Testing AI-Tools: ResearchRabbit

With the "Deus Ex Mach­ina? — Test­ing AI Tools"-series we want to show you dif­fer­ent tools, that aim to sim­pli­fy writ­ing, design and research by using Arti­fi­cial Intel­li­gence. More on the "Deus Ex Machina?"-series can be found here.


"Reima­gine Research" is what the Resear­chRab­bit homepage declares as their motto. As the name sug­gests, it is a tool for sci­entif­ic research that is inten­ded to make the search for suit­able sources and sci­entif­ic input more intu­it­ive and to sim­pli­fy the dis­cov­ery of rel­ev­ant papers. The developers call Resear­chRab­bit a "Spo­ti­fy for Papers", which, with reg­u­lar use, is sup­posed to get to know the search cri­ter­ia of its users and thus optim­ize its suggestions.

To use Resear­chRab­bit, users start with a "seed paper", i.e., a sci­entif­ic text that is marked as the begin­ning of the search. Start­ing from this, the pro­gramme sug­gests fur­ther, match­ing sci­entif­ic texts. These texts can be linked to each oth­er either by com­par­able sub­ject areas, authors or cita­tions, so that the texts sug­ges­ted by Resear­chRab­bit com­ple­ment an already exist­ing text "col­lec­tion" (col­lec­tion of papers on a topic).

Texts that have been assembled into a col­lec­tion can be examined more closely in a second step: in addi­tion to the authors, the pub­lic­a­tion date and medi­um, Resear­chRab­bit shows an abstract and how often the dis­played paper has been cited in oth­er sci­entif­ic works. This is one of the strengths of the tool: with just a few clicks, you can find out who has been cited, who the authors are and in what con­text these cita­tions occur. The side-by-side list­ing of research paths (which can the­or­et­ic­ally be con­tin­ued ad infin­itum) makes it easy to trace the search pro­cess even in ret­ro­spect and, if neces­sary, to adjust it again in the middle of the path.

Bild 1: Screen­shot from Resear­chRab­bit. Shown are the vari­ous options for research with­in the tool.

But its not only the cita­tions that can be tracked via Resear­chRab­bit; the tool can also be used to track which texts the authors of the "seed papers" refer to via the "All Ref­er­ences" but­ton. Under "These Authors", users can take a closer look at the authors of the texts and, for example, search for fre­quent col­lab­or­at­ive rela­tion­ships among authors or view which works were pub­lished before or after the selec­ted text.

Resear­chRab­bit dis­plays these con­nec­tions between the dif­fer­ent authors in a net­work cluster ("author net­works") with con­nect­ing lines between authors linked by their research work. It even visu­al­ises how often authors have col­lab­or­ated: the thick­er the grey con­nect­ing line between them, the more joint papers Resear­chRab­bit has found. This also makes it pos­sible to see which authors are or were par­tic­u­larly influ­en­tial in their field.

How­ever, Resear­chRab­bit only dis­plays the texts that it finds by itself with its search algorithms. There­fore, one should not assume that the search res­ults presen­ted are com­plete. Like­wise, not every text is read out com­pletely, so some con­nec­tions between texts only emerge when you read through them yourself.

The AI behind the application

The algorithms on which the Resear­chRab­bit AI is based have not yet been pub­lished, neither have the data on which they were trained upon. Even when asked, the Resear­chRab­bit team did not provide an answer. The FAQs point out that ResearchRabbit's search engine uses the search algorithms of the NIH (Nation­al Insti­tute of Health, USA) as well as the NIH-based ser­vices PubMed.gov and the Nation­al Lib­rary of Medi­cine, as well as those of Semant­ic Schol­ar, and thus gen­er­ates sug­ges­tions for papers.

ResearchRabbit's own algorithm is trained to make sug­ges­tions for the texts selec­ted by the users, to extract the rel­ev­ant metadata from these and to make fur­ther text sug­ges­tions. These are divided into three cat­egor­ies: Sim­il­ar Work, Earli­er Work and Later Work. Earli­er and Later Work refer to the cita­tion rela­tion­ships between indi­vidu­al texts; accord­ing to Resear­chRab­bit, tem­por­al data (pub­lic­a­tion date) is also read out here to gen­er­ate an over­view for the user. "Sim­il­ar Work", on the oth­er hand, is detached from this and, accord­ing to the FAQs, is not con­gru­ent with Earli­er and Later Work, but refers more to oth­er rela­tion­ships between the texts, for example the cita­tion net­works. Accord­ing to its own inform­a­tion, Resear­chRab­bit ini­tially loads only the 50 most rel­ev­ant texts to a search query in order to keep the research as focused as possible.

