Does my ChatGPT app or MCP work as intended?

---
title: Does my ChatGPT app or MCP work as intended?
canonical: https://chosenbyagent.com/questions/does-my-ai-app-work
updated: 2026-10-22
audience: marketplaces
offers:
  - A3
evidence:
  - chatgpt-session-share
  - chatgpt-conversion-gap
  - answer-instability
  - ucp-adoption
---

# Does my ChatGPT app or MCP work as intended?

Short answer: Usually it works in the demo and drifts in production — stale results, lost attribution, or contact and payment slipping off the platform. The channel is small today, so the checks have to be cheap and repeatable.

## What the research shows

- **0.2%** — ChatGPT drove about 0.2% of e-commerce sessions across 973 stores between August 2024 and July 2025. [Marketing Science (INFORMS), 2025, independent](https://pubsonline.informs.org/doi/full/10.1287/mksc.2025.0489)
- **−13%** — Traffic from ChatGPT converted 13% below organic search and 45% below paid search. [Marketing Science (INFORMS), 2025, independent](https://pubsonline.informs.org/doi/full/10.1287/mksc.2025.0489)
- **64%** — About 64% of the top-5 brands changed between adjacent runs of the same assistant question. [MentionBird, 2026, vendor](https://www.mentionbird.ai/blog/chatgpt-vs-gemini-answer-stability)
- **26** — In May 2026 the /.well-known/ucp file was found on only 26 of more than 3 million sites. [PPC Land, 2026-05, vendor](https://ppc.land/only-26-sites-have-implemented-ucp-and-the-big-names-arent-among-them/)

## AI channel check

A3 · from €3,000 (one-off) · indicative · scoped per engagement

Your ChatGPT app, MCP or agent gateway — freshness, attribution, and contact staying on your platform.

- Correctness and freshness of what your app returns
- Attribution and deep links back to your listings
- Contact and payment kept on the platform

## What you get

We drive your ChatGPT app, MCP server or agent gateway with the same buyer intents a person would type, and compare what it returns with your live site.

**Your AI channel against your site**

| Check                     | Result  | Note                                   |
| ------------------------- | ------- | -------------------------------------- |
| Results match the site    | partly  | filters ignored for price ranges       |
| Freshness                 | lagging | sold listings returned for up to a day |
| Attribution               | kept    | UTM parameters survive the hand-off    |
| Contact stays on platform | kept    | no phone or e-mail in app answers      |

_Illustrative values — your report shows yours._

## What this does not claim

- We test your channel as published, through the assistant that hosts it; we do not review its source code.
- Behaviour of the host assistant changes with its releases, so a pass today is dated, not permanent.

## Request this instrument

We reply with a scope, a fixed price and a start date.

Agents: POST /api/engagements (application/json) with {"offer_id": "A3", "on_behalf_of": {"email": "…"}, "agent": {"name": "…"}}. The person confirms by email before we see the request. Details: /agents.md

### Related questions

- [Do assistants show my listings correctly?](/questions/are-my-listings-shown-right.md)
- [What changed after the last model update?](/questions/what-changed-after-the-update.md)
- [Do buyer and seller agents behave safely on my platform?](/questions/are-agents-safe-in-my-chats.md)