Research AI effectively leverages the power of Statista’s vast and diverse statistics database, making it an excellent feature for existing subscribers. But using it offers less of a conversation and more of a one-question pop quiz.
Research AI, the new large language model (LLM) tool developed by Statista, searches an extensive statistics database and summarizes those statistics based on user-developed prompts. While the tool excels at searching the database and returning relevant materials, the overall functionality is limited by the information housed in the database. This is a great tool for social scientists and those working in business, marketing, and investing, and can help researchers quickly find highly relevant sources.
Research AI is an artificial intelligence LLM search tool that functions on top of the extensive Statista database, the main strengths of which include the following industrial sectors:
In response to a user’s question, an algorithm accesses the Statista database and identifies the 10 most relevant data sources. The LLM then crafts a summary of the information found in those sources. The LLM’s response contains citations linking to source infographics, reports, and data. Research AI also crafts follow up questions, which Statista calls “Extended Query Recommendations,” that allow a user to investigate the data more thoroughly and creates a list of related reports that is similar but not identical to the list of sources that appears in the LLM’s response.
While a majority of Statista’s detailed data focuses on the United States and a few other Western countries, a limited data set provides international context for global issues that impact industrial sectors, like climate change, green energy, and population demographics. The distribution of data is not even. For example, if a company is interested in marketing to the LGBTQIA+ community in various countries in Africa, Statista provides significant data on LGBTQIA+ rights in South Africa but very little data on those rights in Zambia. This is likely due to a lack of curation in some of these countries.
The tool’s performance on Diversity, Equity, Inclusion, and Accessibility (DEIA) issues is also dependent on the data. For example, when asked about disability rights in India and Pakistan over the past 20 years, the LLM was able to provide a helpful answer, albeit one that focused exclusively on India. When asked to project how those rights might change in the future, the LLM instead gave an answer on gender (See Figure 1).
Research AI is clearly designed with current Statista users in mind. People who value Statista’s data and reports will likely appreciate the new AI interface, which excels at finding relevant sources and summarizing those sources quickly. Because it links back to the sources, it will likely prove valuable for people who struggle to search the Statista database and those who need help finding, narrowing, or exploring a topic.
For users who already have a subscription to Statista, Research AI appears as a tab on the main Statista website. On a desktop or laptop, the main landing page for Research AI provides a list of “quick start” prompts that investigate trending research questions, though it’s difficult to tell how often these prompts change. The landing page also features a History panel, which displays a list of the user’s previous, paraphrased prompts. Unfortunately, it’s impossible for Statista to display this information for anyone who is logged in via an institutional IP address.
A second, static history panel on the far left of the desktop page is similarly limited.
The left side of the desktop page also features a “Getting started” guide—a two-minute video that is more of an ad than a how-to. It should also be noted that as of July 2024, the video’s closed captions were poorly crafted. The video does describe how the LLM’s responses provide clickable links to data, one of the tool’s more user-friendly features.
If the user needs to craft their own prompt, they can use either of two “+ New Query” buttons, one on the left and one in the upper right corner of the desktop page, which pull up the prompt text box.
A bolded disclaimer tells the user that the AI can make mistakes, and that information should be validated. This message also appears at the end of each response.
Research AI provides a “prompt-o-meter” status bar at the bottom of the prompt box that encourages users to provide more detailed and precise information to elicit more helpful responses. In practice, however, because the prompt-o-meter doesn’t interpret the content being typed into the box, it ends up working more like a character counter: users can type anything into the prompt box (see Figure 5) and the tool will tell them that it is a great prompt, which is inaccurate (and potentially confusing to new users).
Like many LLM tools, Research AI takes a few seconds to actually craft a response. While you wait, a checklist appears, describing the tool’s progress—a bullet point version of the status bar.
The LLM’s response begins with 3 of the statistics mentioned in the returned reports, which appear in large boxes that are, unfortunately, not linked to their sources. The full response appears below these boxes with clearly marked citations, followed by the list of sources with links.
A couple of tools can help users integrate the LLM’s response into their research. First, a link at the top of the screen can be shared with anyone, whether or not they have access to Statista. A user can also copy the response in its entirety, although, for reasons I could not determine, some source links do not get copied. There is also a refresh or re-prompt button, although when a prompt is refreshed, the same response is “generated,” almost word for word.
