Making sense of large document sets

Procurement records, draft budgets, court files, hundreds of pages of PDF. The information you need is public, and that is the problem: public in a form nobody has time to read.

Summarising is the thing these tools genuinely do well, and the thing they fail at most convincingly. A summary reads smoothly whether or not it is accurate. The tool will not say “page 14 was a scan I could not read”. It summarises the pages it could read and presents that as the whole.

Watch out for this

Never quote a summary in something you publish without opening the original page and reading the sentence yourself. Ask for page numbers and check two at random every time; if the page numbers are wrong, nothing else in that summary is trustworthy either. Look at what you are uploading, too: a document set from a court or a ministry usually has names, ID numbers and addresses buried in the middle of it.