How to Use AI to Sell: Lead Replies, Quotes and Follow-Ups

A small business can use AI to draft three things: a reply to each new enquiry within minutes, quote wording that spells out scope and price, and a short follow-up sequence for quotes that go quiet. ChatGPT or Claude writes the first draft from your notes; a person checks every price, date and promise before it is sent.

The gain is speed and consistency, not magic. Many small firms lose enquiries to slow replies and to quotes that nobody chases, and both are drafting problems. The risk sits in the details AI will happily invent, such as a start date you never offered or a discount you never agreed, so the workflow below puts a human check at the one point where those mistakes become promises.

Follow me on Instagram@sagnikteaches

Three drafting jobs, and which one pays first

Selling by email, DM or text breaks into stages, and AI helps with different ones in different ways. Before you write a single prompt, decide which stage leaks the most.

Connect on LinkedInSagnik Bhattacharya

• The first reply. The enquiry arrives, and the person is still comparing two or three businesses. Speed and a clear next step matter most here.
• The quote. The wording decides whether the buyer understands what they are paying for. Vague quotes get ignored or haggled.
• The follow-up. Most quotes get no answer on day one. A polite sequence of three messages recovers a share of them that would otherwise be lost.

For a business that already replies quickly, the follow-up is usually where the money is. For one that answers enquiries the next afternoon, start with the first reply. Check your own numbers: take the last 20 enquiries and mark how many got a reply within an hour, how many received a written quote and how many of those quotes were chased at least once. The smallest of those three numbers shows you where to begin.

Subscribe on YouTube@sagnikteaches

Step 1: Write down what the AI is allowed to know

AI drafts good sales messages only when it has facts. It cannot know your prices, your availability or what you refuse to do, so give it a short "fact sheet" once and reuse it. This lives in a ChatGPT Project or a Claude Project (a saved workspace with standing instructions and files), so you do not paste it every time. If you have two or more people, a business plan is worth the cost; both ChatGPT Business and Claude Team need at least two seats, and list prices are on the vendors' own pages. A one-person business can use an individual paid plan instead.

A fact sheet for a photography studio might look like this (illustrative, with invented figures):

BUSINESS FACTS (use only these; if something is missing, ask me)
- We are a four-person photography studio doing family portraits, headshots and small-business product shoots.
- Portrait sessions: 60 minutes, up to 2 outfit changes, 15 edited images, $320. Extra edited images $18 each.
- Headshots: 30 minutes per person, 3 edited images each, $140 per person, minimum 4 people for an on-site visit.
- Product shoots: from $450 for up to 10 items, quoted by item count.
- Booking: 25% deposit holds a date. Balance due on delivery of the gallery.
- Delivery: galleries within 10 working days.
- We never promise a delivery date shorter than 10 working days.
- We do not offer printed albums.
- Tone: warm, brief, plain English, no exclamation marks, no emojis.

Two lines in there do most of the protective work: "use only these" and "never promise a delivery date shorter than 10 working days". A fact sheet that only lists what you do leaves gaps, and the AI fills gaps by guessing; listing what you will not do gives it a boundary to stay inside.

Step 2: Draft the first reply from the enquiry itself

The prompt is simple: paste the enquiry, ask for a reply, and tell the AI what the reply must achieve. The reply should answer the question asked, confirm one detail, and offer one next step. Not three.

Using the business facts in this project, draft a reply to the enquiry below.
Rules:
- Answer only what was asked, using only the facts I gave you.
- If a price depends on something the customer has not said (number of people, number of items), ask for that one thing instead of guessing.
- End with one clear next step and a reply-by time I can keep ("I'll hold Saturday 10am until Thursday evening").
- Under 120 words. Plain text, no bullet points.
- If anything in the enquiry is unclear or outside our services, say so and flag it to me before the draft, in capitals.

ENQUIRY:
"Hi, I run a small accountancy practice with 6 staff and we need new profile photos for the website. How much and how soon? Thanks, [first name]"

A typical draft back (illustrative):

Hi [first name],

Thanks for getting in touch. Headshots for six people would be $140 each, so $840 in total, with three edited images per person. We'd come to your office and work through the team at about 30 minutes each, so around three hours on the day.

