Answer questions about images customers send

What image reading does

When a customer sends a picture, your agent reads it and answers about what is in it. A shopper who photographs a torn package and asks "can I return this?" gets an answer about the damage, rather than a request to describe it in words.

Your agent reads the picture and then answers from your trained content as usual. So it can see that a receipt shows order 1043 and a total of $42.10, and it will use your own refund policy to answer the question about it. If something is not in your trained content, it still says so.

This works with whichever AI model your agent uses. There is nothing to switch on and no setting to choose.

Where it works

A picture on its own is enough. The customer does not have to write anything alongside it.

Discord works differently: files shared there go to your team in the Inbox rather than to the agent, so a person sees them instead.

What your agent does with a picture

• Reads the text in it. Order numbers, reference codes, model and serial numbers, prices, dates, and error messages are read out of the image.
• Describes what is visible. Damage, a wrong or missing item, a screen state, a label on a box.
• Answers the actual question. The picture is treated as something the agent saw, so it refers to it naturally instead of saying it cannot see images.
• Remembers it for the rest of the conversation. A follow-up question like "and the one next to it?" still knows what the picture showed.

Your agent never guesses at detail that is not visible. If a photo is too blurry or too dark to read, it says so and asks for a clearer one.

What your team sees

The picture itself appears on the customer's message in your Inbox, so whoever picks up the conversation sees exactly what the agent was looking at. The customer's message shows their own words, not the agent's reading of the image.

If a teammate has already taken over a conversation, the picture is added to the Inbox as normal and the agent stays quiet, just as it does for a text message.

Limits

A customer who sends more than three images in one message gets an answer about the first three. All of the pictures are still stored on the conversation for your team.

Troubleshooting

The agent replied without mentioning the picture. The image may not have been readable. Ask the customer to resend it in better light, straight on, and close enough to read any text. Very dark, heavily cropped, or extremely blurry photos often cannot be read.

The agent said it cannot help with what is in the picture. Reading the image and answering about it are two different things. Your agent can see a photo of a product, but it can only answer using content you have trained it on. If the answer is not in your knowledge, add it under Sources or as an Instant Answer, then retrain.

A customer sent a file and nothing happened. Only images are read. PDFs, documents, and text files are stored on the conversation for your team to open, and the agent acknowledges receiving them without describing the contents.

A screenshot of a long page was only partly answered. Small text in a full-page screenshot is often unreadable. Ask the customer for a closer crop of the part their question is about.

The picture appears in my Inbox but the reply ignored it. This is what happens when reading the image did not finish in time, usually a very large file or a slow connection. The reply still goes out, and your team can see the picture and follow up.

Does this use extra message credits? No. Reading an image does not cost anything beyond the one message the reply already uses.

A customer sent a photo while my teammate was handling the chat. Then the agent does not reply at all, which is the normal handover behaviour. The photo is added to the conversation so your teammate can see it.

Can customers send video? Not for reading. Video sent on a channel that allows it is stored on the conversation for your team.