A client sends a certificate photo on WhatsApp at the end of the day, and the agency still has to decide whether it can quote, translate, and review the document from that image. The risky part isn’t opening the attachment. It’s treating a phone photo as if it were a clean, complete source file.
A reliable intake process checks what the image actually shows before anyone promises a delivery format or starts translating. The team confirms that every page is present, tests whether the text is legible, chooses OCR or human transcription, and checks extracted names and numbers against the image.
This guide sets out that workflow for agencies. It covers the practical choices between an in-house process, a freelancer, and an online document tool, plus the checks that stop a readable-looking photo from turning into an inaccurate translation.
The WhatsApp attachment is an intake decision, not a translation-ready file¶
A WhatsApp photo is an image supplied through a messaging channel; it isn’t automatically a complete, legible, or translation-ready document. The first decision is whether the agency has enough source material to assess the job, not which translation engine to run.
Clients often send the image they have to hand. The photo may show a single page from a multi-page document, a screen reflection, a folded corner, or small print. A screenshot of a photo may add another layer of compression or cropping. The agency doesn’t need to assume any of those problems is present, but it does need to check for them.
Treat intake as a short triage stage with a clear outcome:
- Ready for OCR review: the full page is visible, the image is sharp, the text is legible, and the layout is understandable.
- Needs a better image: blur, glare, shadows, cropping, or low contrast make important content uncertain.
- Needs a human route: the image is handwritten, damaged, unusually complex, or too poor to transcribe safely from the file.
- Needs clarification: the client hasn’t sent every page, or the purpose and requested deliverable remain unclear.
A photo can look readable on a phone screen and still be difficult to inspect at working size. Ask the intake coordinator to open the original attachment rather than judging it from a thumbnail in a chat window. Zoom into the smallest text, seals, names, dates, and any handwritten additions.
Google’s Drive OCR guidance lists sharpness, even lighting, and clear contrast as conditions for better results, and recommends text at least 10 pixels high; these criteria come from the help page accessed on September 23, 2026 (Google Drive OCR guidance). The figures are useful as a practical screen, not as a promise that a particular image will produce correct text.
The intake response should make the next step obvious. If the attachment is usable, confirm what pages and language the agency sees. If it isn’t, name the problem and ask for a replacement rather than sending a vague “please resend.” If the job needs a human transcription route, explain that the image needs to be checked before the agency can confirm scope.
A useful reply might say: “The bottom line on page two is blurred, so we can’t verify the reference number. Please send a new photo or scan with the full page in focus.” That message tells the client what is wrong and what to do next without suggesting the agency has already accepted the text as accurate.
Build a quality gate before you choose OCR¶
OCR, or optical character recognition, converts text in an image into editable text. OCR can speed up handling of a clear scan, but it’s an extraction step, not a quality guarantee and not a replacement for checking the source.
Google Drive supports OCR for image files in JPEG, PNG, and GIF formats, as well as multipage PDFs, according to its help page accessed on September 23, 2026 (Google Drive OCR guidance). The same guidance recommends keeping a file at 2 MB or less and says text should be at least 10 pixels high. Those details describe Google’s Drive workflow; they aren’t universal limits for every OCR tool.
Use a consistent visual checklist before sending any photo into OCR:
| Check | What the intake coordinator looks for | If the check fails |
|---|---|---|
| Completeness | The full page, edges, and all required pages are present. | Ask for the missing page or a new full-page image. |
| Focus | Small text and punctuation look sharp when enlarged. | Request a retake or choose human transcription if a retake isn’t possible. |
| Lighting | Text isn’t obscured by shadows, glare, or reflection. | Ask the client to change the lighting and retake the image. |
| Contrast | The text stands out clearly from the paper or background. | Ask for a cleaner image; don’t rely on OCR to resolve unclear characters. |
| Orientation | The document reads upright rather than sideways or upside down. | Rotate the image before OCR or request an upright version. |
| Text size | The smallest print remains readable in the image. | Request a higher-quality image or escalate to a human check. |
| Structure | Headings, columns, tables, and labels can be related to their values. | Preserve a visual reference and review layout manually. |
| File format and size | The file can be accepted by the chosen OCR workflow. | Convert or rescan through an approved route. |
Google says Drive OCR works best with sharp images, even lighting, and clear contrast, and advises rotating an incorrectly oriented image before upload; that guidance was accessed on September 23, 2026 (Google Drive OCR guidance). Drive also detects the document language during its image-to-text workflow, according to the same dated help page. Automatic language detection can help with extraction, but the agency should still identify the source language and script before assigning the job.
