OCR starts with what is visible in the source
Extraction makes visual information workable. It does not guarantee that every character, value, field, or layout relationship was interpreted correctly.
A supplier sends a scanned quote and you can’t select the text. Open it in Think-Act and ask, “What’s the total, when does the quote expire, and does it include delivery?”
Extraction is not verification. Check important text, fields, summaries, answers, and interpretations against the original document before relying on them.
Windows 10 and 11 now. Mac release planned.
Document workflow
Add a document, let OCR read the text, then ask a question about the part you need.
Extraction makes visual information workable. It does not guarantee that every character, value, field, or layout relationship was interpreted correctly.
Check important values and interpretations before they become inputs for writing, automation, connected systems, or other consequential work.
Source
Bring a supported screenshot, scan, image, receipt, invoice, form, photographed page, or PDF instead of manually recreating the information it contains.
Extract
OCR can make image-based text editable and help prepare useful document details such as names, dates, totals, references, and other visible fields where supported.
Understand
Prepare structured information, summarize supplied content, identify useful details, or ask focused questions about what the document contains.
Review
Check names, dates, totals, identifiers, obligations, missing fields, unclear scans, and important interpretations against the source before treating the result as dependable.
Continue
Copy the reviewed information, use it as Writing Assistant context, bring it into a bounded Desktop Agent task, or use an appropriate configured external capability where the next step requires one.
Source
Visual document
Extract
Usable content
Check
Source comparison
Use OCR on a screenshot, receipt photo, or scanned page. Copy the result into a message or give it to the agent for a follow-up task.
Screenshots & images
Extract visible text from supported screenshots, photographed pages, copied images, interface captures, and similar material.
Scans & image-based PDFs
Work with supported scans and PDFs when the source does not provide useful selectable text.
Forms, receipts & invoices
Prepare names, dates, totals, references, and other visible fields, then verify important values against the original.
Longer documents
Find details, summarize supplied content, or prepare structured information that remains traceable to relevant source pages.
Already working with clean selectable text? The Writing Assistant may be the simpler starting point.
Use OCR for the words on the page. Ask a question when you need a date, a comparison, or a shorter explanation.
OCR
OCR focuses on making words that are visible in the source available as editable text for search, copying, checking, and later document work.
Typical result
Document Intelligence
Use extracted or supplied text to prepare structured drafts, summaries, focused answers, review checklists, or follow-up context where the document workflow supports it.
Typical result
Document Intelligence does not make an OCR error disappear
Interpretation depends on the content it receives. If an important extracted value is wrong, a polished summary or answer can still carry that mistake forward. Review important extraction and important interpretation against the source.
Try “Which charges are optional? Quote the line that says so.” If the document doesn’t say, ask Think-Act to flag the gap.
High-stakes documents need qualified review
Legal, financial, compliance, medical, tax, or regulated interpretation should remain working material until reviewed at the level appropriate to the decision.
Useful document questions
Useful for getting oriented before inspecting the particular clauses, numbers, fields, or passages that matter.
Use the answer as a locator, then verify consequential dates against the corresponding source passage before acting on them.
Prepare a working checklist from the supplied content, while keeping legal or compliance interpretation subject to appropriate review.
Use the result as a review aid rather than proof that the form is complete. Inspect the actual source pages before treating a field as absent.
Question answered does not mean question verified. Confirm the source passages, values, terms, and fields that would matter if the answer were wrong.
Use the document workspace to read the extracted text and ask follow-up questions about the same file.
Source
Original document
Extract
Text & fields
Work
Answers & structure
Review
Checked result
The scan, screenshot, image, receipt, form, or PDF remains the reference for important details.
Extracted text and draft fields are available for searching, correcting, and organizing.
Summaries, answers, structured drafts, and follow-up context stay attached to the source.
Checked details are separated from assumptions before the result is copied, shared, or reused.
Pull the model number off a label or list the due dates from a notice. You don’t need every word to finish the task.
Prepare draft totals, dates, vendor details, references, and other useful values, then confirm financial information before reporting, accounting, or payment-related use.
Extract difficult-to-select text, surface dates or obligations, and prepare summaries or focused answers while keeping legal interpretation subject to qualified review.
Prepare useful field drafts from supported forms or scanned records, then validate missing or consequential information before passing it into another process.
Extract visible text from interfaces, errors, web pages, or screen captures so it can become searchable, editable, or useful as context for the next task.
Locate relevant sections, prepare summaries, or ask focused questions across longer material, then trace important conclusions back to the source pages.
After reviewing the extracted details, use the checked text as context for an email, summary, report, response, or another writing task.
A faint decimal point can change a total. Compare amounts, identifiers, and dates with the original before you reuse them.
Use models, endpoints, connectors, and storage approved for the information involved. Processing and data flow depend on the configuration you choose, so follow your organization’s data-handling policy and review the Think-Act security guidance before processing sensitive material.
Review gate
Check the parts that would matter if they were wrong
Names, dates, totals & references
Compare consequential values with the original document rather than trusting a clean-looking extraction.
Missing or unclear source content
Look again when scans are faint, cropped, noisy, or arranged in a complex layout that makes extraction harder to verify.
Legal, financial & regulated interpretation
Treat summaries, classifications, and interpretations as drafts until the appropriate qualified person has reviewed them.
Original source retained for comparison
Keep the source available so reviewers can trace an extracted value, statement, or answer back to the material it came from.
Downstream use after review
Confirm the extracted content before it becomes input for a message, file change, connected system, or another consequential step.
Read a model number from a photographed label and check it against the photo. Then ask the Desktop Agent to open the matching manual on the manufacturer’s site.
Another Think-Act feature
A reviewed document summary, field set, or extracted passage can become context for writing or a bounded Desktop Agent task without making OCR responsible for the next job.
Skills
Once a document method is understood and reviewed, a Skill can preserve useful instructions, conventions, and checks for recurring work.
Connectors
A connector matters when the reviewed document workflow needs a supported external capability. It must be configured, trusted where required, enabled, and permitted for the action involved.
What OCR can read, where it can make mistakes, and how to use the result.
OCR extracts the words; document questions help you use them. For example, extract an invoice, then ask when payment is due.
Yes. Name the fields you want, such as date, supplier, total, or clause number, and have missing fields marked rather than guessed.
It depends on the original. Blur, handwriting, low contrast, or a cropped page can change the result. Verify consequential details against the image.
Yes. Copy the checked text or supply the document and ask the agent to turn the findings into a report.
Yes. Save which fields to extract, how to label missing data, and how to format the result.
Redact unneeded details and check the selected provider before sending a document that contains private information. See Security & Controls.
You can try Think-Act on 64-bit Windows 10 or 11. The Mac release is planned for Q1 2027; see Downloads for availability. See downloads page.
Choose a receipt and ask for the vendor, total, and date. You can check the answer at a glance.