Start by assigning each job a project label in your own logs, even if the provider account stays shared. Record its start time, request count, transferred bytes, retries, and usable output. Browser sessions can load images and other assets that a simple text request never fetches, so raw request counts alone are a poor basis for comparison. Set a small traffic limit for each trial and stop automatic retries after a defined number of attempts. You can then compare costs by task type and discuss exceptions using evidence, instead of dividing the total bill equally among people doing very different work.