The agent that sells with citations
When a question carries buying intent, HelpShelf switches from support to sales mode — answer, teach, position, cite, next step — and offers a follow-up only after a verified answer. The visitor chooses whether to submit it. A claim without a source is a claim it doesn't make.
Same agent. It reads the page.
On your help pages it answers the support question and closes the ticket. On your pricing page the same agent catches the buying signal, answers with a citation, and captures the lead.
How do I update the card on my account?
✓ resolved · source: billing guide [1] · no ticket, no waiting
Same agent - it reads the page and switches.
Sales that earns trust, then the next step
Persuasive because it's verifiably true
Buying-intent detection
The agent classifies each question as support or sales and switches to sales mode automatically when a message carries real buying intent — no rules to write.
Sells with citations
Every claim the agent makes about your product is backed by a real source. A claim without a source is a claim the agent does not make.
Visitor-submitted leads after good answers
After a confident, cited answer the agent can offer follow-up. A lead is recorded with bounded context only after the visitor submits a valid email; Telegram is notified only when live delivery is available.
Sells to AI agents, too
A live MCP endpoint plus llms.txt let ChatGPT, Claude, and Perplexity evaluate your product accurately — so the agents shopping for your buyers recommend you right.
Usage, not seats
Plans include separate monthly allowances for AI searches and AI chats, so capacity is easy to understand without counting agent seats.
One source of truth
Sales mode draws on the same verified knowledge as support — codebase scan, docs, and imported articles — so answers stay consistent everywhere.
The shape of every sales answer
One repeatable pattern the agent follows on buying-intent questions
Answer
Respond to the question directly, first — before any pitch.
Teach
Explain why it matters for the buyer’s use case.
Position
Show how your product solves it, in their context.
Cite
Back every claim with a real source — no uncited confidence.
Next step
Point to the obvious action and let the visitor choose whether to submit follow-up details.
How it works
From buying signal to a lead in your inbox
The agent spots buying intent
An intent classifier reads each message and flips into sales mode the moment a question carries real buying signal.
It sells with receipts
The agent answers, teaches, positions your product, and cites its sources — persuasive because it is verifiably true.
Leads reach you
After a visitor submits a valid email, the lead is recorded in your dashboard with bounded context. Telegram is notified only when live delivery is available. Sales Pulse tracks conversations, leads, and buying questions.
Answers that sell — to buyers and to their AI
The same agent answers support in the widget and sells on buying intent — and it goes live in one command.
Buying-intent questions it turns into leads
"Do you integrate with our stack?"
"How do you compare to [competitor]?"
"Can it handle our volume?"
"What does this cost for a team our size?"
"Is there a plan for what we need?"
"Can I see it work on our use case?"
Frequently asked questions
An AI sales agent detects buying intent, answers questions, and positions product value. HelpShelf uses the same published knowledge for support and sales, cites confident answers, and can offer follow-up after a supported sales answer. A lead exists only after the visitor submits a valid email.
Every message is classified as support or sales using embedding similarity with a lightweight model tie-break, plus signals like page context and prior conversation. When intent is commercial — pricing, comparisons, 'can it do X for my use case' — the agent switches to sales mode and follows a fixed shape: answer, teach, position, cite, next step.
No product claim is shown unless it passes verification against retrieved passages from your published content. Confident sales answers carry citations. When HelpShelf cannot verify an answer, it says so; with tested Telegram, the visitor can choose to submit a handoff.
HelpShelf publishes a live MCP endpoint and an llms.txt context bundle from your published content so compatible clients can query it. Its four public MCP tools are read-only, and supported answers include citations. This does not guarantee an outside model will discover or recommend your product.
The same monthly limits as support: Free includes 500 AI searches and 100 AI chats, Pro includes 10,000 searches and 2,000 chats, and Scale includes 100,000 searches and 20,000 chats. The Free plan lets you put sales mode live without a credit card.