Chat has a clock running in the reader’s head. Silence that lasts more than twenty seconds is read as being ignored, and a wall of text is read as a form letter. Somewhere in the middle is a small window of time: short sentences that come frequently but each one is something new and true while the agent works.
Most chat libraries remove the ability to message agents in the middle of the script. Agents are left to “improvise” the ninety seconds between the scripted greeting and resolution. Satisfaction scores are actually determined during that time.
Chat is a turn-based system where both parties are able to message simultaneously. A system that assumes each party takes turns messaging breaks the moment a customer posts three messages while an agent is still drafting theirs. Messages must be kept short, so that no line depends on a preceding line delivered first, and the system must provide a means of acknowledging a message that arrived while composing the response, without starting over.
With machine-generated chat lines, the chat is paragraph-shaped. They are polite and complete, and four times longer than anything written by a human, so the customer sees a big block of text and instantly understands that there is no human behind the chat.
The writer looks at your chat transcripts, with timestamps, to see a few hundred examples of what agents send to fill in the gaps and what time segments are empty. Messages are drafted as a bubble, then pasted into a chat window to see what the length will be and how it looks on a phone.
You get sequences, not isolated lines: an opener, the questions that qualify the issue, holding language keyed to the duration of the lookup, and three ways to close. Every line is marked for its position in the flow, and anything containing a variable is written so the variable is inserted in the middle of a sentence.
One rate, whatever the format: $10 per 100 words. You are paying for the writer’s time and judgement, so the price scales with the words rather than with a package tier.
| Typical chat-support script | Words | Writing | Fee (1%) | You pay |
|---|---|---|---|---|
| Live-chat opener set | 400 | $40 | $0.40 | $40.40 |
| Chat flow for one journey | 1,000 | $100 | $1 | $101 |
| Full chat playbook | 2,200 | $220 | $2.20 | $222.20 |
The writer receives 100% of the writing price. Our 1% fee is added on top of it, and 0.5% is donated to tree planting. Full pricing breakdown.
Anything from 100 words up. Because you pay by the word you are never buying padding to fill a package tier.
Yes. Tell us the volume and cadence and we keep the same writer on the account so the voice stays consistent. The rate does not change with volume.
Once the copyright is transferred, you can do what you want with them, but they are written for people. Humans improvise around a line; a bot needs intent coverage and fallback logic, which we are not providing. If a bot is the destination, let us know and we will structure it differently.
Cutting, with some amount of completeness also. The robotic read is due to a compulsion to use full sentences when a fragment would do, and the use of formal connectives no one types. We write more closely to the way your best agents actually chat. This means we’ll send you their transcripts that include our edits that show what we actually kept.
Completely. Copyright transfers to you on delivery, with no attribution requirement and no licence back to us. Nothing written for you is resold, repurposed or republished.
Reviews from completed, paid orders in this category.
Thirty articles with the same structure and with the same voice are difficult to develop for a knowledge base. Screenshot placeholders were described with prose instead of with tags. Unfortunately, our designer had to try to find them with some difficulty. This is a small thing, but I plan to put it in the brief next time.
Twenty-two articles on our top ticket drivers. The writer requested redacted transcripts, and then described the problems in the way customers used the language – thermostat will not hold temperature, instead of thermostat calibration. Help center search finally started working.
Eleven articles covering our top ticket drivers. They requested the original ticket text first, not the internal descriptions, and the difference is significant. The articles use the terms customers input into the search box. There’s been a downward shift of about a fifth in the volume of contacts on those topics.
We do not ask you to take the no-AI promise on trust. Each draft is run through these platforms and the reports are attached to your delivery email.
Originality.aiAI detection + plagiarism
GPTZeroAI detection
TurnitinAI detection + similarity
CopyleaksAI detection + plagiarism
Winston AIAI detection
ZeroGPTAI detection
SaplingAI detection
Content at ScaleAI detection
Send the brief, get a human draft back within 3 days. No account, no subscription, no AI.
$10 per 100 words · all of it to the writer · 0.5% to trees