Galleries are delivered within 10 working days of the shoot. If you'd like to go ahead, a 25% deposit ($210) holds the date. Could you tell me which two or three weekdays would suit you best? I'll hold whichever you choose until Thursday evening.

Kind regards,
[your name]

What a person should fix: the arithmetic is right ($140 × 6 = $840; 25% of $840 = $210), but "around three hours" is the AI's own estimate (6 × 30 minutes) and does not allow for set-up. Either add the set-up time to the fact sheet or delete the sentence. Also note that the draft says "on the day" and "come to your office", which is a reasonable inference from "minimum 4 people for an on-site visit" but not something the customer asked for. A reader checking drafts learns to look for exactly this kind of confident extra.

Step 3: Write quotes people can say yes to

A quote is a short document that answers four questions: what exactly do I get, what does it cost, what is not included, and what do I do next. AI is good at structuring this, and poor at knowing what is included, so supply the scope in your own words and let it do the layout and tone.

Here is a messy set of notes from a driving-school owner who is quoting a block of lessons for a nervous adult learner (illustrative):

Notes: [first name] wants to start from scratch, 20 hrs probably, own pick-ups from home, weekdays after 4. Manual car. $48/hr, 10-hr block $450. Theory test fee not included, they book themselves. Test car hire on the day $120 extra. Block expires 6 months after first lesson.

Prompt:

Turn these notes into a one-page quote for a customer. Structure: (1) what is included, as a short list; (2) price, with the arithmetic shown; (3) what is NOT included; (4) how to accept and what happens next; (5) how long the quote stays valid (14 days). Do not add anything that is not in the notes. If a detail is missing, list it under "QUESTIONS FOR ME" at the top instead of guessing.

A good result will do the sums and, importantly, ask. For this one it should come back asking whether two 10-hour blocks ($900) beat 20 single hours ($960), whether pick-ups are included in the hourly rate, and whether the test-car hire is optional. Those questions are the value: they are the gaps a customer would have found for you. Answer them, run the prompt again, and you have a quote where the customer cannot reasonably misread the price.

Wording that cuts haggling and misunderstandings

Three habits that AI can apply to every quote once you ask for them:

• Show the arithmetic. "2 blocks of 10 hours at $450 = $900" beats "$900 for 20 hours" because it tells the buyer how any change will be priced.
• Name the exclusions. A short "Not included" list prevents the awkward invoice-day conversation.
• State an expiry and a next action. "Valid until 14 March; reply YES and I will send the booking link" gives the buyer something to do.

If your quotes involve complicated pricing across many items, build the numbers in a spreadsheet first and paste only the finished figures into the prompt. Language models are better at wording than at long sums, and a quote with one wrong total costs you more than the time saved. For help structuring the sums, the side-by-side quote comparison tutorial shows the same checking habit from the buyer's side.

Step 4: The three-touch follow-up sequence

When a quote goes unanswered, the temptation is either to do nothing or to send the same "just checking in" message three times. A better sequence gives each touch a different job, spaced so it does not feel like pressure.

Here is the prompt, then the three drafts for a dog groomer following up a quote for a regular grooming plan (illustrative):

Write a three-message follow-up sequence for a quote that has not been answered. The messages go by email, 3 days, 7 days and 14 days after the quote.
Message 1: check the quote arrived, answer the most likely question about the quote, ask nothing else.
Message 2: add one genuinely useful thing (information, or a date or slot that is filling), not a repeat of message 1.
Message 3: a short, polite close. Say I will not chase again, leave the door open, no guilt, no pressure.
Each under 90 words, first person, warm and plain, no exclamation marks. Use [first name] and [your name] as placeholders. Use only facts from the quote and business facts supplied.

QUOTE: Six-weekly full groom for a medium-sized dog, $65 a visit, first visit includes a 15-minute coat assessment at no charge. Booking: reply with preferred day. Slots are limited on Saturdays.