The core policy is simple: don’t ask OCR to decide what an unclear character says. A faint 1 could be confused with a capital I, for example, but the right response is to compare the image and extracted text, not to guess which one the client intended. If the image doesn’t support a confident reading, flag the uncertainty and ask for a clearer source.
A historical study can help explain why the image gate matters, but it shouldn’t be misused as a present-day product benchmark. A NIST experiment published on December 1, 1997, tested three OCR products against pages selected from the 1994 Federal Register and reported character-recognition error rates from 1% to 74% as print and image quality declined (NIST study, published December 1, 1997). Those results don’t estimate error rates for today’s client photos or modern OCR engines. They do show why image degradation needs to be treated as an operational risk rather than a cosmetic issue.
Use the checklist to decide whether a photo is ready for OCR, not to promise a specific accuracy score. If the page passes basic checks, OCR can produce a working text layer for translation and review. If the page fails, the next step is a better source image or a human route.
Choose the right route: agency team, freelancer, or document tool¶
No single route fits every WhatsApp attachment. An agency should choose based on image quality, document structure, client confidentiality requirements, review needs, and whether the final output needs a formal human service.
| Route | Useful when | Main trade-off |
|---|---|---|
| In-house agency workflow | The team needs control over intake, escalation, review, and client communication. | Staff need a repeatable process and a clear way to manage transcription or OCR exceptions. |
| Freelance translator | A specialist can handle the language, document type, and manual review required. | The agency still needs to brief the freelancer, transfer files appropriately, and check the returned work. |
| Online document tool | The image is legible enough for OCR-assisted processing and the team can review the output. | A tool can’t resolve unclear source text or replace a human decision about uncertain content. |
An in-house workflow is valuable when the agency needs a single intake standard across different translators. The coordinator can check the attachment, note missing pages, ask for a retake, and send only readable material to the assigned linguist. A written handoff also prevents the translator from receiving an image without context about the intended deliverable.
A freelancer may be the right choice when the document needs language or subject-matter expertise, or when the agency’s own team can’t read the source script. The project manager should still send the original image alongside any OCR text. A transcription file without the image removes the translator’s ability to verify punctuation, labels, and unclear characters.
An online document tool can fit a clear, scan-only job where the agency wants a formatted draft to review. In ChatsControl, a typical workflow is to upload a photo or scanned PDF, let OCR extract the text, and receive a translated document with its layout rebuilt. The tool also provides a bilingual side-by-side review view, so a linguist can compare the source and translation while checking the draft. ChatsControl is a document-translation tool rather than a full CAT tool or TMS, and it isn’t built for handwritten or very poor-quality scans. An agency should keep human review in the workflow and avoid using OCR as a substitute for an unreadable original.
The practical question isn’t whether a tool or a person is “better.” The question is whether the source image supports the route. Clear printed text may be suitable for OCR-assisted processing. Unclear text needs clarification or human attention. A complex form needs layout review even when the words are easy to extract.
This is also where the agency can set expectations with the client. Explain whether the quote assumes a readable source, whether additional transcription may be needed, and whether the requested output is a translation draft or a formally certified document. For work that needs formal certification, the human service route is separate from an automated document-translation workflow; see the agency’s certified translation option when that requirement applies.
Preserve context: OCR text can lose document structure¶
OCR often gives an agency a useful text layer, but a list of extracted words isn’t the same thing as a faithful representation of the original page. Layout carries meaning, especially when a label, value, column, or handwritten note belongs to a particular field.
Google says bold, italics, font size, font type, and line breaks are likely to be retained when an image is converted to a Google Doc. The same guidance says lists, tables, columns, footnotes, and endnotes are not likely to be detected by that conversion; the information was on the Google Drive help page accessed on September 23, 2026 (Google Drive OCR guidance). Agencies should therefore verify the relationships between parts of the page instead of assuming that OCR preserves structure.
A two-column document shows why. The extracted text may put every word into a single sequence, leaving a translator unsure which value belongs to which heading. A form can have a similar problem when several entries sit beside different field labels. Even if every character has been recognized, the meaning can be wrong if the associations are lost.
Keep the image and OCR text together through review. The reviewer should be able to see the source page while reading the extraction, and should flag places where the order or grouping is uncertain. Don’t remove seals, stamps, marginal notes, or handwritten additions from the review just because they aren’t part of the main printed text.
A useful handoff includes:
- The original image or PDF, not only the extracted text.
- A page-by-page note showing whether the source appears complete.
- An indication of the source language, if known.
- A list of unreadable or uncertain areas.
- A note about tables, columns, labels, or other structure that may need manual reconstruction.
- The required output format and review level.
