar API of over 200 million papers. If youre doing a heavy lit review, Elicit Plus Research Plan is even better for under 15 bucks. It extracts specific data like methodology or sample sizes directly into a table so you can verify if the AI is hallucinating immediately. To fix the robotic tone, I usually use their custom instructions to mimic a human draft. Its a grind for sure, but these tools should keep you from pulling your hair out before the Philly deadline.