Anyone who runs WhatsApp for a business in Lisbon, Madrid or Buenos Aires knows the pattern. The customer does not type "What is the delivery time to Porto for order 4471?". They hold the microphone button and talk for forty seconds, or they send a photo of the label and a thumbs-up. A chat that only understands typed text answers half of these people badly and the other half not at all.
Since October, voice notes, photos and documents work both ways in SuperCognit: customers can send them on every channel, and your team can send them back from the Inbox. Here is what happens to each kind of message, depending on who answers.
What the customer can send
- Photos, on the web chat, the widget, WhatsApp and Telegram. They are always accepted.
- PDFs and text files, on the same channels, when the agent has file attachments turned on.
- Voice notes. On WhatsApp and Telegram people use the app's own microphone. On your web chat and widget there is now a record button next to the message box, with recordings of up to two minutes. The button only shows when the browser can record, and it is disabled if the person refuses the microphone.
When the AI agent answers
An AI agent treats each of these as part of the message, so the reply takes them into account:
File attachments are a per-agent setting on the General tab, "Accept file attachments". Reading a long document costs more than reading a sentence, so turn it off for agents that only ever need text. Photos are not affected by that switch.
When your team answers
The other half of the story is the moment a person takes over, whether the AI handed the conversation on or the agent is answered only by people. A voice note that nobody can play is a dead end, so the Inbox now behaves like the phone in the customer's pocket.
See what was sent
On person agents, the ones answered only by your team, photos show as thumbnails in the transcript, documents as file chips with their name and size, and voice notes as a player with a waveform. The files are kept with the conversation, so they are still there next week.
Listen at your speed
The player has play and pause, a waveform you can click to jump, and a speed toggle on the time label: 1×, 1.5× or 2×. The same player shows on your chat page and widget, so the customer hears their own note the same way.
Reply with a file or your voice
The Inbox reply box has an attach button and a voice-note recorder. A reply can be text, files, a voice note or only files: up to five files per reply, 5 MB each, as images, audio, PDF or text.
It arrives where they are
The reply goes to the channel the customer is using. On WhatsApp and Telegram a voice note from your team shows as a normal round voice message, not as a strange attachment. On the web chat and widget it plays in the same player.
For agents that are answered only by people, voice notes are kept as audio and not transcribed: the team listens, as they would on their own phone. That kind of agent is covered in chat with no AI at all.
Where this matters most
- Repairs and field service. A technician or a customer sends a photo of the part and a voice note explaining what it does. Your team sends back a voice note with the fix, because explaining it out loud is faster than typing it.
- Hotels and restaurants. Guests send a photo of the booking confirmation or ask, by voice, whether the terrace is open tonight. The AI agent answers both, and books when it can, as in an AI reservation agent for hotels.
- Shops. "Do you have this in blue?" with a screenshot. The agent sees the product; if a person needs to step in, they can reply with a photo of the item on the shelf.
- Paperwork. A customer sends an invoice or a contract and asks one question about it. With file attachments on, the agent reads the document instead of asking the customer to copy the numbers.
Privacy, briefly
Stored files are private. They open only for signed-in members of your workspace with access to the Inbox, and for the visitor in their own conversation. When you delete a conversation, a workspace or an account, its stored files are deleted with it.
Set it up
Decide on documents
On the agent's General tab, turn "Accept file attachments" on if customers send PDFs or text files you want the agent to read. Leave it off for text-only agents.
Connect your channels
Voice notes and photos on WhatsApp and Telegram need the channel connected on the Channels tab. If you are starting from scratch, how to automate WhatsApp Business walks through it.
Turn on hand-overs
So a person can answer the messages the agent should not, set up human support on the agent's Team tab. Your team can now take over from the AI agent explains how hand-overs work.
Team replies with files are also in the API and the platform MCP tools, so a system of your own can send a document into a conversation. The October roundup lists everything else that shipped this month.
Frequently asked questions
Can an AI chatbot understand WhatsApp voice notes?
Yes. A SuperCognit AI agent transcribes a voice note sent on WhatsApp, Telegram, the web chat or the widget, and answers the words as if they had been typed. The transcript goes through the same guardrails as a typed message.
Can customers send photos and PDFs to the AI agent?
Photos are always accepted on every channel. PDFs and text files are accepted when the agent has "Accept file attachments" on, on its General tab; the agent then reads the document as part of the message.
Can my team reply with a voice note?
Yes. The Inbox reply box has a voice-note recorder and an attach button: up to five files per reply, 5 MB each. On WhatsApp and Telegram a voice note from the team arrives as a normal voice message.
Who can open the files customers send?
Only signed-in members of your workspace with access to the Inbox, and the visitor inside their own conversation. Stored files are deleted when the conversation, the workspace or the account is deleted.