A typical result:

Message 1
Hi [first name], I wanted to check my quote came through properly. If you're wondering about the first visit, it includes a 15-minute coat assessment at no extra charge, so we can agree what suits your dog's coat before we start. Just reply with a day that works. [your name]

Message 2
Hi [first name], one thing that might help: Saturdays fill first for six-weekly plans, so if you prefer weekends it's worth choosing a slot soon. Weekdays are usually easier to move if plans change. Happy to hold a day for you while you decide. [your name]

Message 3
Hi [first name], I'll assume the timing isn't right for now and won't keep emailing. If you'd like to start a grooming plan later, just reply here and I'll find you a slot. Wishing you and your dog well. [your name]

What a person should fix: "Saturdays fill first" and "weekdays are usually easier to move" are claims the AI invented from the phrase "slots are limited on Saturdays". If they are true, keep them; if not, delete them. Also check the "happy to hold a day" offer, because it is a commitment. The structure is what you are borrowing from the AI, the facts are yours to confirm.

Step 5: Keep it all in one place, and stop when asked to stop

The drafting is the easy half. The half that decides results is a tracking habit: every enquiry gets a row, every row has a "next action" and a date. The free HubSpot CRM covers two users and 1,000 contacts, per HubSpot's own pricing page, which is plenty for a small studio or school. A spreadsheet works equally well while volume is low. What matters is that the three touches are scheduled when the quote goes out, not remembered later.

A simple layout:

• Name and contact (a column your AI prompts never need to see in full: paste only first name and the enquiry text).
• Enquiry date and source (Instagram, website form, phone).
• Quote value and date sent.
• Touch 1, 2, 3 due dates with a tick when sent.
• Outcome: won, lost, no reply, asked to stop.

"Asked to stop" deserves its own value. If someone replies "please don't contact me again", the sequence ends immediately and nobody on your team should ever message them about the same job. Marketing-email rules in many places also expect an easy opt-out, so if you send the sequence from an email tool rather than your own inbox, check what that tool adds and what rules apply to your customers. For a full automated build of the sequence, see the lead follow-up sequence tutorial, and for getting each enquiry into a sheet without retyping, the enquiry pipeline tutorial covers it.

A café catering enquiry, start to finish

To see the pieces working together, take a café that sells office lunches (illustrative figures). An office manager emails on Monday morning asking about sandwiches and salads for 18 people on Friday.

1. Monday, 9:12. The enquiry arrives. The owner pastes it into the café's Project, which holds the menu, minimum order (10 portions), lead time (48 hours) and delivery fee ($12 under $200, free above).
2. Monday, 9:15. The draft asks one question the enquiry left open: any dietary requirements? It quotes $13.50 a head for a sandwich-and-salad box, so $243 for 18, delivery free.
3. Monday, 9:20. The owner checks the sum ($13.50 × 18 = $243), confirms Friday is feasible, and sends it with a reply-by of Tuesday noon, which leaves a day's margin inside the 48-hour lead time for a Friday lunch.
4. Tuesday, 12:00. No reply. The row's "touch 1" date was set to Tuesday, and the AI draft is a two-line nudge: "Just checking this reached you, and happy to adjust numbers or swap fillings." It goes out.
5. Tuesday, 15:40. The manager replies: two vegans, one nut allergy. The owner asks the AI to redraft the quote with an allergen line, then reads it against the actual recipes before sending, since allergens are a safety matter and not a wording one.

That last step matters: AI saved time on layout and tone, but the allergen content was checked by the person who knows what is in the kitchen. Make "safety and legal content is never taken from AI unchecked" a written rule for your own team.

Where drafts go wrong, and how to catch it

These are the slips that show up most often when owners first rely on drafts. Check for them on every message, and tell the AI to flag them.

The learner who was quoted the wrong package

Take a driving school (an illustration) that let AI draft a reply to a parent asking about lessons for a 16-year-old. The draft quoted adult prices, because the fact sheet had only one price list and the enquiry did not say that a learner under 17 books through a different package. The owner skimmed, sent it, and the parent replied with a price from a competitor. Nothing in the draft was false; it simply answered a question that had not been asked precisely. The fix was a single line in the project: "If the learner is under 17, or the age is unstated, ask the age before quoting." Most failures of this type are fixed by one rule, which is why it pays to keep a short "mistakes log" and turn each into an instruction.