Incid­ent­ally, Resear­chRab­bit takes its name from the idea of diving deep­er and deep­er into the next research "rab­bit hole" (a ref­er­ence to Alice's jour­ney down the rab­bit hole in Lewis Carol's "Alice in Won­der­land") with the help of the research flow that the tool's design is sup­posed to promote.

The rhetorical potential of the tool

In the research phase, Resear­chRab­bit is a sup­port in find­ing suit­able texts and inter­est­ing ref­er­ences for one's own aca­dem­ic work. In par­tic­u­lar, the "works cited" func­tion and the author net­work can give an indic­a­tion of which texts are rel­ev­ant in the field being researched and have gen­er­ated a cor­res­pond­ing echo. Nev­er­the­less, the qual­ity of the texts and their rel­ev­ance to one's own top­ic must be checked manu­ally, since the pro­gram does not provide any assess­ments of the qual­ity of the texts. Fur­ther­more, users can add com­ments to the texts them­selves, but not view the com­ments of oth­er users.

The same rule of thumb applies to this tool that should also apply to oth­ers: if pos­sible, one should start using the tool with pri­or know­ledge of a top­ic and a manu­al search that has already been car­ried out. On the one hand, this is because it is much easi­er to recog­nize texts that are rel­ev­ant to one's own work and to sep­ar­ate them from papers that can­not be used, and because research­ers rel­ev­ant to one's own work can be found more quickly via the author net­works. Oth­er­wise, users may quickly be over­whelmed by the amount of options.

Usage in science communication

Resear­chRab­bit is aimed at research­ers, sci­ent­ists, stu­dents in high­er semesters and oth­er people with a sci­entif­ic back­ground. Import­ant for the suc­cess­ful use of Resear­chRab­bit are pre­vi­ous know­ledge in the researched field and a reflec­tion in advance on the res­ults Resear­chRab­bit should deliv­er in order to keep the search time for suit­able lit­er­at­ure as short as possible.

For each sci­entif­ic text, Resear­chRab­bit provides an abstract includ­ing cita­tion and a link to the respect­ive text: either as a PDF or by for­ward­ing to JSTOR, Spring­er Link and oth­er portals. For those who do not have access to these portals via their own uni­ver­sity or research insti­tu­tion, Resear­chRab­bit will only be of lim­ited use, as the pro­gramme may be able to find many texts, but can only out­put them dir­ectly in rare instances where a PDF is stored in Resear­chRab­bit itself or the text sought has been pub­lished open access.

Pri­or know­ledge also helps the user to feed the "Rab­bit" with mean­ing­ful ini­tial texts that determ­ine the ori­ent­a­tion of the lit­er­at­ure it sug­gests for the "Col­lab­or­a­tion" (text col­lec­tion). At the time of writ­ing, all that is required to use Resear­chRab­bit is an account on the site; the ser­vice is free for research­ers and, accord­ing to the FAQ, is sup­posed to remain free in the future.

Finally, Resear­chRab­bit offers the option to net­work with oth­ers on the plat­form and cre­ate joint text col­lec­tions. This opens up the pos­sib­il­ity to con­duct research togeth­er for a joint pro­ject. Ones own col­lec­tions can be shared with oth­er people and groups for 14 days even if they do not have their own Resear­chRab­bit account. In this case, how­ever, they can only view the col­lec­tions, not edit them themselves.

And finally, the "Linked Con­tent" func­tion allows users to see where texts from their own col­lec­tion are also used in non-aca­dem­ic con­texts — for example, on Wiki­pe­dia, per­son­al blogs or offi­cial web­sites of prac­tice centres. How­ever, the inform­a­tion in Resear­chRab­bit itself is not yet very soph­ist­ic­ated. In addi­tion to the chosen text, in many cases only the type of web­site on which the text was used or named (Word­Press, for example) is dis­played, but not the insti­tu­tion to which the respect­ive web­site belongs.


The name Resear­chRab­bit and the meta­phor­ic­al jour­ney down the sci­entif­ic rab­bit hole is spot on with this tool. The intu­it­ive hand­ling of "dip­ping fur­ther and fur­ther into the rab­bit hole" is facil­it­ated by the fact that inform­a­tion that might be of interest to research­ers, such as pre­vi­ous works or authors cited in a paper, is dir­ectly access­ible with just a few clicks.

How­ever, even Resear­chRab­bit is not free of errors — until the "seed paper" has pro­grammed the algorithm to its own interests, a lot of input has to be giv­en. And not every text searched for can be found by Resear­chRab­bit or is returned by the tool as the first search res­ult, or even as a search res­ult at all — even with a very pre­cise keyword search includ­ing title(s), name of the author(s) and pub­lic­a­tion date. If you keep these points in mind and also use the con­ven­tion­al lib­rary cata­logues and oth­er search masks, Resear­chRab­bit can be a valu­able help in research work and in present­ing text and author contexts.