Skip to content

Media Reading for Connected AI Apps and Weave ​

INFO

Requires the Starter plan.

At a glance ​

LocationSettings → AI Assistant (MCP) → Media reading
PlanStarter
Related features6
media readingphotoimagestickerhandwritingscanned pdfchatgptwithheldreaderautoquickcarefulmonthly limitpauseread ahead

The Media reading page (Settings, AI Assistant (MCP), Media reading) decides which files in your chats Spun reads into text for your AI assistants, so ChatGPT, other connected AI apps and Weave can answer questions about them. Only an administrator can open it.

Nothing is read until an administrator switches media reading on (agreeing to the sentence shown) and chooses the chats: chats to include and chats to exclude, each picked by chat, by saved filter or by label, and an excluded chat is never read.

Each kind of file has a Read switch and a Show in results switch: Photos (a short description and the printed text in the picture), Handwriting (off by default; off, Spun names a handwritten note without copying it; on, it is copied and marked as possibly containing mistakes), Stickers (one short label for each different sticker, read once and reused), Documents (the text inside PDF and Word files; spreadsheets and presentations are not read) and Scanned PDFs (read page by page like photos, the first pages of each file, 3 unless you choose another number).

With Read on and Show in results off, the AI is told only what the item is, for example a handwritten note. Identity documents and cards are always recognised, and Show in Weave results decides whether Weave may show their details. Voice notes are read from their transcripts and follow your transcription setting.

Videos are not read: an AI only sees that a video was sent and how long it is. Under How photos are read, the reader is Auto (recommended: the quick reader first, the careful reader only for photos that are hard to read), Quick (clear photos and printed text) or Careful (small print and tables).

When to read is Only when an AI asks (the default: a photo is read the first time an AI asks for it, so that answer takes a few seconds longer) or Ahead, as media arrives (every photo in the chosen chats is read when it comes in, so answers are instant), and with Ahead, How far back can also read the stored files of the last days once.

While those older files are being read, a photo that arrives now is read before them. Get better reading in advance, with the Auto reader, gives photos that need the careful reader that reading right away in the background, so nobody waits for it later. Only files Spun has stored can be read.

The page shows what your choices are expected to cost before you save, and has a Monthly limit (reading pauses when it reaches the limit or when your AI credit runs out, and you get a notice), Pause reading and Resume reading, Read again (one chat or a date range with the Quick or Careful reader, showing the cost before anything is read) and View changes to this page, since every change is recorded.

Reading done in the background (Ahead, How far back and a Read again started on this page) is planned work: it uses AI credit like any reading, but the safety guard that watches for unusually fast AI use does not count it, so reading a large number of photos does not pause your other AI features.

If the safety guard pauses AI on your line for another reason, reading stops with the rest of AI, the page shows Paused: AI is paused by the safety guard, and administrators get a notice that says so.

That is not a lack of AI credit: messaging and the readings already made keep working, and an administrator can resume AI with Resume AI now on Plan & Usage, or it resumes by itself within 24 hours. The notice Media reading stopped: no AI credit left is sent only when the credit is really used up.

Readings are part of paid plans: during the free trial and on Free MCP nothing is read.

What each AI app receives for a photo:

  • ChatGPT, and any other OpenAI app such as Codex, gets the reading as text, and gets the image itself only for an organization Spun approved, when its administrator switches it on
  • Weave also gets the reading as text and never sees the picture
  • Other connected AI apps, such as Claude, get the image itself

A ChatGPT connection gets photo readings only when its access includes photos and voice-note transcripts.

On ChatGPT connections ID, card and password numbers are always withheld, a rule of OpenAI for apps in ChatGPT that cannot be turned off: a marker such as [passport number withheld] takes the number's place, and identity documents and medical documents are named, not described.

When something is withheld, the answer ends with an Open in Spun link to a Withheld from ChatGPT page, where after signing in you see the withheld items of the last 24 hours in full (each opens only for someone with access to that chat), with Open in chat, Ask Weave about this chat and Report a wrong removal.

An administrator can also open the same page without the link: in Settings, AI Assistant (MCP), Connected AI apps, each ChatGPT or Codex connection has a Withheld in the last 24 hours button with the number of items, and it shows only items from chats that administrator can open.

