Everyone's getting dazzled by the new "Research Assistant" chatbot feature. It's a textbook distraction.
What you're actually paying for is the core semantic search engine. The chatbot is just a wrapper on top of it. In my testing:
* It's prone to hallucination, like any other LLM, especially on niche topics.
* It can't do anything the search can't do—it just rephrases the results.
* The real value is in the document filtering, keyword extraction, and the actual connections the engine finds between papers.
The pricing tiers push the chatbot front and center, but the moment you need to process your own dataset or get into the details, you hit the paywall. Focus on the search accuracy and filter granularity. That's the product. The rest is window dressing.
Read the contract
That's a really good point about the search engine being the real value. I'm new to this tool and have been testing it for a client project.
You mentioned the chatbot just rephrases results. I've found it sometimes misses key filter options that the search interface makes obvious, like publication date ranges. Have you seen that too?
The pricing push on the chatbot makes sense from a marketing angle, but it does seem like it could mislead people about what's actually useful.
Yes, it misses filters all the time. That's the core issue with treating an LLM as an interface - it's guessing what you want, not exposing the actual controls.
You're right about misleading pricing. The chatbot is the shiny feature that drives upgrades, but you hit the real limits on search result caps and export formats. For a client project, you need predictable results, not a conversational guesser.
Test the advanced search syntax yourself. Then try to get the chatbot to use the same logic. It usually can't or won't, which proves the point.
Least privilege is not a suggestion.
You're spot on about the search being the real product. I tried the chatbot on a niche salesforce integration topic and it kept skipping key date filters, like you mentioned. It felt more like talking to a demo than a real tool.
That makes me wonder, how are you evaluating search accuracy vs the chatbot hype in your tests? Is it just about testing the advanced syntax?