# When your business shows up inside ChatGPT: how to wire the operation behind it

> OpenAI now lets companies manage how they show up in ChatGPT. Here is how to connect that new front door to the records, owners and stages your team already uses.

- Type: Technical
- Published: 2026-11-12
- Author: Sentynel
- URL: https://sentynel.ai/sygnals/when-your-business-shows-up-in-chatgpt/
- Language: en

On 10 November 2026, OpenAI introduced Businesses in ChatGPT, which it describes as a way to "manage your business information, get insights about how your business shows up, and turn interest into action in ChatGPT." Customers can now meet your company inside a conversation you do not host. **That is a new intake channel, and it only works if what it says comes from the same records your team uses, and if every interested customer lands in your operation with context, in a stage someone owns.**

## A channel you do not host is still a channel

Most companies will treat this as a marketing task: claim the profile, polish the description, move on. That misses the operational half. Interest that starts in an assistant ends somewhere: on your website, in your WhatsApp, in a call to your store. If that arrival is not designed, it becomes one more conversation that nobody counts.

The safe working assumption is that you will see very little of what happened before the customer reached you. Whatever a platform shares back, design as if the only thing you control is the moment the customer crosses into your systems. That moment needs four parts:

1. **One source of truth** for every fact the assistant can repeat.
2. **A handoff that carries context**, so the customer never starts from zero.
3. **A stage and an owner** for every arrival.
4. **Measurement from your side**, because your own records are the only numbers you can audit.

**From an assistant conversation to resolved work**

1. Assistant conversation — a channel you do not host
2. Tagged handoff — product, intent, source
3. Record created — same CRM as every channel
4. Owned stage — New · In progress · Resolved
5. Outcome recorded — resolved, lost or escalated

## Part 1: one source of truth for what it says

An assistant that describes your business repeats facts: prices, stock, opening hours, delivery areas, return rules. Each of those already lives somewhere in your company, usually in more than one place, and the copies rarely agree.

Before you publish anything to an outside platform:

- **List every fact the platform can show** and mark where it lives today: ERP, e-commerce catalog, a spreadsheet, someone's memory.
- **Pick one system of record per fact.** If prices live in the catalog, the catalog feeds the platform. Nobody types prices into a profile by hand.
- **Give each fact an owner and a refresh rule.** "Stock updates when the catalog syncs" is a rule. "Marketing keeps it current" is a hope.
- **Write the policies as the agent should state them**: short, unambiguous, with the exceptions included. A return policy that only a store manager can interpret will be interpreted by the assistant instead.

**If a fact has no owner inside your company, it should not be published outside it.**

## Part 2: a handoff that carries context

The handoff is where most of the value leaks. A customer who already asked about a product, a size and a delivery date should not have to explain it again to your team.

Design the handoff as a contract:

- **Links that carry intent.** Every link you publish to the platform points to a specific product or action, with parameters for the source and the intent (for example, a source tag for the assistant channel and a product or service ID).
- **A landing step that creates a record.** The page or the WhatsApp entry point writes a contact or a case into the CRM with the source already set, before anyone replies.
- **A first message that shows you know why they came.** "You were looking at the 2-seater in grey" beats "How can we help you?"
- **A fallback for missing context.** If the parameters are lost, the record still opens, tagged as "assistant, unknown intent", so it is counted rather than mixed into general traffic.

## Part 3: a stage and an owner for every arrival

Once the customer is in your systems, the new channel should look exactly like every other one: same queue, same stages, same people. A separate inbox for "AI traffic" is how a new channel becomes a blind spot.

This is the pattern in Sentynel's Rodo case: WhatsApp, social and web enquiries were brought into one service operation with a visible path to resolution, New, In progress, Resolved. In the first measured window, 82% of 604 live-created enquiries reached Resolved, with a median of about one hour. An assistant-sourced enquiry belongs in the same pipeline, with its own source tag, so it can be compared with every other channel.

**Rodo · first measured window**

- 82%: of live-created enquiries reached Resolved
- ~1 h: median to resolved

Source: Sentynel: Rodo operational report, 24 September 2026. The resolution cohort covers 31 August–24 September, 604 live-created enquiries. Figures describe pipeline records, not all conversations. First measured window, no comparable earlier baseline.

## Part 4: measure from your side

A platform may give you aggregated insights. Use them, but run the channel on numbers you can audit in your own records:

- **Arrivals by source**: how many records were created from the assistant channel each week.
- **Share reaching Resolved**, and the median time to get there, compared with your other channels.
- **Context loss**: how many arrivals came in as "unknown intent". A high share means your links or landing step are broken.
- **Mismatches**: every time a customer quotes a price, a stock level or a policy that differs from your record. Log each one with the fact, the date and the owner who fixed it.

The mismatch log is the most useful of the four. It tells you which system of record is stale, and it is the only early warning you will get that the assistant is repeating something you no longer offer.

## Test it before you announce it

Run a short test before you point any traffic at the channel:

1. Write ten questions a real customer asks your team every week.
2. Ask them about your business in the assistant and record each answer next to the fact in your system of record.
3. Follow each link to its landing step and check that a record is created with the right source and intent.
4. Have a team member pick up three of those test records and reply without asking the customer to repeat anything.

Anything that fails goes back to Part 1 or Part 2, not to the profile text.

## What to do next

Pick the one fact customers ask about most, usually price or availability, and trace it from your system of record to what the assistant says. If you cannot name the owner of that fact and the rule that keeps it current, start there. A new front door is only as good as the operation it opens onto.

## Sources

- https://openai.com/index/introducing-businesses-chatgpt
