GUIDE

An AI response generator is only as good as the price list you give it

Any AI response generator can write a polite reply. The one that books the job answers the price and the date, and it can only do that if you give it a one-page sheet of your real facts. On your own API key, each draft costs about a cent or less.

WHAT THIS ADDSThe generator itself is a commodity that costs roughly a cent per reply on your own key. All of the value sits in a one-page facts sheet and a written rule for which messages AI never answers first. A warm reply that dodges the price question loses more jobs than a blunt one that answers it.

What an AI response generator actually does for a small business

An AI response generator drafts replies to the messages customers send you. That means quote requests, questions like are you free on the 14th, do you take my insurance, or how long will the repair take. The writing part is now cheap and decent everywhere. What decides whether the reply books the job is whether the AI knows your prices, your calendar, and your rules, and most free generators know none of them.

Search the phrase and you get dozens of web tools built around one box. You paste the customer's message, pick a tone (friendly, professional, apologetic), and get a reply. They all work, in the sense that they produce fluent, polite text. That is the trap.

Fluent and polite is not what the customer asked for. They asked for a price and a date. A reply that sounds lovely and answers neither question reads to them as a business that is slow, expensive, or hiding something.

So the useful version of the question is not which generator writes best. It is how you get any of them to write the reply you would have written yourself if you had ten quiet minutes. That depends on three things: what you feed it, which messages you let it touch, and where it runs. The rest of this guide takes them in that order.

If you came here for review replies specifically, that is a narrower job with its own risks. It is covered in who should press post on AI review replies. This guide is about the inbound messages that turn into paid work.

Why the free generators write replies that lose jobs

Here is a made-up but typical example. Marisol is a wedding photographer in Tucson. She has two packages: $3,200 for six hours with one photographer, and $4,800 for eight hours with a second shooter and an engagement session. An inquiry arrives:

Hi! Are you available June 12, 2027? What do your packages cost? We're getting married at a vineyard in Sonoita, about 150 guests.

Paste that into a generic response generator and you get something like this:

Thank you so much for reaching out, and congratulations on your engagement! I'd love to hear more about your special day. My packages are tailored to each couple's unique needs. Could we set up a quick call to discuss your vision?

Nothing in that is wrong, and that is exactly why it is bad. It answers neither question. It adds a step, a phone call, before the couple learns whether they can afford her. The couple almost certainly emailed four or five photographers that evening. The ones who replied with a date and a number made the shortlist. The tailored-to-your-needs reply quietly did not.

Now the same message, drafted by an AI that was handed Marisol's facts:

Congratulations! June 12, 2027 is open as of today. I have two packages: $3,200 (6 hours, one photographer) and $4,800 (8 hours, second shooter, engagement session included). For 150 guests at a vineyard, most couples choose the 8-hour package because the ceremony and golden-hour portraits tend to be far apart. Sonoita is inside my travel area, so there is no extra fee. A $1,000 retainer holds the date. Want me to send the contract?

Same AI model, same tone setting. The only difference is the information it was given.

The worse failure: invented prices

There is a second way generic generators go wrong, and it is more expensive. Ask one to include pricing when it has no pricing, and some models will produce a plausible number anyway: starting at $2,500, a standard deposit of 25 percent. It reads like a fact. If you skim and send it, you are now negotiating against a number you never set.

Body shops, contractors, and clinics get hit hardest. A made-up repair estimate or a made-up insurance answer is not a typo. It is a promise. The fix is not a smarter model. It is never letting the AI answer a money or medical question without your facts in front of it, and telling it in writing to say I'll confirm when it does not know.

Generic generatorGiven a facts sheetAnswers the dateNoYes, June 12, 2027 is openAnswers the priceNo, says packages are tailYes, $3,200 or $4,800 withNext stepSchedule a call$1,000 retainer, send the RiskCustomer moves on to someoWrong only if the facts sh
The same wedding inquiry, answered by a generic generator and by one given the photographer's facts sheet

The one-page facts sheet that makes the replies good

The single biggest improvement you can make is a plain document, one or two pages, that you paste into or attach to whatever AI you use. It is the brief you would give a sharp new front-desk hire on their first morning. The longer argument for why this matters is in Claude is good for business writing once you hand it your facts. Here is what actually goes on the page.