Check that it is working after a month

Treat this as a small experiment, not a permanent decision. Four numbers show whether the workflow earns its place, and all of them come from your tracking sheet:

1. Time to first reply. Median minutes from enquiry to reply. If it was 5 hours and is now 20 minutes, the first-reply draft is working.
2. Quotes sent within a day. The share of qualified enquiries that receive a written quote within 24 hours.
3. Follow-up coverage. The share of unanswered quotes that received all three touches. Aim for close to all.
4. Quote-to-job rate. Won jobs divided by quotes sent, compared with the month before. Small samples bounce around, so look at three months where you can before drawing a conclusion.

A rough sum shows what is at stake (illustrative): a studio sending 30 quotes a month at an average of $400, winning 40% of them, books $4,800. If a consistent follow-up lifts the rate to 45%, that is $600 extra a month, and the cost is a prompt library plus one paid AI plan. If it lifts nothing after three months, the problem is probably the price, the offer or the lead quality, not the follow-up, and that is useful to learn cheaply.

A weekly rhythm that takes under an hour

You do not need software to run this. A weekly slot is enough for a small firm:

• Monday, 15 minutes. Open the tracker, filter for rows whose next action is due, and paste each into the Project with the instruction "draft the next touch".
• Same sitting, 20 minutes. Read every draft against the quote and the fact sheet. Edit, send, and tick the row.
• Friday, 10 minutes. Move won, lost and silent rows to their outcome, and add any new "never do this" rule that the week's drafts exposed.

The weekly routine for open quotes goes deeper on this rhythm, including how to prioritise by quote value. And if enquiries arrive through several channels, a single AI inbox for WhatsApp, Instagram and Facebook keeps replies from being missed before any of this starts.

Keeping your own voice

Buyers can tell a template from a person, and small firms win partly because they sound like a person. Two things keep drafts sounding like you. First, give the AI real samples: paste three of your own sent replies that you are happy with and tell it to match their length, vocabulary and sign-off. Second, edit every draft lightly before sending: change one phrase, add one detail only you would know, such as "the studio is quiet on Thursday mornings". A single human sentence in each message does more for trust than any amount of polish.

Finally, be honest about AI where it matters. If a customer asks whether a person is writing to them, say yes, because a person is: AI drafted, you decided. If you ever move to a chatbot that answers on its own, say so clearly; that is a different tool with different risks, and the customer should know which one they are talking to.

Questions owners ask before letting AI near a quote

Should AI send replies to leads without me reading them?

Not while you are learning what it gets wrong. Let AI draft and you press send for the first month, then loosen only the lowest-risk replies, such as an acknowledgement that confirms receipt and promises a reply time. Anything containing a price, date or promise stays with a person.

How many follow-ups is too many?

Three touches over about two weeks suits most small businesses, and a short final message that closes the loop often draws a reply when the earlier two did not. Stop at once if the person says no or asks to be left alone, and record that in your CRM so nobody chases them later.

Do I need a CRM to do this?

No. A spreadsheet with columns for name, date, quote value, last contact and next action is enough to start. A CRM (customer relationship manager) helps once you have more than a few dozen open enquiries, because it can remind you and keep every message in one place.

Can I paste a customer's email into ChatGPT or Claude?

Check the plan's data settings first. Business plans of both do not train on your content by default, and consumer plans let you switch training off in privacy settings. Even so, leave out anything you would not put on a shared drive, such as card numbers, health details or ID documents.

Further reads

• How Much Does It Cost to Automate Lead Follow-Up With AI? — What automating the three touches costs per month.
• How to Automate Buyer and Vendor Follow-Up With AI — Apply the same follow-up thinking to suppliers.
• Remodelling Leads: An AI-Drafted Follow-Up Sequence for Open Quotes — A longer sequence for high-value open quotes.
• Following Up Silent Quotes With AI: A Decorator's Playbook — See the silent-quote problem handled in one trade.
• How to Write Instagram Captions With AI That Sound Like You — Keep the brand voice consistent beyond sales emails.
• Instagram for Small Businesses: Plan, Caption and Reply With AI — A month-plan prompt, caption drafting, a reply library for comments and DMs, and the review step that keeps it sounding like you.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI ChatGPT Business pricing page and Data controls help article, Claude pricing page, HubSpot free CRM pricing page (all checked November 2026).