Photos on WhatsApp and the AI agent: what it can read from an image (products, receipts, screenshots, damage, documents), what it cannot (measurements, authenticity, medical judgement), how to instruct it, and when a photo should always reach a person
A large share of WhatsApp messages to a business are photos: the thing that broke, the receipt, the screenshot of an error, the item they want to match. An AI agent can look at them, which is useful, and the usefulness has a boundary that an owner needs to know before relying on it.
Quick answer
An AI agent can look at a photo a customer sends and describe what it sees: identify a product or part, read text on a receipt, label or screenshot, describe visible damage, and use that to answer or to collect what a person needs. It cannot measure precisely, verify that an item is genuine, assess whether damage is dangerous, or make medical or structural judgements, and it should be told to hand those to a person. Instruct it on what to do with photos in your business ("describe the damage and ask for a second angle; a person quotes"), keep sensitive images off automated handling, and define the categories that always go to a human. ReplyKit agents see images sent on WhatsApp and follow those rules.
What it can read
| Photo | What the agent can do | Reliability |
|---|---|---|
| A product or part | Identify the type, brand if visible, model if printed | Good when the label is visible |
| A receipt or invoice | Read amounts, dates, order numbers | Good for clear photos |
| A screenshot of an error | Read the message and answer from your help content | Good |
| Visible damage | Describe what is visible and ask for angles a person needs | Good for description, not for severity |
| A document | Read printed text; handwriting less reliably | Variable |
| A room or space | Describe the scene for a quote intake | Fair; no measurements |
What it should not do
- Measure. "How big is this" from a photo is a guess.
- Authenticate. Whether a watch or handbag is genuine is a person's call with the item in hand; the agent describes and hands over, as in luxury resale.
- Judge safety. A cracked beam, a burn, a rash: describe, give any safety line you wrote, and hand over immediately.
- Diagnose. Medical, dental, veterinary and structural assessments are for professionals; the agent books the appointment.
- Approve a claim. Warranty and insurance decisions are human, with the photo attached; see after-sales and warranty.
Writing the photo instructions
When a customer sends a photo: describe what you can see in one line and confirm it with them. For {your category, e.g. appliance repair}, ask for the model label and a photo of the fault, then hand over for a quote. For anything that looks like a safety issue ({your list}), say {your safety line} and hand over immediately. Never estimate size, value, authenticity or severity from a photo.
Test with ten real photos from past chats, including a blurry one, and check the agent describes rather than judges. The general testing routine is in testing before going live.
Privacy
Photos can contain faces, documents, addresses and medical detail. Keep the categories you do not want processed automatically out of the chat (ask for them through a secure form), and set retention so images are not kept longer than the conversation needs. The data rules are in customer data safety.
Frequently asked questions
Can an AI agent see photos sent on WhatsApp?
Yes. It describes what is visible, reads printed text, identifies products and uses that to answer or collect what a person needs.
Can it tell if damage is serious?
No. It describes and hands over. Safety and severity judgements are for people.
Can it verify an item is genuine?
No. Authentication needs the item in hand; the agent collects photos and hands over.
Can it read receipts and screenshots?
Yes, amounts, dates, order numbers and error messages, when the photo is clear.
How do I tell it what to do with photos?
A short instruction block: describe, confirm, ask for what a person needs, hand over the categories you list, never estimate size, value or severity.
Sees the photo, knows its limits
Describes, collects what a person needs, hands over what matters. 7-day free trial, no card.
Sources and further reading
- WhatsApp Business Messaging Policy · opt-in, opt-out and prohibited-use rules every business must follow
- ReplyKit Privacy Policy
ReplyKit is an independent product and is not affiliated with or endorsed by Meta or WhatsApp. This article is general guidance for running a small business, not legal, financial or regulatory advice.