Skip to content
Home » The Effortless Blog » Review: Consensus AI – An Intelligent Literature Review Companion

Review: Consensus AI – An Intelligent Literature Review Companion

The way we extract information for literature reviews has completely changed in the age of AI. While reading and understanding are still essential, finding the correct information quickly becomes accelerated with tools like Consensus. At it’s core Consensus is a search engine with AI built within and around it to speed up your research. It scans millions of papers to find the ones that match your query and sort them by which support or reject your argument or employ specific methods. This makes searching for the right paper faster, more enjoyable, and effortless.


Join the Effortless Newsletter. Receive free tips on note-taking, literature review, AI tools, and other productivity topics specifically tailored for academics and students.


In this review, we will break down the main use cases, limitations, and features of consensus in detail. First, we will look at what Consensus is and what it does, then dive deep into every feature and finish with an even more innovative use case: ConsensusGPT, which combines ChatGPT’s conversational capabilities with Consensus’ expansive database.

  • What problem does Consensus solve?
  • What does Consensus AI do?
  • The Consensus meter for a quick overview
  • Synthesize and CoPilot boxes to grasp a topic
  • Badges/Summaries to sort single studies
  • Filtering options in Consensus
  • ConsensusGPT: Consensus as a ChatGPT plugin
  • What is the difference between ChatGPT and Consensus?
  • Academic writing use-case: Asking Consensus for evidence
  • Academic writing use-case: Filling in Citations

How to use Consensus AI?

Like all AI tools, Consensus is as simple as typing a question into the textbox. You can start using it for free, and if you need it, get the premium membership, which allows you to use AI to summarize findings and systematically analyze single papers (more than one pricing in the last section of this post and a 30% off discount code!). However, it is crucial to understand this technology’s limitations and know which results you can fully trust and which you need to double-check.

To use Consensus you will require a free account. The results have four main elements: Summary, Consensus Meter, Copilot, and about ten papers. If you are not a premium member, you will always see the papers, but the other elements are limited to a certain amount of monthly searches.

Compare the result in the screenshot above with the Google search for yourself (“Does climate change cause range shifts in forests?” ). Despite being a big and somewhat generalizing question, I got a concise reply. The devil is, of course, in the detail with every scientific question, but you see that finding tentative answers to complex questions becomes increasingly simple. If I am brainstorming parts of my research, I might ask dozens of questions like this, and understanding what works in seconds rather than hours makes the research process smooth.

At the end of your search you will find a number of studies sorted by relevance clicking on each of those brings the detailed study view:

This view brings together all the different pieces of Information has on this study which includes metadata, summaries, quality indicators, citation counts and a link to the publication itself.

Can you trust what Consensus AI tells you?

This is a general debate about how much we should trust AI. In my opinion, it is not different from asking a human. If you ask a five-year-old kid a serious question, you will certainly not bet on the correctness. If the kid is 16 and their answer is surprising but not unrealistic, you might trust but double-check. If a professor with 40 years of experience answers a question in their domain, your answer is as firm as it gets. The problem with AI is we don’t know which personas we are talking to.

My rule of thumb is to consider how much data has been published on a topic. If I am asking a question that has been debated for decades, there is likely a large amount of data available, and I tend to trust AI (like in the example question above). The more recent and specialized topic you are interested in, the less you should trust any AI on this. Luckily, Consensus spots these cases sometimes:

Overall, the speed of AI is increasing very fast, and the training datasets are growing daily. Don’t hesitate to be specific and treat Consensus as a new colleague you know is knowledgeable but has yet to talk to them much. Get to know one another.

Questions Consensus AI works best with

To get the most out of Consensus ask supported questions:
✅ Yes / No
✅ Relationship (e.g. Is Sauna beneficial for blood pressure)
✅ Benefits of… (e.g. “Sauna”)
Open-ended questions or ones that require numbers/explanations MAY or MAY NOT work. This is a general case with AI. Questions requiring logical reasoning and thinking rarely produce excellent results (at least as of early 2024). Bad example: “How much climate change can European forests tolerate?”. While you will almost always get a reply from the AI, the value will diminish the more you deviate from questions AI can answer. This is not a particular limitation of consensus, but it will generally apply to large language models since they deal with facts as language rather than logically connected entities.

You can check out the much more detailed guide on the Consensus FAQ page. The best way to find out, however is to just try it out for yourself.

Domains Consensus AI works best with

According to the Consensus website, the best questions are in the medical and social policy domain. I have been using it for my research (forest ecology), and it has always come up with good studies and answers despite needing to be added to this list. It is best to try out the domain you already know and see if the search finds the essential papers you know are central to it.

