"Do it again, but good"

Miranda Priestly in The Devil Wears Prada is one of the most quoted bosses in cinema, and mostly for feedback that tells you a lot about how she feels about your efforts and practically nothing about what to change to make it better. Her feedback on a new layout is "not wonderful yet" – with no explanation of what wonderful would have looked like. Ask her to clarify and the answer is "details of your incompetence do not interest me". Hilarious, but not very helpful.

I'm guessing (hoping?) few of us have ever spoken like that to a staff member, but chances are you have to a chatbot. A chatbot comes back with an answer that considerably misses the mark and you write back saying "that was awful, do it again but good". There's no risk attached to it - nobody's watching, and the chatbot won't hold it against you - so it's an easy way to blow off some steam and unleash your frustration. But there's no feedback in it, any more than there was in Miranda's line. It doesn't say what was wrong with the work, what "good" would look like, or which part to fix. It just asks for another attempt and leaves the model to guess.

Try it on a person and things may end up quite differently. Tell someone on your team "that wasn't good enough, do it again properly" and you're heading for a difficult conversation with HR, and rightly so - that kind of talk is more likely to get you a grievance than a better piece of work - even with no repercussions at all, would it produce a better second draft?

Useful feedback does three things: it says what happened, what effect it had or may have, and what should happen instead. That's not a softer version of criticism, it's just the information someone needs to do it differently. Vague feedback leaves them guessing, and every wrong guess costs time on both sides. Teamshaper's Managing Performance masterclass is built around a four-stage way of giving feedback that does this properly between people, so the correction happens once instead of several times.

Lack of specificity in feedback costs more than time - it damages confidence and it's basically the opposite of nurturing or mentoring. It doesn't mean people should be handed things on a silver platter, and there are few careers worth having that weren't built on some hardship or overcoming obstacles through determination and grit. It does mean that constructive feedback produces better results quicker and achieves better long-term results.

The same logic applies to a chatbot, even though it's tempting to skip it because there's no one watching. The model has the same problem as an under-briefed staff member: no more information about what to fix than it had the first time, so it changes something - not necessarily the right thing – and you're back to square one.

Spending the time explaining what's wrong - the tone's off, the second paragraph repeats the first, it's ignored the word count, it used the wrong example, it misunderstood your argument - takes a bit longer to type or say but it gets you the right results a lot sooner, for the same reason it works with a person: fewer guesses, less rework.

In a way it’s even more essential to provide specific feedback to a chatbot than a person. The person will have access to wider context than the chatbot about your preferences which they will be able to draw on to complete the missing information from your feedback. They may pick up on facial expressions or tone of voice if the feedback was provided verbally, which will be completely lost on the chatbot. If you’re asking the person for a rework essentially because you think they didn’t put enough effort into it, that reasoning won’t apply to a chatbot because a chatbot doesn’t slack off, it works off the information you provide it.

Which brings us neatly to memory. Tell a staff member once that you prefer a report to open with the recommendation rather than the background, and you don't have to say it again - they remember (perhaps with the occasional reminder), and the next piece of work starts from there. A chatbot can do the same, but only if what you tell it gets stored somewhere it can use again, rather than said once and lost when the chat closes: your style, your preferences, the conclusions you've already reached. Without that, every new chat starts from nothing, and the same instructions get typed in every time. Teamshaper's Writing with Copilot programme explores this: how to build memory of preferences within your Copilot account so that it’s enriched by cumulative learning achieved through your work together.

None of this is really about software, it's about what feedback is for. Venting feels good for a moment, then you're re-explaining yourself anyway, whether you're talking to a person or a machine. A clearer sentence the first time gets you there quicker, and with less drama.

That’s all.