At the bottom of the prompt there are two panels, one that suggests extended prompts to gain more information, and one that supplies curated content.
The extended prompts on the left explore nuances of the initial prompt. However, depending on the depth of the statistics, many of these prompts return answers very similar to the original response. It’s unclear if the prompts are also AI generated, or if they are a selection of human crafted questions curated. Either way, the prompts can be refreshed via a button at the top of the panel which provides further, or slightly reworded, follow up questions.
One of the tool’s more useful features is the curated content panel, which provides links to related reports so that users can do a deeper dive. This is similar to the “recommended articles” feature in other databases in that it can help the user more thoroughly explore the data or prompt the user down a different, more fruitful track. Arrows allow the user to scroll through multiple pages of resources, if available.
It’s important to note that, unlike other AI tools, Research AI is less a conversation and more a one question pop quiz. You can create follow up questions through the “+ New Query” button, or click on the extended prompts, but there is not a prompt box below the response, and once you ask a new question, all previous answers disappear.
For existing Statista subscribers, the Research AI tool appears at the top of the database home page as a new tab, alongside familiar tabs like Statistics and Reports. Unlike other databases, Statista has no current plans to charge for the tool.
New users can register for a free Statista account, which offers access to approximately 7% of the total statistics in the database. All three available account types—Business, Academic, and Private—offer statistics as graphics and allow users to save their favorite searchers. The business and academic options also provide access to statistics through an XLSX file and offer advanced search features.
To access the full statistics suite, including Research AI, a business can pay $199 per year for an individual or $959 per year for a five-person team. The individual account has limits: users cannot publish statistics, use the automatic citations feature, or access Statista Reports or Market Insights. These features, as well as on-demand training, are offered under the team license.
Universities, schools, public libraries, and governmental organizations can also obtain a license that will give access to all members of those respective communities. You must contact Statista to get a price estimate.
A quick note on the registration process: Statista asks you about your gender and they only give two choices—male and female. It’s possible they use this information to come up with honorifics, but that is a guess. It seems odd to ask for gender rather than the honorific.
While COUNTER statistics and internal user stats are kept and provided for the Statista database, a Statista representative was unsure if any usage stats related specifically to Research AI are available to subscribing institutions.
IP access and remote access are available to universities and schools. Shibboleth, SAML, and OpenAthens are the most commonly used authentication tools, but other methods are also supported on a case-by-case basis.
Institutions with campus licenses have the right to download Statista content and publish that content for non-commercial purposes within the context of their institutions.
A plethora of AI tools are currently available, with more being developed and released every day. While I’m unaware of any other AI tools that fit within the exact niche of Research AI, there are many search and summary tools for databases, and even more that analyze statistics.
Scopus (Scopus AI, n.d.), Web of Science (Buckland, 2023), and JSTOR (“Explore New,” n.d.) have all developed AI tools to search and summarize the literature in their collections. These summaries are often less precise than Research AI because they deal with scientific writing—which contains more uncertainty and hedging—as opposed to raw data. However, both scan the literature and report on resources for further reading.
As a field, statistical analysis is rich with tools because AI can parse quantitative data quickly and easily. Both Survey Monkey (“SurveyMonkey Announces,” 2023) and Poll the People (“10X More,” n.d.) have AI tools that allow for summaries of original survey data.
And there are countless tools that help people understand data. Most of these are designed for original data, though, and are less effective at searching data that has already been collected and analyzed.
As the popularity of AI tools surges, databases that don’t develop their own may be left behind in favor of databases that have AI tools and make them easier to use. This pressure is likely to push databases that don’t really need AI tools to develop them.
This is particularly true when it comes to LLMs that are simply search skins over the functionality of a database. The power in these tools is their ability to quickly skim the information and return the salient points, saving the reader hours they might otherwise spend wading through the literature. There are, of course, problems with this method—including the possibility that students won’t read the cited articles, that readers will miss context or subtext that is excluded in the summary, or that a researcher will end the search after using AI and assume there are no other relevant results—but it seems as if most AI tools are headed in this direction.
