Understand customer voice notes, pictures and PDFs
Receiving a file and understanding it are different things. This guide helps you choose how omotto handles attachments customers send and check what it understood. For files your business sends out, see sending files; for reusable business information, see knowledge.
How it works
Section titled “How it works”Understanding first turns supported media into text for the conversation. It does not guarantee that every detail was recognized correctly or that every file format can be read.
| Customer message | How omotto handles it |
|---|---|
| Voice note | With Understand voice messages active, speech is transcribed into written words. Successful text becomes part of the conversation context. This is not a switch for sending voice replies. |
| Photo or supported picture | With Understand images active, omotto creates a text description, including visible writing where it can be read. |
| Sticker | Supported stickers can be described through image understanding. Not every sticker format can be understood. An unreadable sticker does not by itself trigger the unreadable-media rule. |
| Video or GIF | Image understanding uses one early still picture, not a complete viewing of the clip. Do not rely on it to understand later action or everything said in the recording. An item recognized as a GIF has the same quiet unreadable fallback as a sticker; an item treated as a video uses the normal unreadable-media rule. |
| PDF document | omotto first tries to read the text stored in the PDF. Pages that need visual reading, such as scans, also depend on image understanding being available. A document can be fully read, partly read, or unreadable. |
| Other files or a shared location | Do not assume the bot can read them because they arrived. Unsupported or unreadable items follow your unreadable-media choice. |
A voice note can be available to play even when automatic transcription is off. A PDF can be available to open even when omotto could not read its contents. A transcription or description is an interpretation, not the original. Check the original before relying on a name, number, payment detail or other important evidence.
Where to find it
Section titled “Where to find it”Unreadable-media and human-review guidance: open the Handbook and find When … needs the team. The heading uses your bot’s name. Click the relevant setting’s value to edit it.
Voice and image switches: Settings → Bot → Handbook → What [assistant name] does, on the voice-message and image rows. Reading results: the customer’s conversation in the Inbox.
Saved links to the old Support Team page take you to the matching Handbook setting.
Set it up
Section titled “Set it up”Turn voice and image understanding on or off
Section titled “Turn voice and image understanding on or off”These are workspace settings, not separate switches for each line. Their availability depends on your plan and on the service being configured.
- Open the voice-message or image row listed above.
- Find Understand voice messages or Understand images.
- Change Enabled.
- Check the save status at the top. These switches save automatically.
Choose what happens when an attachment cannot be read
Section titled “Choose what happens when an attachment cannot be read”- Open the team-involvement section of the Handbook described above.
- Choose Show advanced options if the unreadable-media section is hidden.
- Find When a file, voice note, image or location cannot be read and click its current choice.
- Choose the action described in the settings table below.
- If relevant, click the wording after Asking for text is guided by to edit Ask-for-text guidance.
- Check the save status at the top before leaving.
Inspect what was understood
Section titled “Inspect what was understood”- Open the customer’s conversation in the Inbox.
- Find the original attachment message.
- For a voice note, use Transcript in the voice player when available. Its button is marked A. Other audio displays can show a Transcript disclosure below the message.
- For a picture or snapshot, open What the assistant saw.
- For a PDF, open What the assistant read.
Settings
Section titled “Settings”Voice and image understanding
Section titled “Voice and image understanding”| Setting | What turning it on does | Default |
|---|---|---|
| Understand voice messages | Enables transcription of incoming voice notes when the feature is available. Turning it off sends new voice notes to your unreadable-media rule instead. | On, subject to plan and service availability. |
| Understand images | Enables descriptions of incoming supported pictures, stickers and video snapshots. It also enables the visual part of PDF reading when available. Turning it off does not disable reading text already stored inside a PDF. | On, subject to plan and service availability. |
A plan without the relevant feature does not show its capability card. A card saying that the service is not configured cannot be made usable simply by changing Enabled. Missing editing permission can also prevent changes. Ask your workspace administrator about permissions and plan eligibility, or contact support if configuration is unavailable.
These settings control understanding, not whether your line receives messages. They also do not grant permission to forward customer attachments to another person.
Unreadable-media choice
Section titled “Unreadable-media choice”| Choice or setting | What it means | Default |
|---|---|---|
| Involve the team | Supplies the unreadable item as evidence for team involvement while omotto keeps helping. This is not an automatic human takeover. See Support Team. | Selected. |
| Ask the customer for text | Guides a reply asking the customer to describe what they sent in writing. This fallback does not itself alert staff. | Not selected. |
| Ask-for-text guidance | Instructions for writing that request. This is guidance, not a fixed message sent word for word. It applies to the ask-for-text choice. | Blank uses omotto’s built-in guidance: explain the reading limitation, ask for a text description, and mention that a teammate can look if needed. |
For certain failed voice notes, the built-in reply guidance is more specific: record again clearly, send shorter notes, or type the main points. Your custom Ask-for-text guidance, if filled in, takes precedence over that built-in guidance.
For example, you might write: “Ask the customer to type the order number and describe the issue. Do not ask them to resend private payment details.” This is an example to adapt, not the default.
When a photo should still reach a person
Section titled “When a photo should still reach a person”When a photo should still reach a person gives guidance about decisions that need a person even after the picture has been understood. It does not control whether omotto tries to describe the picture. In the Handbook, find When an image needs a person. This advanced setting appears when image understanding is active.
