Honestly, seeing your advisor freak out about fake citations is basically the new grad student rite of passage lol. Been there. If you want to avoid the hallucination trap without spending a fortune, you gotta look at how these tools actually pull data. Most generic AI just predicts the next word, but research-specific ones use something called RAG to actually look at the text first. I switched to Elicit Plus Plan a few months back because it actually extracts data from the full text instead of just guessing based on the title. It is way better for your urban green space stuff because you can ask specific questions like what was the sample size? or what were the limitations? and it gives you a table based on the actual PDF content. If you're worried about the cost, you can usually get a few searches for free to see if it catches those city planning details you mentioned. For the synthesis part, Scite.ai Assistant is probably the winner for me. It uses something called Smart Citations which tells you if a paper was supported or contrasted by later work. Super helpful for a lit review so you dont accidentally cite something that was debunked in 2022. Another one to check is Connected Papers Visualization Tool. It doesn't write for you, but it builds a visual map of how papers are linked. It's awesome for finding that one seminal paper you might have missed in your Zotero pile. Just remember that even with the best tools, you still gotta be the one to do the final sanity check on the methodology sections... AI is still kinda dumb with local context sometimes. It basically just pattern-matches, so if your Seattle documents are super niche, it might struggle to connect the dots without you guiding it.

Building on the earlier suggestion, I actually disagree that those paid subscriptions are worth the hit to a grad student budget. Unfortunately, I've had issues with the premium tiers hallucinating even on simple tasks, which is just unacceptable for the price. Not as good as expected tbh. If you want to keep your data safe and your wallet full, I'd suggest a different route:

• ResearchRabbit Research Discovery Platform is totally free and doesn't make things up because it maps real citations.
• Zotero 7 Reference Manager now has much better local indexing that doesn't require sending your files to a cloud.
• Semantic Scholar Free Academic Search is far more reliable for verifying if a paper actually exists. I've found that staying local and using free tools is way more reliable than trustin a $20/month black box that ignores your city planning data anyway.

Omg I totally feel your pain!! I had a massive scare last semester when my assistant made up a whole book chapter that didnt exist... yikes! The one I got now is a lifesaver tho:

• It only uses the PDFs I actually upload
• The summaries are legit amazing
• No more fake citations to freak out my prof! It seriously saved my sanity during finals week!

Honestly I have spent way too much time testing these different stacks over the years and the tech literally changes every two weeks. If you want a real side-by-side comparison of how the specialized research brands actually handle city planning data versus generic models, youre better off just searching for academic AI workflow 2024 on YouTube. There is this one creator who did a massive deep dive on this exact topic like a month ago... cant remember the name but it was the top result. They basically show how to bypass those hallucination issues by setting up a local index which is what you actually need for your Seattle documents. Honestly just check out some of the niche subreddits for grad students too because they usually have whole community spreadsheets comparing the accuracy between different brands. Its way more reliable than just taking someones word for it here since everyones dataset is different and what works for one person might fail for your specific city planning docs.

Gonna try this over the weekend. Will report back if it works!