Send the actual image to ChatGPT, under Rules for ChatGPT, is off and locked until Spun opens this choice for the organization after looking at its history; until then Tell us you need this records the request so Spun looks at it.

Once Spun has opened it, an administrator can switch it on by agreeing to a sentence saying that every photo ChatGPT asks for in the chats of this organization is then sent to it as it is, identity documents and cards included, with no check by Spun, that anyone whose ChatGPT is connected to the organization can ask for photos, that text is still filtered and that chats excluded on this page are not sent.

If Spun withdraws the choice, or opens it again after a withdrawal, a switch that was on has no effect and the row says so; it must be switched on again and confirmed. This switch is separate from what Spun reads on this page.

In the inbox, a photo that has a reading shows Show photo text under it and in its message menu: it opens the description, the printed text and, for an identity document, the name, number, expiry and whether the check digits match, marked Read by AI, may contain mistakes, with Copy text from photo.

It shows only to people who can open that chat, stays closed until opened (each browser remembers which ones you opened), and never shows in an excluded chat or when Read is off for photos.

For a photo not read yet, Read this photo in the message menu shows the cost in dollars and AI credit tokens and reads that one photo when you confirm, even in a chat that is not chosen (never in an excluded one); anyone allowed to use AI can press it, and a photo that already has a reading is never charged again.

The inbox search also finds a photo by what is in it: type at least 3 characters, and a photo whose printed text or description holds them is listed with a Photo mark and the matching words, and opens at that photo.

It searches only readings that are already stored (nothing is read and nothing is charged), in chats that are chosen on this page and not excluded, while Read is on for photos, and only in chats you can open; a chat excluded later stops matching at once.

Inside the inbox a photo of an identity document is found by its number or by the name on it, and handwriting is found only while Handwriting is read. A search made by ChatGPT or another connected AI app never looks in photo readings.

The Chat Summary of a chat whose photos may be read also uses the stored readings, and can read the photos that have none after showing the cost (see AI Summarize).

How to use ​

  1. As an administrator, open Settings, AI Assistant (MCP) and choose Media reading.
  2. Under Which chats, add the chats, saved filters or labels to include, and any to exclude.
  3. Under What to read, switch Read on for the kinds you want (Photos, Handwriting, Stickers, Documents, Scanned PDFs) and decide Show in results for each; for identity documents and cards, tick Show in Weave results if Weave may show their details.
  4. Under How photos are read, choose the reader (Auto, Quick or Careful), When to read (Only when an AI asks, or Ahead, as media arrives with How far back), and whether to Get better reading in advance.
  5. Check the estimated cost, set a Monthly limit if you want one, and press Save; the first time, confirm with Agree and switch on.
  6. Use Pause reading to stop all reading, Read again to read one chat or a date range again, and View changes to this page to see who changed what.
  7. When ChatGPT shows an Open in Spun link, open it and sign in to see what was withheld.
  8. In a chat, press Show photo text under a photo to see what Spun read in it, or open the message menu and choose Read this photo to read one photo after seeing its cost.
  9. In the inbox search, type words that are printed in a photo (at least 3 characters): the photo is listed with a Photo mark and the matching words.

FAQ ​

  • How do I let ChatGPT read photos in my chats?
  • Which chats does Spun read photos in?
  • Why does ChatGPT get a description instead of the picture?
  • Why was a passport number withheld from ChatGPT?
  • What is the Withheld from ChatGPT page?
  • What is the difference between the Auto, Quick and Careful reader?
  • What does Get better reading in advance do?
  • How do I pause media reading?
  • Why did media reading stop?
  • What does Paused: AI is paused by the safety guard mean on the Media reading page?
  • How do I set a monthly limit for media reading?
  • Can Claude get the real image?
  • Can Spun read stickers and handwriting?
  • Why are photos not read during my trial?
  • Can I send the actual image to ChatGPT?
  • How do I see the text Spun read in a photo?
  • How do I read one photo in the inbox?
  • Can I search for the text inside a photo?
  • Why does the search not find a photo by its text?
Was this page helpful?

Spun Docs - the documentation for Spun, the WhatsApp team inbox by Spun Life LLC.