For a clean scan, Google Drive’s mobile app can scan paper documents and save them as searchable PDFs; scanning is available in the Android and iOS apps, not the Drive web version, according to the help page accessed on September 23, 2026 (Google Drive scanning guidance). A searchable PDF can be easier to handle than a series of camera photos, but scanning won’t make unreadable text reliable by itself.
The biggest pitfall is confusing appearance with structure. A translated page may look tidy while placing a date beside the wrong label. A reviewer should check how each item relates to its headings and neighboring fields, not just whether the translated text reads smoothly.
Verify names, dates, amounts, and uncertain characters¶
OCR review should focus first on details that can change the identity or meaning of a document. Names, dates, amounts, reference numbers, addresses, and identifiers deserve direct comparison against the original image because a small recognition error can affect the translation’s usefulness.
A translator shouldn’t review only the final translated prose. Check the extracted source text against the image first, then check that the translation carries the verified details across accurately. That order separates two different risks: OCR may misread the source, and translation may alter or omit a correctly recognized detail.
A practical review pass can follow this sequence:
- Compare every name against the image, including spelling, order, and diacritics.
- Verify dates, amounts, reference numbers, and other numeric strings character by character.
- Check punctuation and symbols that may change a value or identifier.
- Confirm that each label still belongs to the correct value in a form, table, or column.
- Mark any source content that remains unclear instead of silently choosing a likely reading.
- Compare the translated version against the verified source text and confirm that critical details remain consistent.
An agency can add a second QA step when the first pass produces a draft. ChatsControl includes a separate QA validator that checks translation issues such as numbers, names, terminology, and omissions. That feature can help a reviewer find items to inspect, but the original image remains the authority when OCR has misread a character. A validator can’t recover information that isn’t visible in the source.
Names need particular care when a client sends several document photos together. A name may appear in different forms across documents, or a handwritten correction may conflict with the printed version. The agency should record what is visible and ask the client or document owner to resolve a genuine ambiguity rather than standardizing it by guesswork.
Numbers also need context. A date can use different ordering conventions, a decimal marker can be hard to see, and a number in a table may belong to a neighboring row. The reviewer should check the full line or field, not just the isolated characters. That approach is especially important when OCR has flattened columns or separated a label from its value.
A document can be legible overall while one critical region is not. In that situation, don’t describe the whole file as “unreadable” if only a small area needs attention. Point to the specific location, ask for a close-up or replacement page, and keep the uncertain segment visibly flagged in the working file.
The reviewer should also record which source was used. If the client replaces a photo after the agency has already extracted text, the team needs to know whether the OCR and translation draft refer to the old image or the new one. Clear file naming and version notes help prevent a polished translation from being checked against the wrong attachment.
Turn the intake checks into a repeatable agency workflow¶
A good workflow gives the project manager a predictable way to handle both easy photos and difficult exceptions. The process shouldn’t depend on one coordinator remembering which images to inspect or one translator guessing what the client meant.
Use these steps as a practical intake sequence:
- Save and identify the attachment. Follow the agency’s approved file-handling process, give the file a job reference, and keep the original image available for review.
- Confirm scope. Ask what the client needs translated, whether all pages have been sent, and what output they expect.
- Inspect the image. Check completeness, focus, lighting, contrast, orientation, text size, and layout.
- Choose the route. Send a usable printed image to OCR-assisted processing, ask for a replacement when the source can be improved, or escalate to human transcription and review.
- Keep the source with the extracted text. Make sure the assigned translator can see the image while checking OCR output.
- Review critical details. Verify names, dates, numbers, identifiers, and table or form associations against the image.
- Resolve uncertainty. Ask the client for clarification when a character or field can’t be read confidently.
- Confirm the deliverable. Tell the client what the agency can produce from the material received, and note any limitation caused by the source.
That workflow also gives the agency a consistent way to respond to incomplete submissions. If the attachment contains only one page, the coordinator can pause before quoting a multi-page job. If the client sends an image with a cut-off edge, the team can request the missing area before translation starts. If the document is clear but structurally complex, the coordinator can flag the need for layout review.
A short message template helps keep the tone direct:
Thanks for sending the document. The text at [specific location] isn’t clear enough to verify, and we don’t want to guess. Please resend that page in focus with the full page visible, or let us know if a clearer original isn’t available.
Replace the bracketed text with a concrete description, such as “the date near the bottom of the page” or “the right-hand column.” Don’t tell the client merely to “improve the quality.” A specific request is easier to act on and creates a clear record of why the job is waiting.