  • Prices and what each includes. Real numbers, not ranges, wherever you can. If a price genuinely depends on something, say what it depends on: a bumper repair is quoted from photos, usually between X and Y.
  • What you do not do. No commercial jobs. No cosmetic procedures. No weekend installs. AI is eager to say yes, and this is the list that stops it.
  • Service area and travel fees. Towns, zip codes, or a radius in miles.
  • How booking and payment work. Deposit amount, when the balance is due, how far ahead you book up, your cancellation rule.
  • The one question you always need answered. Every trade has one. A body shop needs year, make, model and photos. A clinic needs the insurance carrier. A contractor needs the address and pictures of the space. Tell the AI to ask for it in every reply that does not already include it.
  • Three replies you have sent that worked. Paste them in full. This teaches tone better than any setting labeled friendly.
  • Hard rules. Never promise a date that is not confirmed open. Never quote a repair without photos. Never give medical advice. If unsure, say you will confirm and stop.

Two things people get wrong here. First, they write the sheet in marketing language, along the lines of passionate about quality, family-owned since 1998. The AI then writes replies in marketing language. Write it the way you would explain it to a new hire.

Second, they never update it. A stale facts sheet is worse than none, because the AI will quote last year's prices with total confidence. Put a date at the top. Change it the same day your prices change, not when you remember.

Sort your messages into three piles before you automate anything

The pitch for most AI responders is that they answer everything, instantly, around the clock. For a small business that is usually the wrong goal. The cost of a mistake is lopsided. A wrong answer about your hours costs you a confused phone call. A wrong quote costs you margin or trust, and a wrong answer to an angry customer can cost you the customer and a review.

So before you switch anything on, sort what lands in your inbox into three piles.

Pile one: AI drafts, you send with a glance

Hours, location, parking, what to bring, do you do this kind of job, how long does it usually take. Low stakes, and the answers do not change week to week. Even here, read every draft for the first month. You will spot the patterns where your facts sheet is thin.

Pile two: AI drafts, you check every line

Anything with a number or a date: quotes, availability, deposits, rescheduling. This is where AI saves the most time, because these replies take you longest to write. It is also where you never skip the read. Check the price against your sheet and the date against your calendar.

Pile three: you write it, AI stays out

Complaints, refund requests, anything with a legal edge, anything about symptoms or medication, and anyone you have already disagreed with. These replies need judgment about a specific relationship. A model that has never met the customer will smooth things over in a way that reads as dismissive. Use AI afterward to tidy your grammar if you like, but not to decide what to say.

The honest win is not zero-touch replies. It is going from ten minutes per reply to about one, while still answering within the hour instead of that evening. For most owners that is the whole value, and it is plenty.

1Write a one-pagefacts sheetPrices, exclusions,area, deposit, theone question youalways ask2Add three repliesthat workedPasted in full, toset the tone3Sort messages intothree pilesGlance, check everyline, or neverAI-first4Test on your lastten inquiriesCompare drafts withwhat you actuallysent5Fix the sheet, notthe draftsEvery repeated editis a missing fact
A safe order for setting up AI-drafted customer replies

What it costs when you use your own API key

The draft itself is close to free. The real cost is your review time and whatever markup a tool adds on top of the AI.

Here is the arithmetic, so you can redo it with your own numbers. AI providers bill by the token, which is roughly three-quarters of a word. A one-page facts sheet of about 1,500 words is around 2,000 tokens. The customer's message plus a little thread history adds maybe 200 more. A solid reply comes back at around 300 tokens.

When we checked Anthropic's published pricing, Claude Haiku 4.5 was listed at $1 per million input tokens and $5 per million output tokens. Claude Sonnet 4.5 was $3 and $15. Working it through:

  • Haiku: 2,200 input tokens is about $0.0022, plus 300 output tokens at about $0.0015. Call it $0.004 per reply.
  • Sonnet: 2,200 input is about $0.0066, plus 300 output at about $0.0045. Call it $0.011 per reply, roughly a cent.