However, the interface is designed with medical studies in mind. For example, you can filter by human/animal studies or find randomized control trials, concepts that don’t exist in ecology. However, the UI will likely change in the future as Consensus gets more adopted and expanded across all domains.

Limitations of Consensus AI

The two most significant limitations of all AI tools are incompleteness and irreproducibility. Incompleteness means that the results you get are filtered by an AI model with a certain level of randomness. We don’t truly understand how this process works, so it might sometimes miss something that a human would deem necessary. This is especially true since we need to know which papers Consensus can access.

This brings us to the second issue: Irreproducibility, which is of particular concern when conducting a systematic literature review. Such a review involves documenting your paper search process, theoretically allowing for replication. However, if you were to state, for instance, ‘I searched for X on Consensus’, and someone attempted to replicate this a year later, their findings might differ from yours. This variation directly results from the stochastic nature inherent in all AI models.

Remember, searching for an AI is like asking somebody for directions to a restaurant. You will generally provide the same reply to different people, but you will never use the exact words and add/omit specific, more minor details.

The Consensus Meter for yes/no Questions

The Consensus Meter consists of three bars: Yes, No, and Possibly. The way it works is that the Consensus AI first finds several papers related to your query, scans them, and then sorts them into one of the three categories. If most papers affirm your question, the yes bar will grow. This is why Consensus invites you to use yes/no questions. Otherwise, the meter would make little sense. To use this feature, you need to enable the “Synthesize” checkbox right below the search bar. You get 20 of those summaries for free monthly and must purchase a premium account if you need more.

This feature is handy even when researching what supplements to take or what exercise to prioritize daily. In my experience, I tend to trust this reply over clickbait Google articles written primarily for SEO (Search Engine Optimisation) and “approved” by somebody with a doctoral degree, as their purpose is mainly to rank high in the Google search and only secondary to providing scientifically correct output.

Synthesize and CoPilot for Topical Overview

The synthesize (or summary), and Copilot features are more in-depth AI features that process the search results (i.e., the top 10 papers answering your question) and summarize them. The summary feature creates a very short summary (~50 words), while Copilot creates a lengthy text of about 400 words with references to the search results below.

To use these features, activate the “Synthesize” and “Copilot” checkboxes right below the search bar in Consensus. Be aware that they both cost “AI credits.” 20 AI credits are free every month. Synthesize costs one credit per query, and Copilot costs two. If you are in a free account, I suggest you save these questions for the relevant questions and use them selectively and judiciously.

The Copilot is 100% free of hallucinations as it can only use the papers retrieved from the database.

Customizing Copilot with queries

While Syntehesize is fixed feature, Copilot can be customized to your needs by adding instructions to it. Think of it as an upgraded ChatGPT version that has access to the contents of all the papers that Consensus found for your query. Essentially this allows you to write a mini literature review. Here, I ask to generate a bullet-point list of the main drivers of forest range shifts. And Copilot solves the task for me:

When we keep the question but change the prompt the result changes. For example, here I ask to draft an outline of a literature review. Notice how useful this can be when starting out with a topic and seeking general guidance on how to solve a problem:

Outline of a literature review with consensus copilot.

However, Consensus is primarily a search engine not a conversational AI. If you are learning a topic and want to have a back and forth conversation use the Consensus GPT bot (section below) instead.

Badges and Study Snapshops to Identify Relevant Studies

Badges and snapshots allow users to investigate a single study of interest. This is where Consensus’s strengths for medical-type questions shine, as the snapshots and badges lean towards medical studies and feature options like “animal trials” or “randomized controlled trials.” To generate a study snapshot, click on “Study Snapshot” right below the paper.

In the medical example above, you immediately see who the test subjects were, how many, which methods were used, and many other details that would likely take a while to find in the paper’s methods section.

Badges follow a similar idea but highlight different important properties of studies. For example, some studies might have more citations, while others employ a particular method. Badges make it easier to find papers that stand out.

There are many different badges and when you hover them with the mouse a small popup explains what they mean.

Filtering Options in Consensus

Since Consensus is primarily a search engine, you can filter your results like on PubMed. However, there is a big difference because Consensus filters the results using semantic search and AI, which allows filtering by content rather than words in the title/abstract.

Especially for medical questions where sample size or human studies are essential distinguishing factors, this search is unique. I hope that one day, Consensus will implement similar filters to search in my domain (ecology), for example, by study location, taxa, or statistical methods.