Agencies can document these steps in their intake SOP, including who can accept a photo, who approves an exception, how uncertainty is recorded, and who checks the final output. For a deeper guide to setting those internal rules, see SOPs for Translation Agencies: How to Document Your Processes. The same process should cover file access and transfer rules, especially when client documents contain personal information; GDPR for Translation Agencies: NDA, DPA and Cloud Tools can help teams think through that side of the workflow.
Common pitfalls and how to handle them¶
Most intake problems don’t come from OCR alone. They come from assumptions made before anyone checks the original image or from a handoff that loses the context a reviewer needs.
Treating OCR output as the source of truth. Extracted text is a working version of the image, not proof that the words have been read correctly. Keep the image beside the text and verify critical details against it.
Accepting a cropped page because the main text looks complete. A cut-off edge can hide a stamp, page number, note, or part of a field. Ask for a full-page replacement when a missing area could affect the task.
Flattening a table into plain text and translating the result without checking associations. Google warns that tables, columns, lists, footnotes, and endnotes are not likely to be detected in Drive’s image-to-Doc conversion, according to its help page accessed on September 23, 2026 (Google Drive OCR guidance). Review the original layout before deciding which value belongs to which label.
Guessing a blurry name or amount from context. A plausible guess can still be wrong. Mark the uncertainty, ask for a clearer source, or use human transcription and review when the client can’t provide a better image.
Forwarding the photo without an intake note. The translator may not know whether pages are missing or which line the client needs. Include the job scope, visible issues, requested output, and any client clarification with the handoff.
Assuming a searchable PDF is a verified document. Searchable means the file has text that can be found or selected; it doesn’t establish that every character matches the image. Compare the text to the page before translation.
Skipping the client follow-up because the deadline is tight. A rushed guess transfers uncertainty into the translation and makes later correction harder. A short, specific request for a better image is often the fastest responsible next step.
A failed OCR result doesn’t mean every document needs manual transcription from the start. First determine whether the client can send a clearer photo or a scan. If not, inspect whether a human can read the original image. If neither OCR nor a person can establish the text with confidence, explain that the agency needs a better source before it can translate the affected content.
For a practical comparison of scan handling and OCR limits, see Translating PDFs and Scans with AI-OCR: What Works and What Doesn’t. The key is to keep the exception visible. Don’t silently remove an unreadable line or deliver a clean-looking file that hides unresolved source text.
FAQ¶
How should a translation agency handle a document photo sent on WhatsApp?¶
Save the image through an approved intake process, check that every page and critical detail is legible, and decide whether OCR is safe to use. Ask for a new scan or arrange human transcription when the image fails that check.
What image quality does OCR need to translate a phone photo accurately?¶
Google recommends sharp images, even lighting, clear contrast, correct orientation, and text at least 10 pixels high for Drive OCR; the help page was accessed on September 23, 2026 (Google Drive OCR guidance). Those are useful triage criteria, not a guarantee that every photo will be recognized correctly.
Should an agency use OCR or manually transcribe a blurry document photo?¶
Return a photo that fails basic legibility checks for rescanning when the client can provide a better image. Use human transcription and review when the document can’t be rescanned or critical details remain unclear.
How can an agency check names, dates, and numbers after OCR?¶
Compare extracted text against the original image, focusing on names, dates, amounts, reference numbers, and symbols. Check tables and labels in context because OCR may not preserve their relationships.
What should a client do if a document photo is too blurry to translate?¶
Ask the client to retake the photo in focus with even lighting, clear contrast, and the page upright. A scan made in the Google Drive mobile app can also be saved as a searchable PDF, according to its help page accessed on September 23, 2026 (Google Drive scanning guidance).
Can an agency translate a photo without sending it to an OCR service?¶
Yes. A translator can transcribe the visible text manually, or the agency can ask the client for a clearer image or original file. The right option depends on legibility, confidentiality rules, and the document’s required review level.
Does OCR preserve tables and document layout?¶
OCR may extract words without preserving how tables, columns, lists, footnotes, or endnotes connect. Review the original layout and verify every label-value relationship before delivering the translation.
Can an agency rely on older OCR research to estimate current accuracy?¶
No. A NIST study published on December 1, 1997, reported error rates from 1% to 74% in a historical experiment using pages from the 1994 Federal Register; the study doesn’t establish the performance of modern OCR products (NIST study). Use it as a reminder that image quality affects recognition, not as a current forecast for a client’s photo.
The agency’s best response to a WhatsApp photo starts with one question: can the team read and verify the source, not just extract its text? A clear intake gate, a human review of critical details, and a direct request when the image falls short let the agency choose the right route without turning uncertainty into an unmarked translation error.