Buying Gift Cards Inside ChatGPT: How We Brought Faydo to AI Assistants

Faydo exists to do one simple thing: help you pay less, every time. You buy a digital gift card for a brand you already shop with — Amazon, Myntra, Starbucks, BigBasket, PVR — at a discount, pay by UPI, and redeem it at checkout like cash. The saving is quiet but real, and it repeats every month.

Lately, how people shop has started to change. A growing number of our customers don’t open an app to look something up — they ask an AI assistant. “What’s the discount on Starbucks?” “Buy me a ₹1,000 Uber gift card.” So we asked ourselves a straightforward question: if that’s where people are, why should they have to leave the conversation to use Faydo?

This post is about how we brought the Faydo catalog inside ChatGPT, Claude, and WhatsApp — and the one problem that turned out to be far harder than the engineering.

The idea: Faydo as a set of tools an AI can use

There’s now an open standard for exactly this, called the Model Context Protocol (MCP). Think of it as a universal adapter: it lets an AI assistant discover and use a set of “tools” that a service like Faydo provides — search brands, get a brand’s discount and denominations, start a purchase, check payment, view my gift cards.

We built a Faydo MCP server that exposes exactly those tools. Once you connect it, your AI assistant can browse our live catalogue and walk you through a purchase — the same commerce engine that powers our website and app, just reachable from wherever you’re already typing.

That part is the plumbing. It works. The interesting problem was somewhere else.

The hard problem: AI assistants are confident, even when they’re wrong

Large language models are trained to be helpful and fluent. Ask one about a discount and — if you don’t give it the facts — it will happily invent a plausible-sounding answer. Early in our testing, a question like “how much can I save on Starbucks?” would produce something like:

“Starbucks gift cards offer 12% off for in-store use. For example, on a ₹750 bill, a ₹750 gift card would cost about ₹660, saving you around ₹90…”

Every number after the discount was made up. The model had invented an example amount, invented the math, and invented a restriction we never stated. In commerce, that’s not a quirk — it’s unacceptable. Nobody should be quoted a price we didn’t set.

The root cause was subtle: our tools were handing the model too little and letting it fill the gaps. So we flipped the design principle:

Give the model facts, not freedom.

Concretely, that meant rebuilding what our tools return:

• Real prices, computed by us. The brand tool now returns the exact price table — what you pay and what you save at each amount — derived from the live discount on our servers. The model doesn’t calculate anything; it relays.
• A clear “don’t invent” instruction baked into every tool: never fabricate prices, savings examples, store restrictions, or offers that aren’t in the result.
• The real redemption steps and terms, pulled straight from the brand’s record, so “how do I redeem this?” is answered from truth, not guesswork.
• A link to the brand’s Faydo page in every response, so you can always verify and complete the purchase on our site.

The difference is night and day. The same question now returns a clean, accurate answer sourced entirely from live data — no invented examples, no phantom restrictions.

Confirm before you buy

The second principle was just as important: an AI should never spend your money on an assumption.

So the purchase tool is gated. Before any order is created, the assistant has to show you the exact amount and final price and get your explicit “yes.” And the server itself validates the amount against what the brand actually offers — if you ask for an amount a brand doesn’t sell, the order is refused with the valid choices, rather than quietly created. Confirmation isn’t just a polite prompt; it’s enforced.

What it feels like now

Connect Faydo to your assistant once, and then, in plain language:

• “What’s the best discount on fashion gift cards right now?” → a real, ranked answer from our catalogue.
• “Show me Starbucks options.” → the actual denominations, each with its exact price and saving.
• “Buy the ₹1,000 one.” → a confirmation, a UPI link, and your gift card delivered after you pay.

All inside the chat. All at Faydo’s discounted prices. Nothing invented.

Where this is going

Bringing a store into an AI assistant is less about connecting APIs and more about trust — making sure the assistant only ever speaks facts about prices, availability, and terms, and never acts on your behalf without a clear yes. Get that right, and “shopping by conversation” stops being a gimmick and starts being genuinely useful.

We’re just getting started. If you use ChatGPT or Claude, you can connect Faydo today and turn “what’s the discount on…?” into an actual, discounted purchase — without ever leaving the chat.