ConsensusGPT: Consensus as a ChatGPT plugin

Not too long ago, ChatGPT released so-called GPTs or customized AI assistants that use ChatGPT but add the ability to communicate with other services (like Consensus). The result is an entire ecosystem of assistants for hundreds of use-cases. Here are my top 5 GPTs for academic purposes. Consensus has been featured in the GPT store from the first day and remains among the top GPTs, which is a testimony to its usefulness.

To use ConsensusGPT you will need a premium subscription to ChatGPT. Should you get one? I wrote a thread with my thoughts on this.

What is the difference between ChatGPT and Consensus?

While you will find the same papers using ConsensusGPT or a website, you will generally have a different approach to using them. A GPT is just ChatGPT with a few extra instructions and the ability to call other services like consensus to ask, for example, for papers. So, while Consensus provides the data and some instructions, ChatGPT generates the text.

This allows for significant use cases like finding evidence for questions and having long conversations where you alternate between accessing knowledge from ChatGPT and querying papers from consensus. If you are curious, check out the Effortless Literature Review 3 webinar, where I heavily relied on this technique.

The downside is that you do not have access to the Consensus meter or the content of the papers like Consensus does on the website. Overall, this means that you can do a broader literature review that is slightly shallower. However, Consensus GPT provides a link to their website where you can continue to use the features introduced above.

If you want to start building a team of AI helpers for your academic career, check out my tutorial on building GPTs or dive as deep as possible by taking the Effortless AI course.

Is ConsensusGPT free?

Yes it is if you have a ChatGPT Premium subscription. It delivers more or less the same functionality when searching on the Consensus website (which is free there, too). However, to get insights from inside the single papers, you will need to use the Consensus website and their AI to analyse the results in-depth.

Academic writing use-case: Asking Consensus for Evidence

Below is a use case I often use when writing a paper. Usually, I know certain things are true because I have read them many times but don’t have a specific citation in mind to back them up in a published paper. This is where you can use ConsensusGPT. Use the straightforward prompt “Find evidence for …”. You can use The results of several studies as a reference in your writing.

It goes without saying that you should always double check the reply by at least skimming the paper. Keep in mind that often AIs will not have access to the entire PDF but only the publicly available abstract and its conclusions might therefore be incomplete.

Consensus AI vs SciSpace

Both companies have similar capabilities and the same goal. While SciSpace shines at using AI to help you read papers and discover related papers, Consensus extracts essential information from a collection of documents. It displays the final results as a quick, easily digestible summary. Consensus shines regarding Yes/No questions in the medical domain as its UI is optimized towards this, as evidenced by various filter options, the consensus meter, and badges. If you want to learn more about SciSpace, read my deep dive on SciSpace.

Consensus AI Pricing

Consensus works on a so-called freemium model. Some free features, including search results (papers) and badges, are forever free. This lets you know which papers are relevant but uses something other than AI to extract the appropriate insights. For this, you need the premium plan. However, Consensus allows you to try out the advanced features: Summary and Copilot up to 20 times a month. For this so-called AI, credits are consumed, of which you have 20 free every month. You use one of those credits (two for Copilot) whenever you create a summary.

Feel free to use the discount code Effortless30 for a 30% discount (I don’t even get any commission from it, by the way). Students can purchase the premium account at a 40% discount and the general price is 9$ per month.

Summary

Overall Consensus is a novel way to discover literature that particularly shines when you ask it medical yes/no questions as its UI has been designed with these papers in mind. It is equally useful in an academic context as well as just for you personally to answer some health related question.

  • It starts with a plain text query and delivers a set of papers that help answer this query.
  • If you ask it a yes/no question, Consensus will analyze how each paper leans and display that as a small yes/no/possibly graph called the Consensus Meter
  • If you are a premium member, AI will analyze the resulting papers and create an outline with citations in an academic fashion.
  • Consensus is available as GPT in ChatGPT, which allows for unique use cases as you can leverage ChatGPT conversational capabilities and Consensus database of studies.
  • Regular price is 9$ with a 40% student discount or 30% with the “Effortless30” code. Searches are always free, and the AI analysis is what you are paying for. Twenty uses of AI are free per month.

Consensus is featured as part of my literature Review webinar and you can get a recording here:

Workflow for the Literature Review webianr

Leverage semantic and citation search with AI to find the most impactful literature quickly. Uncover reference gaps combining multiple tools. Use ChatGPT assistants to get rid of hallucinations and use AI to aid faithfully and ethically in your lit review and writing process.