Politicker is a brand new platform bringing robustly non-partisan data on the UK House of Commons. It is wholly distinct to traditional sources of political journalism and consultancy firms, which rely on policy experts and journalists to determine what's important, and what's not. Relying on experts to select the most useful bits of information is, and will always be, a vital step in communicating politics to citizens and organisations. However, journalists and other communicators can be subject to personal biases; they can also focus disproportionately on certain dimensions of politics, such as conflict, scandal, individual gaffes, or whatever the zeitgeist may be. This has the effect of distorting politics, making it appear more turbulent than it actually is. Most importantly, however, is the fact that most gatekeepers of political information have vested interests in one way or another.
We are not so audacious to claim that our data is 'objective', however it's about as close as you can get.
Politicker filters data from the original source - the UK Hansard database containing transcripts of every word spoken in Parliament - standardises it via a data pipeline and then uses LLM/AI technology to analyse, interpret and construct the output data. Politicker isn't the first to attempt this process. One organisation claims to have built a similar tool directed at the Spanish legislature, while a handful of academics are attempting to use LLMs to analyse political speech and social media content. However, it is difficult to be certain of their success without being able to see and scrutinise the results. It has become somewhat of a cliche for organisations to say that they are 'harnessing AI/LLM technology', however this can mean a range of things, and absent published results, we are unsure that they have achieved reliable and user-ready results.
There are many steps which Politicker has taken to force AI to deliver reliable data. A tireless commitment to prompt engineering has been essential, in terms of asking the right questions and in the right way. Other steps we reserve as trade secrets, though we challenge other organisations to see if they can achieve the same. Anyone who's used an LLM service such as ChatGPT, Gemini, Co-Pilot, Claude etc., knows that their capacity for language and interpretation is incredible, however that for larger tasks which require precision and consistency, AI models invariably produce 'hallucinations'. Sometimes a task which ought to have been made much easier by AI, actually takes longer than simply visiting a website and collecting the information yourself.
Human testing against LLM output is essential. Only then can you ensure that the data is as reliable as a human's, but produced instantly and at-scale.
It is for this reason that we at Politicker are highly sceptical of AI results, and approached our task with the view to never rely on them without independent verification. LLMs have been instructed to express certainty and behave in a manner which pleases users. This can often result in false information and a failure to acknowledge its limitations. Politicker's methodology addresses AI's shortcomings, and overcomes them. Using well-established methods of reliability testing used by political scientists, behind-the-scenes human researchers conduct their own analyses of random samples of parliamentary data. The results are then compared against LLM output and statistical formulas determine whether or not the results are similar enough, accounting for the possibility of false positives. Politicker conducts these tests regularly and commissions independent academics to do the same, ensuring that the data is verified both internally and externally.
With this watertight methodology, Politicker has the capability to ask all manner of questions of the raw data. Shown publicly on the home page is a tracker of issue salience, meaning we can see which topics MPs focus on and how this changes over time. The exchanges between Ministers and MPs can also be analysed in greater depth, for instance at weekly Prime Minister's Questions, Politicker can show which subjects the Prime Minister is asked about, how questions are asked (e.g. plant questions, attacking questions etc.) how he answers (e.g. defensively, evading accountability, blaming others, not answering the substance of the question etc.) and much more.
On single issues of interest to citizens and organisations, such as EU relations or Net Zero, Politicker can track parliamentary discourse over time, revealing the positions of parties and individual politicians, how these issues intersect with others, and what the key arguments are on both sides. The tool can also gauge the 'temperature' on single issues, showing where there is polarisation and where fault lines lie.
So far, Politicker is just scratching the surface. We invite you watch our development, use our publicly available data, and contact us directly if there's something you or your organisation want to know about in greater detail.
