Genuinely asking because I’ve trialed a lot of these over the past year and my feelings are mixed. I’m in a PhD program, so my use case is pretty specific. I need to organize large amounts of literature, find connections across papers, and manage citations. I don’t need something to write my dissertation for me.
The tools that have actually held up for me are the ones that do one thing well. Summarizing long PDFs into something navigable. Pulling out methodology sections so I can do quick comparisons. Generating a first-pass literature map that I then correct. Every tool that tries to do everything ends up doing nothing particularly well.
Where I’ve wasted time is with tools that market themselves as AI research assistants but are basically chatbot wrappers that hallucinate citations and produce confident summaries of papers they clearly didn’t read. I’ve started testing any new tool by asking it about a paper I know very well and seeing what it gets wrong.
What’s working for other people, especially in contexts where the accuracy bar is high?
The single-function tools you mention are consistent with what holds up in my workflow too. I’m skeptical of the ‘AI research co-pilot’ framing that implies something close to a reasoning partner. What exists right now is useful for organizing and surface-level pattern-finding. Anything requiring deep synthesis still needs a human to drive. That framing matters because it changes how you evaluate and use the tools.
From a professional standpoint, the same principle applies in client research. The tools that surface relevant information quickly are valuable. The ones that try to produce polished analysis are usually worse than a competent junior. Knowing which category a tool falls into before you commit to it in a workflow matters a lot.
for me it’s mostly been the pdf summarizers. i can get through way more background reading in less time. the problem is when i try to cite something from the summary and realize i don’t actually understand the nuance. so now i use them to figure out which papers are worth actually reading, not to replace reading them
In a classroom context I’ve found the tools that help students engage with source material are the ones worth keeping. Anything that lets a student ask questions about a text they’re trying to understand. Not summarizers that let them skip reading, but tools that make the reading more interactive. Those seem to actually improve comprehension rather than replace it.
The citation hallucination test is a good one. I use a similar approach. Give it a paper with an unusual title and see if it can accurately describe it. Any tool that confidently describes a paper it doesn’t have access to is not a research tool. The confidence level is often inversely correlated with the accuracy.