blog
Conversational AI for Sales: How to Turn Your Offer Into Knowledge the Agent Can Sell From
An AI sales agent can only sell from what you give it. Here is how to write the FAQs, inclusions, context and tiers that actually do the selling.
Mike Davis · 9 min read ·
Conversational AI for sales is only as good as what it knows about your offer. So skip tone and scripts for now. Write down what you sell the way a buyer asks about it: direct FAQ answers, a What's Included list with real deliverables, honest context on who it suits, tier names that explain themselves, and an identity that sets how the agent behaves.
This guide covers the knowledge side and nothing else. You'll see what goes in the required FAQs, how custom FAQs can answer objections before anyone raises them, what belongs in Additional Context, how tiers appear in the DM, and why editing a live offer doesn't reach conversations already underway.
What does conversational AI for sales actually know about your offer?
An AI sales agent knows what you configured. Nothing more. In Cloziq, the agent answers from five places: the offer details, the pricing, the What's Included list, the FAQs and the Additional Context. Ask it something none of those cover and it says it doesn't have that detail. It won't invent one.
That's the right behavior, and it changes your job. You stop writing replies and start writing the source material an AI sales agent for Instagram DMs pulls from. Give it a thin offer and you get an agent that keeps saying "I don't have that detail." Give it a well-documented one and it handles the payment-plan question at midnight and keeps the conversation moving.
When the agent can't ground an answer in your offer at all, it stops replying and flags the contact Needs Human Intervention. Safe, yes. But each of those handoffs is a lead you now answer yourself, and most of them trace back to one missing FAQ.
Conversational AI vs generative AI vs predictive AI
People use these terms interchangeably. They aren't, and the difference tells you what you're actually setting up.
- Predictive AI scores or forecasts. It reads past data and guesses which lead is likely to buy. CRM lead scoring works this way.
- Generative AI produces new text, images or audio from a prompt. A general chat assistant is generative.
- Conversational AI runs a back-and-forth exchange toward a goal. A DM sales agent uses generative models underneath, but it's held to a structure: questions to ask, facts to stay inside, an outcome to reach.
That old quiz line, "conversational AI is only useful in customer service departments, true or false," is false. Sales is where structure pays off most, because every reply either moves a buyer toward a call or checkout or loses them. Voice assistants sit in the same category, but if you're a coach or creator selling on Instagram, your conversation happens in text, in the DM thread, which business tools reach through Meta's Instagram Messaging API.
Start with the manual audit: mine your own DMs
Don't open a builder yet. Open Instagram and scroll back through a few weeks of sales conversations. Copy every question a buyer asked into one document, word for word, typos included.
Sort them into three piles:
- What they get (format, length, access, contents)
- Fit ("is this for beginners", "I work nights", "I've tried this before")
- Money and risk (price, payment plans, refunds, what if it doesn't work)
That document is your real FAQ. It's also the clearest picture you'll get of where your Instagram DM marketing leaks, because the questions you keep getting are the ones your content never answers. Keep it open. Everything below gets filled in from it.
What to put in the five required FAQs
Cloziq's offer builder asks five preset questions you have to answer. They're fixed, and they differ between a Service offer and a Digital Product offer, so a coaching program and a template pack get different prompts. Treat them as the floor, not a form to rush.
Three rules make an answer usable in a live DM:
- Answer first, then add one supporting detail. The agent rephrases your answer for the thread, so hand it something direct.
- Specifics, not adjectives. "Lifetime access, including future updates" beats "flexible access." "Weekly group call plus voice-note feedback" beats "ongoing support."
- No outcome promises you can't back. Anything you write here can be repeated to a buyer. If it wouldn't go on your sales page, cut it.
Here's the difference in practice. Weak: "You'll have plenty of support throughout the program." Usable: "You get a weekly group call, and between calls you can send voice notes in the client chat." Only the second gives the agent something true to say when someone asks "how much help do I actually get?"
Use custom FAQs to answer objections before they come up
Beyond the five presets, you can add up to 20 FAQs of your own. Spend most of them on objections. That's where sales conversations stall, and where a generic chatbot falls apart.
When a buyer pushes back, the agent follows one pattern. It acknowledges the concern, gives one genuine reframe grounded in the offer, then checks whether that was the real blocker. It won't run the same argument twice and it won't badger. The reframe is only as strong as what you documented.
Good objection FAQs, lifted from the fit and money piles:
- "I've done a program like this before and it didn't stick. What's different?"
- "I travel a lot for work. Can I still keep up?"
- "Is there a payment plan?"
- "I'm a complete beginner. Is this too advanced?"
- "What happens if I fall behind?"
Phrase each question the way buyers say it, not the way you'd title it on a sales page. Then answer with the fact that settles the fear. For the travel one, that could be that every session is recorded and can be watched on your own schedule. For the in-thread side of this, see the framework for selling your offer in the DMs.
