How to handle customer voice notes, photos and PDFs in chat
Customers send voice notes, screenshots and PDFs instead of typing. How an AI assistant transcribes and reads them, replies in text and leaves video to your team.
Product page: Voice notes and media
On WhatsApp, a lot of customers would rather talk than type. Others send a screenshot of the error, a photo of the shipping label, or the PDF invoice they were emailed. A text-only bot answers all of that with "Sorry, I can only read text messages", and the customer has to type out what they already sent you.
Risolva reads voice notes, images and documents so the conversation keeps moving. Below is how each one is handled, whether it arrives on WhatsApp, Instagram, Messenger or web chat. See AI for WhatsApp voice messages for the overview.
How should a business reply to WhatsApp voice notes?
When a voice note arrives, Risolva transcribes it and hands the transcript to the assistant, which answers like it would a typed message. The original note stays in the inbox, so anyone on your team can listen to it.
Fernwood HomeWhatsApp · example conversationReplies are always in text. Risolva does not speak back, and it does not handle phone calls. In practice that is what most customers expect from a WhatsApp chatbot anyway: they send a quick voice note while walking and read the answer later.
What can AI read from a customer's screenshot?
Images are read with a vision model that pulls out all the readable text and writes a short summary. That text goes to the assistant as context. It works well for the images customers actually send to support:
- Order confirmations and receipts
- Screenshots of an error message or a checkout page
- Photos of a label, a serial number or a price tag
- Stickers
Screenshots often arrive in Instagram DMs, like this order confirmation:
Fernwood HomeInstagram · example conversationThe AI works from the text and summary it extracts. It does not verify payments or check that a receipt is genuine, so do not build a process that depends on that.
PDFs and documents
PDFs are read locally first, which is fast and cheap. If a PDF turns out to be a scan with no text layer, Risolva falls back to the vision model. Word documents (.docx) and text-based files (.txt, .csv, .json, .xml and .md) are read directly.
If a customer sends something the assistant cannot read, such as a spreadsheet or an unusual format, it says so politely instead of pretending.
Captions come first
People often send a file with a message attached: "here's my invoice, why was I charged twice?" Risolva answers the caption right away, so the customer is not left waiting while the file is processed, then follows up once the document has been read.
Which media should the AI answer, and which should a person?
Under each bot's channel settings, Media handling has four switches: Respond to voice notes, Respond to images, Respond to videos and Respond to documents. They are set per channel, so you might answer voice notes on WhatsApp but not on web chat.
When a switch is off, the media still shows up in the inbox and your team can open it. The AI simply does not process it or reply to it, which is useful for media you would rather a person handled, like photos of damaged goods that need a judgment call.
What it does not do
Video is not analyzed. Replies are text, not audio. And extracted text is used as context for an answer, not turned into structured fields in another system. If you need details from a document to land in your own system, a custom action can send them to your API. Our custom actions guide has examples.
To try voice notes and media on your own channels, book a demo.
Frequently asked questions
Can Risolva reply to WhatsApp voice notes?
Yes. Voice notes are transcribed and the AI answers in text, using the same knowledge it uses for typed questions. It does not reply with audio.
Which files can the AI read?
Images and screenshots (including stickers), PDFs including scanned ones, Word documents, and text, CSV, JSON, XML and Markdown files. Other formats are acknowledged, and the assistant says it cannot read them.
Does the AI understand videos?
No. Videos are shown in the inbox for your team, but the AI does not analyze them, so a question asked only inside a video needs a typed follow-up or a teammate.
Can I turn media replies off?
Yes. Each bot and channel has separate switches for responding to voice notes, images, videos and documents. When one is off, the media still appears in the inbox for your team, but the AI does not process it or reply.
Do customers expect a voice reply to a voice note?
Rarely. Most people send a voice note because it's quicker than typing, and they're happy to read a written answer. Risolva replies in text and keeps the original audio in the inbox for your team to play.
Should a person handle photos of damaged items?
Usually, because they need judgment. You can switch off image replies per channel so photos still show in the inbox while the AI stays quiet, or have a handoff rule send damage complaints to a person.
