The airline chat has no sound
The airline cancelled my flight and refunded the ticket, not the dearer flight home. Their support chat has no sound, so I missed the agent and got disconnected. I asked my agent for a sound alert first, then for the claim. Every dollar came back.
- Where
- In the web chat
- Size
- A longer one
- Short link
- piy.show/s129
A refund, a silent chat, and a disconnect
The airline cancelled our flight home from Tokyo. We flew with another airline instead. The refund came back, but the new tickets cost about two thousand dollars more, and that part did not.
So I opened their complaints chat. Twenty minutes in the queue. I switched to another window to work, and the chat has no sound. The agent arrived, waited, and disconnected. Back to the queue.
That is the moment this Short is about. The claim was the real job, but the first problem was the tool. A chat you cannot hear is a chat you cannot win. Fix the tool first, then argue.
Why fix the chat before the claim? Prompt it yourself
I am waiting in a support chat that has no sound. What is the smallest thing I can build in my browser so I never miss the agent?
Give the agent the page, not a description
My agent cannot see my screen. It can read the page. So I opened the page source of the chat window, copied the whole thing, and pasted it in with my request: a script I can paste into the browser console that plays a sound when a new message arrives.
From the source it found the chat widget by name, and knew where the messages appear. Without the source it would have guessed, and a guess in a browser script is a script that watches the wrong box.
This is the trick for any website you need to automate for yourself. Right-click, view source, copy all, paste. Then say what you want in plain words. The source is your screen, in text.
How do I copy a page's source for the agent? Prompt it yourself
Here is the page source of the chat window. Write a script I can paste into the browser console that plays a sound when a new message arrives, and ignores my own messages.
Ask for the test buttons with the feature
I did not ask for a sound. I asked for a sound, a Test sound button, and a Self test button with a report. The script came back as a small dark panel on the page. Test sound played three tones. I switched to another window, and it repeated every few seconds until I came back. Ping, ping, ping.
The self test is the part that matters. It checks the sound, the clipboard, whether the chat box was found, and whether a fake message sets off the alarm. Each check prints PASS or FAIL, with a Copy report button. I pasted the report back, and the agent fixed the script without seeing my browser. Two rounds, done.
Ask for the self test every time you ask for a tool. The agent cannot watch you use it. The report is its eyes.
What should a self test report? Prompt it yourself
Add a Self test button to the script. It must check each part, print PASS or FAIL per check, and have a Copy report button so I can paste the report back to you.
The bug the transcript found
The panel also has a Copy chat button. It copies the chat text and every new message with its arrival time. I built it to paste the chat back to my agent for help with the case. It did something else first: it exposed a bug.
The airline's chat redraws the whole history on every update. To the script every redraw looked like a pile of new messages, and the alarm went off for old lines. The transcript showed it plainly. The fix was to compare the chat text with the text at the last check, and count only the new lines.
Keep a button that copies what the tool saw. A copy of what the tool saw is the bug report, and it is better than any description you can type.
Why did old messages set off the alarm? Prompt it yourself
Here is the chat text the tool captured, with times. The alarm fired on an old message. Find out why, fix it, and add a test that replays this exact case.
I can help you build yours
I build things like this with an AI agent, on my own computer. In a private session I help you do the same with your idea. Or bring it to the show, free.
Then the claim itself
With the alarm working, I gave my agent the evidence. The cancellation email, the airline's invoice, the refund email, the new ticket and the receipt for the ride to the airport. Files, not a summary.
It read the cancellation reason out of the email and matched it to the airline's own policy, which lists that reason as within their control. It worked out the loss. It found a gap in the numbers, asked me about it, I explained it, and we left that part out. A claim with a wrong number invites questions.
Then it did the research on consumer rules and wrote the claim: what happened, in numbered steps, the amount, the documents I can send, and a request for a case number. One message, with everything an agent would ask for already in it.
What makes a claim message hard to refuse? Prompt it yourself
Here are my cancellation email, the invoice, the refund email and the new ticket. Work out exactly what I lost. Flag any number that does not add up before you use it. Then write the claim with the documents listed.
Every dollar back, in one chat
I pasted the claim into the chat late at night. Nineteen minutes in the queue, and this time the alarm called me back. The agent asked four things. Each one was already answered in the claim, or I answered it in one line that my agent had ready.
The agent even found a small part of the refund that had been missed the first time. About an hour after I pasted the claim, the full amount was agreed in the same chat. Approved the next morning. No escalation, no "we will email you".
When the message answers the questions before they are asked, there is nothing left to escalate. That is what the preparation buys.
How do I answer a support agent fast? Prompt it yourself
The support agent asked me this. Write a one-line answer from the facts in my claim, and tell me if anything in my claim needs to change.
Start with your own chat
Next time you wait in a support chat, do these in order. Copy the page source and ask for the sound alert with its test buttons. Then give the evidence files and ask for the claim. Then let the alarm wait for the agent, not you.
What do I do next? Prompt it yourself
I have a complaint to make in a live chat. Here is the page source and here are my documents. First build me a sound alert with a self test, then write the claim.
Prompty