Make the What's Included list specific enough to quote
Every offer, and every tier if you use them, carries a What's Included list of up to 12 items. "What do I get?" is one of the most common questions in any sales DM, and the agent answers it from this list. Write each item as a deliverable.
Same item, two versions:
- Vague: "Nutrition guidance"
- Quotable: "A personalized macro plan, adjusted every month based on your check-ins"
You don't need all 12 slots. A handful of sharp items beats a padded list, and filler like "community" or "accountability" gives the agent nothing to repeat. If an item needs a sentence of explanation, keep the item here and move the explanation to an FAQ.
What belongs in Additional Context
Additional Context gives you up to 7 optional sections for anything the agent should know that isn't a direct FAQ. Write it like the briefing you'd give a new setter in week one.
Worth including:
- Who the offer is for, and who it isn't for
- What happens after purchase: onboarding, when access starts, who reaches out
- Your background and method, stated plainly, so the agent can explain why your approach works the way it does
- Policies buyers ask about, like refund terms
- Misconceptions about your niche that buyers walk in with
Not worth including: pricing (it has its own step), qualification questions (Sales Flow), and rules like "never discuss competitors" (Guardrails). Keep each kind of information in its own box. Dump everything into one and the agent has a harder time telling a fact from an instruction. If you're still deciding which questions should qualify a lead, read the guide on lead qualification questions and the order to ask them.
How do pricing tiers show up in an AI sales conversation?
Variant-Based Pricing lets you set up to 3 tiers, each with its own name, price and What's Included list. The agent presents tiers by name in the DM, and when the prospect picks one, it creates checkout for exactly that tier. The tier name is something the buyer reads and chooses from.
So name tiers by what changes. "Self-Guided", "Coached" and "Coached plus Weekly 1:1" explain the difference before anyone reads a line. "Silver, Gold, Platinum" forces the agent to explain all three, every time.
How price comes up depends on the goal:
- A Close Sale agent discusses price openly from the pricing you configured. Checkout runs on Stripe Checkout, where Stripe promotion codes work, so a launch discount doesn't need its own tier.
- A Book Call agent holds price for the call on purpose and steers there. With Calendly connected and Book in chat switched on, it books the call in the thread. Your tier details still shape how it describes the offer.
The pricing modes are fixed: Single Price, Subscription (monthly or yearly), or Variants. A subscription can run ongoing or for a fixed number of payments.
Agent Identity sets behavior, not knowledge
Agent Identity, 50 to 3000 characters, describes the agent's personality and how it carries itself. The role is inferred from the offer and goal, so you don't need to explain that it sells your program. A Generate with AI button drafts a first version you can edit.
The usual mistake is stuffing offer facts in here: prices, deliverables, testimonials. Don't. Facts belong in the offer, where they're structured and versioned. Identity is for lines like "warm but brief, asks one thing at a time, never pushes someone who says they need to think." Tone is a separate preset on the same tab.
Whatever you write, the agent won't claim lived experience or pose as a specific person. Asked if it's a bot, it answers with the role you gave it. Write with that in mind. The wording of the replies themselves is its own job, covered in building an AI conversation flow that reads like a human exchange.
Why editing a live offer creates a new version
Offers are always editable. If one is already in use, saving creates a new version. Conversations in progress stay on the version they started with, and new ones get the update.
That protects the buyer. Someone halfway through a thread about your program won't watch the price change mid-conversation. The cost: fixes don't reach people already talking. Correct a wrong FAQ answer and anyone mid-thread still gets the old one.
Time your edits accordingly. Price changes, new tiers and removed deliverables go in between launches, not in the middle of one. For anything time-sensitive, keep an eye on in-progress contacts for a day or two after you save.
Where automation fits in conversational AI for sales
All of the above works on paper. The audit, the FAQ answers and the tier names will sharpen your own sales calls even if you never automate a thing. Automation earns its place once that material exists and DMs arrive faster than you can answer them.
In Cloziq, the offer is a five-step wizard: Basic Information, Pricing, FAQs, Additional Context, Payment. You can also import an offer from a URL, and Cloziq prefills the wizard from your sales page for you to review. Qualification steps live on the agent's Sales Flow tab. And Cloziq is free, no card needed.
Comparing tools for Instagram lead generation? Judge them on how they handle offer knowledge, not on how many trigger types they list. The roundup of the best AI appointment setters compared on price, channels and booking is a fair place to start.
Key takeaways
- An AI sales agent answers only from your offer details, pricing, What's Included, FAQs and Additional Context, so every gap becomes a stalled DM or a handoff to you.
- Build FAQs from real buyer questions, and spend most of the 20 custom slots on objections, phrased the way buyers say them.
- Write What's Included items as deliverables the agent can quote, and name tiers by what changes between them.
- Keep facts in the offer and personality in the Agent Identity. Pricing, qualification and rules each have their own place.
- Saving a live offer creates a new version. Conversations in progress stay on the old one, so make big edits between launches.