Statista already has two tools that help users understand the statistics that are the backbone of their content—Reports or Insights. Reports contain over 20,000 documents that provide an in-depth look into a wide variety of topics. These reports highlight the main questions or issues explored, use statistics to answer these questions, and focus on trending topics. At the time of this writing, the trending topic was “Sober Curious?”, an investigation into how UK and US consumers are exploring a “mindful approach to drinking.” Insights explore potential revenue and growth, consumer attitudes and behaviors, and business information for companies and digital marketplaces.
While Research AI might help undergraduates halve the time they spend developing research papers, for those who still need to fully ingest the literature, including market and business analysts and public policy researchers, Reports and Market Insights are likely to be the more helpful tools.
There are a few reasons for this. First, because the follow up prompts supplied by Research AI are often very similar to the original question, the answers these prompts produce seem repetitive. As noted previously, available statistics are limited in many areas, so there’s only so much an AI can say before it all starts sounding the same. In the absence of human intervention, many questions are answered with similar summaries.
Second, because the AI uses the data found in Statista, the answers don’t offer insights beyond the statistical. For example, in a search on the long-term impacts of imprisonment in the United States, the answer covered only financial losses, failing to address other considerations—including impacts on mental health, family connections, and community support—that might be more evident in qualitative studies. For students not familiar with these topics, this lack of information, and Statista’s failure to prompt the user to search for this type of information elsewhere, may undermine a more holistic understanding of complex issues.
Which is to say: those who use this tool need to understand that it only investigates statistics, which, by their nature, can’t provide a complete picture of a topic. Other tools, such as Perplexity, or even larger literature databases that collect a variety of data sources, may better address this gap.
That said, several useful features of Research AI might appeal to Statista users. While the Extended Query Recommendations help the user about half the time, the Curated Contents Recommendations consistently push the user to explore more deeply or consider new aspects of a topic. The information summary can help a user determine if there is enough data available to justify a line of research or point to gaps in current literature. Where the research exists, the summary can also suggest sources for further reading.
Like many AI tools, Research AI raises the specter of plagiarism; to help instructors validate answers, Statista makes it especially easy to both copy a prompt and provide a link to the prompt.
Despite the need for more testing of the platform, overall, Research AI is very easy to use—a simple prompt and answer search. However, with this simplicity comes frustration at the inability to have a “conversation,” as is possible with other AI tools. It would also be helpful if all the statistics used, including those used to craft the graphical buttons at the top of the summary, were clickable links to the original sources. I found the checklist “status bar” a little trying when conducting many searches back-to-back.
Per the representative I spoke to, the whole of the Statista database, including Research AI, is currently being evaluated with a Voluntary Product Accessibility Template (VPAT), though there isn’t a deadline for completion. I was unable to confirm whether any accessibility testing was conducted before Research AI was released.
It’s difficult to provide a critical evaluation of price, since Research AI is only one tool embedded among several, all of which explore Statista’s robust statistics. Overall, the value will depend on whether all of the other tools, as well as the data, will be useful to a particular subscriber.
For researchers and students who already subscribe to Statista and use it regularly, Research AI is an excellent tool that will help them save time. It is easy to use, and while its responses may seem repetitive, it surfaces highly relevant and trustworthy sources. Potential users include undergraduate and graduate students, professionals working in business, marketing, or demographic or social science research, and maybe even the general public.
But while Research AI effectively leverages the power of Statista’s vast and diverse statistics database, it doesn’t on its own justify a subscription. In other words, the tool keeps Statista at the top of its game, but it won’t entice new customers.
10X More Effective Surveys, Powered By ChatGPT. (n.d.). Poll the People. Retrieved July 29, 2024, from https://pollthepeople.app/
Buckland, F. (2023, November 7). Web of Science Research Assistant—Bringing Generative AI. Clarivate. https://clarivate.com/blog/bringing-generative-ai-to-the-web-of-science/
Explore new technologies on JSTOR. (n.d.). JSTOR. Retrieved July 29, 2024, from https://about.jstor.org/gen-ai/
Scopus AI: Trusted content. Powered by responsible AI. (n.d.). Elsevier. Retrieved July 29, 2024, from https://www.elsevier.com/products/scopus/scopus-ai
SurveyMonkey announces AI generated surveys. (2023, October 25). SurveyMonkey. https://www.surveymonkey.com/newsroom/surveymonkey-announces-ai-generated-surveys/
10.1146/katina-100124-6
