How we make Infercline
How a story is made
Every morning our system reads the newsrooms and official blogs we follow in each of our five markets. AI tools pick the stories that matter most to founders and operators, and draft each one in our house style from the source article. Each story links to its sources.
Drafts are ready at 05:00 South Africa time. Our editor reviews them, and anything not rejected is published at 07:00.
Our rules for every story: only facts stated in the sources, no invented numbers or quotes, facts attributed to the publisher that reported them, and a summary in our own words rather than a copy.
Corrections
If we get something wrong, tell us at editor@infercline.com. We fix it and say so.
The Invoice Index
What it measures: what it costs AI to check 100 supplier invoices correctly. The job: read an invoice and its purchase order, find the mismatch, and write the supplier a short email with the corrected figure.
How: the same task, the same documents and the same instructions go to each AI lab's flagship model, ten runs each. A computer checks every answer against an answer key we write in advance. The price is total spend divided by the number of correct answers, so mistakes are priced in.
Sources of truth: token counts reported by each lab's API, multiplied by each lab's public price list. Exchange rates are pinned to the Monday of each week. The tasks, answer keys and code will be published so anyone can rerun them.
Status: the first reading is coming soon. We publish the method before the first number.