At 300 replies a month, a busy shop, that is about $1.10 on Haiku or $3.30 on Sonnet. OpenAI's smaller models land in the same range, and you can check their current rates on OpenAI's pricing page. Model names and prices change, usually downward, so look up Anthropic's current rates and swap your numbers into the same sum.

Two practical notes. For routine replies the small, cheap model is fine. Save the larger one for long, tricky quotes where the reasoning matters. And if a tool bundles AI into a subscription or sells reply credits, divide its monthly fee by the number of replies you actually send. That tells you what you are paying per draft compared with a cent on your own key.

Then do the sum that matters more. If you read each draft for 30 seconds, 300 replies is two and a half hours a month. That time is the real expense, and it is the reason the facts sheet matters. Every line you have to fix by hand is a line the sheet should have answered.

Claude Haiku 4.5~$1.10/moClaude Sonnet 4.5~$3.30/mo
Approximate AI cost for 300 replies a month on your own key, using about 2,200 input and 300 output tokens per reply at Anthropic's listed rates when checked

Where to run it: a chat project, your inbox, or your customer records

Pick the setup based on where your messages actually arrive and how many you get. There is no single best generator.

Under about ten inquiries a week: a chat project

Both ChatGPT and Claude let you create a project and attach files to it. Put your facts sheet there once, then paste each customer message in and copy the draft back out. It costs nothing beyond the plan you may already pay for. The downside is the copying, and the fact that it cannot see your calendar, so you check dates yourself. For a solo photographer or a two-chair clinic, this is often all you need. Do not let anyone sell you more.

Mostly email, low volume: the AI built into Gmail

Gmail's built-in writing help is convenient because it sits right in the reply box. It does not know your prices unless you tell it each time, so you end up pasting your facts in again and again. Fine for occasional use, tiresome as a system. If the AI features are not showing for you, see why Gmail's AI Inbox may not be showing up.

Steady volume: inside the system that holds the customer record

Past a handful of inquiries a day, the copying becomes the bottleneck. The draft gets better when the AI can see who this person is, what you quoted them last time, and which slots are open. That is the case for running it where your contacts, calendar and quotes already live. Fynl does this with its Gmail-connected inbox and your own Claude or ChatGPT key, so the drafts use your facts and you pay the provider's rate with no markup.

Be clear about the limit. Fynl does not handle SMS or phone calls yet. If most of your leads arrive by text message or Instagram DM, a tool built around those channels will serve you better, and you should choose that instead. The best response generator is the one sitting where your customers already write to you.

DO THISWrite a one-page facts sheet with your real prices, exclusions, and booking rules, then test any generator on your last ten inquiries before trusting it. Choose where to run it based on where your messages actually arrive, and keep complaints and refunds in your own hands.

Questions people also ask

Can I let an AI response generator send replies without checking them?

Only for the lowest-stakes questions, such as hours, location, or what to bring, and only after a month of reading every draft. Anything with a price, a date, a complaint, or a medical or legal angle should be read by a person before it goes out.

Is a free online AI response generator good enough?

For tone, yes. For answering what customers actually ask, usually not, because it does not know your prices or availability. It will either dodge the question or invent an answer. Either use one that lets you attach your facts, or paste them in every time.

Do I need to know how to code to use my own API key?

No. You create the key on the provider's website, add a payment method, and paste the key into whichever tool supports it. The provider then bills you directly for what you use, typically well under a cent to about a cent per reply.

Will customers be able to tell the reply was written by AI?

They notice vague, gushing replies that do not answer the question, whoever wrote them. A short reply with your real price, a real date, and a clear next step reads like a competent business. Feeding the AI three of your own past replies makes it sound like you.

Sources

One portal instead of five subscriptions

CRM, calendar, quotes, invoices, website and email — with the AI running on your own Claude or ChatGPT API key, so there is no markup on it.

See pricingMore posts