Skip to content

AI Deep Read

Scans, photos and handwriting into files you can edit

Standard conversion needs a text layer. When a PDF is only page images, or the input is a screenshot, receipt or notebook page, AI Deep Read looks at the page the way you do and writes out the text, tables and headings as Word, Excel, Markdown or plain text. One credit per page.

Reads Writes 1 credit per page · needs a free account

How it works

Standard reads the file; Deep Read reads the page

Standard

Opens the file with the matching format engine (office, document, table, image) and moves the structure it finds. Fast, exact, and free within your allowance. Nothing to read when the PDF is a picture of a page: a scanned contract comes out as a blank document or one image per page.

AI Deep Read

Renders each page to an image and sends it to a vision model that transcribes text, rebuilds tables as cells or pipe tables, marks headings and lists, and reads handwriting. The transcript is then written into the target format by the same engines Standard uses, so a DOCX from Deep Read is a real Word document with headings and tables, not an image wrapped in a file.

  1. Step 1

    Drop the file

    PDF or image. The converter checks for a text layer and suggests Deep Read when a PDF has none.

  2. Step 2

    Pick the output

    Word for editing, Excel or CSV for tables, Markdown or text for notes and RAG pipelines.

  3. Step 3

    See the cost

    The page count and credit cost show before anything runs. Nothing is charged for a failed read.

  4. Step 4

    Review

    Warnings list what was uncertain. Check totals and dates before you rely on them.

From real runs

What it reads well, what to check, what it does not keep

  • Printed pages, 200–300 dpi Reads wellBody text, headings and lists come back in reading order with paragraph breaks restored. Two-column articles are re-flowed into one column.
  • Tables with clear rules Reads wellRows and columns become spreadsheet cells or Markdown pipe tables. Numbers are typed; amounts in parentheses become negatives.
  • Phone photos with a slight tilt Reads wellRotation up to a few degrees and uneven lighting do not stop the read. Very dark or blurred photos lose characters first in the smallest text.
  • Screenshots of chats, emails, apps Reads wellText is extracted line by line. Interface chrome (buttons, icons) is dropped rather than described.
  • Handwriting Check itNeat block capitals and cursive on lined paper read well; cramped notes, unusual abbreviations and numbers written over corrections are the usual errors. Always check totals.
  • Multi-page scans Check itEach page costs one credit and is read on its own; a table that continues across a page break becomes two tables with the header repeated.
  • Stamps, signatures, faint watermarks Not keptFaint stamps and signatures are usually skipped or transcribed as a short label. They are not reproduced as images.
  • Exact fonts, colours and page layout Not keptThe output is editable structure, not a facsimile. Keep the original scan if the visual record matters.

Three real conversions

Chase checking statement → Excel

Input

4-page PDF, 87 posted transactions, two transaction blocks per page, repeated Date / Description / Amount / Balance headers.

Output

One sheet with 87 typed rows. Two parenthesised fees became negative numbers. Two 'Ending balance' summary rows were also captured and needed deleting.

Open this converter

Release-note screenshot → text

Input

1400 × 1000 PNG with a heading, one paragraph, three bullets and a blue owner / date / status callout.

Output

'# Quarterly Release Notes', the paragraph re-flowed, three dash bullets, and separate Owner, Due and Status lines. The callout colour and border are gone.

Open this converter

Compressed field note → text

Input

1300 × 900 JPG with grey inspection notes, two pressure readings, a received stamp and a handwritten follow-up line.

Output

Title, valve note, '42 psi' and '39 psi', the received date and the handwritten 'Sam okayed temp patch'. The drawn underline did not appear.

Open this converter

Credits

What it costs

Free

0

credits included — buy a pack when needed

Starter

50

credits a month, included

Pro

500

credits a month, included

Scale

2,000

credits a month, included

Unused monthly credits do not roll over. A one-off pack of 1,500 credits costs $50 and never expires. One credit reads one page or one image regardless of the output format. Compare plans →

Preset for the job

Deep Read workflows

These converters have the output and options already chosen and state up front what the result will and will not contain.

By file pair

Questions about AI Deep Read

What is a credit and how many do I need?

One credit reads one page or one image. A 12-page scanned contract costs 12 credits; a single receipt photo costs one. Starter includes 50 credits a month, Pro 500, Scale 2,000, and a one-off pack of 1,500 credits costs $50. Failed reads are not charged.

How do I know whether a PDF needs AI Deep Read?

Open it and try to select a line of text. If nothing highlights, or the whole page highlights as one block, it is a scan. The converter also checks for a text layer when you drop a PDF and suggests Deep Read when it finds none. A digital PDF converts in Standard mode for free.

Is the output accurate enough for accounting?

Treat it as a fast first draft, not a source of record. On clean statements the transaction rows and amounts match; the errors we see are summary rows captured as transactions, dates read in the wrong locale, and characters lost in faint print. Reconcile totals against the original before you rely on them.

Where do my pages go?

Page images are sent to the vision model for reading and the transcript comes back; they are not used to train anything. Your uploaded input is deleted from our storage within 24 hours and the converted output follows your workspace retention setting.

Which languages does it read?

Latin-script languages read reliably; Cyrillic, Greek, CJK and Arabic scripts read but with more errors in small or stylised text. Pass a language hint (an option on the converter) when a page mixes scripts or uses accented characters heavily.

Can I use it without signing in?

No. Standard conversions run without an account; AI Deep Read needs a free account so credits can be tracked, and credits come with a paid plan or a one-off pack.

Can I call it from code or from an AI assistant?

Yes. Add deepRead=true (or vision=true) to a /api/v1/convert request, or use the convert_file tool with deepRead on the MCP server. Both count credits the same way as the web converter.