Doubling a team is a request for 100 percent more input against a document that has grown by a factor of 19 to 27 [3][5], which means the association is implicitly promising to give each incoming page somewhere between 7.4 and 10.5 percent of the attention the old single sheet used to get (two staff-units over 27 pages, then two over 19) [9]. That would work if the added pages were skimmable padding, but they are not, because of the specific reason the BBC identified on 29 August: the tools sometimes cite the wrong provision or state something that is simply untrue, which is why civil servants end up verifying each item and buying legal advice [1][4].
That is the whole mechanism, and it is worth being precise about it. A long but accurate filing can be answered by pattern, because the twentieth citation of a correct statute costs the reader nothing new. A long filing with an unknown error rate cannot be triaged at all, because you do not know which line is the wrong one until you have read every line. Length is cheap to produce and cheap to receive; unverified authority is the expensive part. The Information Commissioner's Office and the national association of data protection officers describe the same effect in freedom of information work as hidden workload that delays processing [7], and the association of school leaders reports teaching staff absorbing multiple long parent complaints [14].
Note also what the doubling estimate actually covers. It is framed as the staffing needed to handle the surge in complaints [5], and the account does not separate more complaints from longer ones, which matters because those two curves have very different fixes. Volume growth is real enough on the sending side, where examples of people using AI to contest fines and appeal benefit decisions are circulating [11].
The responses now visible price differently. Length-capped complaint forms have been proposed [6], which is the cheapest to implement and the most exposed, since a cap is exactly the sort of administrative decision a chatbot drafts a good appeal against [11], and since the same report credits AI with restoring the process to neurodivergent people and others who had given up filing anything once correspondence moved to email [12]. Summarising the AI complaint with AI and issuing a one-page reply is already happening [8], which converts a labour cost into a liability, because the summary that drops a cited authority is the summary you have to defend. Going back to meeting residents and parents in person [8] is the honest answer, and it requires staff time no current budget allocates.
This is probably wrong, but the summariser wins on cash and the cap dies on challenge, and the constraint that decides it is documentary rather than technological. What would prove it wrong is a published error rate: if councils find that AI-cited authorities are accurate often enough that sampling survives review, the second headcount never gets hired and the caseload just gets slower [13]. If, instead, an unanswered cited authority starts being treated as a procedural failure, doubling a legal team turns out to be the discount option.
