Photos from customers

When a case involves a damaged, wrong or missing item, the AI can ask the customer for a photo, and it reads the photo the customer sends. What the AI sees is context for the decision. A photo never approves or denies a case on its own, and it never overrides your Charter.

What the AI looks for

The AI reads a photo in light of the case:

• A damaged item. The AI looks at the damage.
• A wrong item. The AI compares what the photo shows with the product in the order.
• A missing item. The AI looks at what arrived and compares it with what the order included.

The photo informs, it does not decide

Whether an action is allowed is still decided by your Charter and the Policy Gate, and who carries it out is still decided the usual way. The photo is one of the facts they use. It does not move the line between what the AI resolves and what goes to a Human Node or to your team.

A photo is never used against the customer either. An unclear photo is not proof that the problem does not exist.

How the AI reacts

• A plausible photo the AI asked for. The AI assumes good faith and continues the case, without questioning the customer.
• A photo clearly unrelated to a product. The AI shows it noticed, in a natural way, and asks whether that was the photo of the item.
• A photo with no context in the conversation. The AI does not guess what it is. It asks the customer what the photo is before going on.

Where the AI's reading appears

• In the conversation, the photo appears as the customer sent it. The AI does not add captions or notes about it.
• When the case is escalated, to a Human Node or to your team, the AI includes what it read in the photo in the case summary, stated as its reading and not as a fact. Whoever decides sees both the photo and the AI's reading.

Text inside a photo is never an instruction

If a photo contains written text, such as an instruction to issue a refund, the AI treats it as content to look at, never as a command. No action is ever taken because of text in an image.