The worst support failure of my career was not rudeness, and it was not a slow reply. It was a dropped promise, and it was entirely avoidable. A customer had been told by a previous agent, in writing, that we would refund their last month. The ticket got reassigned to me over a weekend. I opened it, skimmed the three most recent messages because the queue was on fire, missed the refund commitment buried fourteen messages up, and confidently told the customer we did not offer refunds for that situation.
They were, with total justification, livid. Not because of the policy — because we had contradicted ourselves to their face. Two agents, two opposite answers, same company. There is no faster way to destroy trust than that, and I had done it by skimming. The refund itself was a small amount of money. The damage was that the customer now believed nobody at our company actually read their file, and honestly, in that moment, they were right.
Long threads are minefields, and the mines are in the middle
A ticket with twenty messages, three agents, and a week of back-and-forth is a genuine hazard. The single most important detail — a promise, a deadline, a workaround already attempted — is almost never in the most recent message. It is somewhere in the soft middle, and when you pick up the ticket cold under time pressure, the middle is exactly what you skim past. Every reassignment is another chance to drop something that someone, somewhere up the thread, already committed to.
The structural problem is that the urgency of a ticket lives at the bottom (the newest message) while the obligations of a ticket are scattered throughout the whole thing. Your eyes go to the urgent part. The binding part hides where you are least likely to look. It is almost designed to fail.
After the refund disaster I did not trust myself to read carefully under pressure anymore, because the evidence said I could not. And that turned out to be the healthiest professional conclusion I have ever reached: instead of vowing to 'just be more careful next time' — a promise I had made and broken before — I admitted the failure was systematic and went looking for a system.
Reading the thread for me, before I reply
Now, before I touch any long or reassigned ticket, I run it through the Summarize a Long Ticket Thread and Surface Every Promise prompt on Gemini 2.5 Pro. I picked Gemini specifically because this is long-context synthesis — twenty messages in, a clean structured table out — and it formats the brief as a table so I can scan it in fifteen seconds: one-line status, what has been tried, the current blocker, the customer's sentiment, and the single best next action.
The row that earns its entire keep is the one in shouting capitals: WATCH OUT FOR. It pulls every promise, refund, deadline, and commitment made anywhere in the thread up to the top where I cannot miss it. The exact failure that haunted me is the thing this prompt is built to prevent.
- Every earlier commitment gets surfaced under WATCH OUT FOR, front and center
- I read the customer's sentiment before replying, so I match my tone instead of fumbling it
- The recommended next action stops me re-litigating things already tried
- Picking up a reassigned ticket now takes two minutes instead of ten of nervous re-reading
Consistency is the thing customers actually pay for
I think a lot now about how being told two different things by two agents destroys trust faster than almost any single rude reply could. A customer can forgive a slow response. They struggle to forgive being made to feel like nobody is reading their file. This brief keeps our story straight across the messiest, most-reassigned threads, which is exactly where our story used to fall apart.
When the brief tells me the customer is escalating, I hand the actual reply to my De-escalate a Furious Customer Without Over-Promising prompt — the summary tells me what was promised, the reply tool helps me hold the line without caving. The two together are the closest thing I have to a safety net on a tangled, angry, reassigned ticket: one reads the history correctly, the other responds to it like a calm human.
I will add the caveat I give every colleague I show this to: the brief is only as honest as the thread. If a commitment was made on a phone call that never got logged, no summarizer on earth can surface it, because it is not in the text. The prompt fixed the failure mode where the promise was written down and I skimmed past it — which was my actual failure — but it is not a substitute for logging commitments in the ticket in the first place. Garbage in, garbage out; complete thread in, trustworthy brief out.
Grab the Summarize a Long Ticket Thread and Surface Every Promise prompt on Prompt Dock and run it on the most tangled open ticket in your queue right now. I have not dropped a promise on handoff since, and I sleep noticeably better for it.