The default guidance calls for a person when the photo is evidence for a decision the bot cannot make itself. Examples include damage, defects, wrong or missing items, and matters heading toward a refund, replacement or allowance. It also includes identity or medical documents and apparent payment confirmations that would otherwise be taken on trust. For other pictures, the guidance is to use what is visible, ask for a better photo or a missing detail when needed, and not guess.
You can replace that guidance with your business’s criteria. Leaving it blank uses the built-in default; it does not mean “never involve anyone.” See complaints and allowances for the separate decision and approval rules.
What happens when…
Section titled “What happens when…”An attachment cannot be read
Section titled “An attachment cannot be read”The unreadable-media rule is used when understanding is off, unavailable, rejected or unsuccessful. It also covers unsupported files and shared locations. Stickers and items recognized as GIFs are exceptions: failure to understand them alone does not automatically ask for text or involve the team.
Team involvement and human control are different. A person who wants to answer directly must take over the conversation. Configure where team requests appear through the team channel.
A readable photo still needs a person
Section titled “A readable photo still needs a person”Requiring a person does not turn off the image-reading work or its usage cost. The description can still help your team review the picture.
A customer sends a PDF
Section titled “A customer sends a PDF”A PDF is a document file. Some PDFs contain selectable text. Others are mostly pictures of pages, such as a scanned receipt.
- Text-based PDFs: omotto can read stored text without image understanding, provided document reading is available.
- Scans and visually complex pages: these may need visual reading. That part depends on Understand images, plan eligibility and service availability.
- Mixed PDFs: text may be available on some pages even when others cannot be read. Turning off image understanding can leave useful text while preventing the visual part from completing.
There is no separate owner-facing PDF-reading switch in these settings. Reading starts after a supported incoming PDF is available and has passed its safety check. The same private-document reading is used for supported PDFs received through customer messaging channels and website chat. A channel accepting a document does not guarantee that omotto can read it.
PDF reading is limited to the first 10 pages. The reader rejects files larger than 16 MiB (16,777,216 bytes). A clear PDF with more than 10 pages still cannot be read in full. Ask the customer to send the relevant pages in a smaller document of no more than 10 pages and no larger than 16 MiB.
Password-protected, damaged or difficult documents may also be only partly read or not read at all. Some useful text can be retained even when no page counts as fully read. Check the actual reading result rather than assuming that receiving the file means every page was understood.
A customer’s PDF is evidence for that customer’s conversation. Uploading it does not add it to your knowledge. Use the separate knowledge import workflow for documents intended to teach reusable business information.
A PDF is only partly read
Section titled “A PDF is only partly read”The reading disclosure can show Total pages, Pages read, Pages not read, the available text, and one of these statuses:
| Status | Meaning |
|---|---|
| Waiting to read | Reading has not started yet. |
| Reading | Processing is still in progress. |
| Read completely | The whole known document was fully read. |
| Partially read | Some text is available, but coverage is incomplete. |
| Could not read | No usable reading was obtained. |
Read the text as well as the page lists. Partial text from a page does not mean that page was read in full. For a sensitive decision, inspect the original file or take over the conversation.
Processing takes time or does not finish
Section titled “Processing takes time or does not finish”Attachments can take time to download, pass a virus check and process. A message may appear before its preview or understanding result is ready. omotto waits for pending understanding for a limited time; it does not wait indefinitely for one attachment.
If processing cannot finish, the result may be incomplete or follow your unreadable-media rule. A result arriving late is not a promise that another automatic reply will be sent. If a customer is waiting, check the conversation and handle the next step rather than repeatedly resending the file.
Understanding and usage
Section titled “Understanding and usage”Transcribing voice notes and describing images can incur usage costs. Visual PDF reading uses image understanding too. Reading text already stored in a PDF is separate from that paid visual work. An attachment arriving, being safe to open and being successfully understood are three different outcomes. Check your plan and usage rather than treating a successful upload as proof of included understanding.
You share files with an Expert
Section titled “You share files with an Expert”Experts have their own Photos & files sharing setting. It controls forwarding eligible customer files to that expert, not whether the customer bot understands incoming attachments. Keep the two decisions separate.
Troubleshooting
Section titled “Troubleshooting”- A voice note has no transcript: check Understand voice messages, availability and any failure note in the conversation. Ask for a clearer or shorter recording, or for text. Do not assume every uploaded audio file is a supported voice note. Being able to play it is not proof that transcription succeeded.
- A picture has no description: check Understand images and whether the file is still processing. Ask for a clearer picture or the important details in text. A sticker or video format may be unsupported even if it can be displayed. A preview is not proof that a description was produced.
- The team was involved despite a readable picture: review When a photo should still reach a person and the separate team-involvement guidance. Reading succeeded; a human decision may still be needed.
- A PDF is partly read or too large: check Pages not read, the available text and the file size. Ask for the relevant pages in a smaller, clear, unprotected PDF: no more than 10 pages and no larger than 16 MiB. Alternatively, ask the customer to type the missing facts. Never assume the missing pages contain nothing relevant.
- The file remains unavailable or processing repeatedly fails: ask for the essential information in text, and contact support with the affected conversation and what happened. Do not send passwords or private payment information as troubleshooting material.
Related
Section titled “Related”- Use the Inbox
- Edit the Handbook
- Configure Support Team involvement
- Send files to customers
- Import reusable knowledge
- ‘app/javascript/components/assistant/VoiceCapabilityCard.tsx’