For an ordinary, unsigned PDF that you are allowed to upload, start with moderate compression and inspect the result before trying a stronger setting. For a confidential file, use a permitted local application and give the AI only a description of the problem. If you already know which setting you need, you can go straight to the compressor.
Choose a route for your kind of PDF
Try selecting a sentence and searching for a word you can see. Repeat this on a few different pages. A PDF can contain selectable text on one page and an image-only scan on the next. Some scans have a searchable text layer added by optical character recognition (OCR). Their visible lettering is still part of the image, so check the page itself.
| Your file | Where to start | What deserves extra checking |
|---|---|---|
| Report, résumé, or letter exported from an office app | A normal compression preset; re-export from the source if the result barely shrinks | Searchable text, fonts, hyperlinks, and page breaks |
| Scanned paperwork or phone photographs of documents | Adjust image resolution cautiously; use grayscale only when color carries no necessary information | Small lettering, dates, amounts, faint stamps, and cropped edges |
| Brochure, portfolio, or presentation with photographs | Reduce image resolution or JPEG quality while preserving text and vector artwork | Product details, gradients, chart labels, and intended print quality |
| A mixture of text pages, photos, and scans | A moderate preset, then inspect the weakest scan before accepting any stronger version | Each page type separately; a clean cover says little about the scanned appendix |
| A file containing private or confidential information | A local tool you are allowed to use for that file | Where processing occurs; an online compressor still receives the file even if AI never sees it |
| A fillable, digitally signed, archival, or accessibility-dependent PDF | Check its special requirements before choosing a tool | Whether fields still work, signatures remain valid, and any required PDF standard or reading order survives |
| A PDF already close to the required size | Try modest structural optimization before sacrificing image detail | Whether the size change is worth any visible degradation |
Image-heavy PDFs usually have more room for reduction than text-only files. Adobe's Audit space usage report can show whether images, fonts, or other elements occupy most of the file. If images are a small part of the total, repeatedly lowering image quality is unlikely to solve the main problem.
Set an exact size target and keep the original
Use the destination's stated limit. A decimal 2 MB is 2,000,000 bytes; 2 MiB is 2,097,152 bytes. A rounded folder label such as "2 MB" can hide a file that is slightly too large. If a portal gives only an ambiguous 2 MB label, aim below 2,000,000 bytes and test the actual upload. For this guide's practice job, the working target is 1,800,000 bytes. That buffer is a practical choice, not a requirement for every portal.
Make a copy named something like report-compressed-v1.pdf. Keep report-original.pdf unchanged so that every later attempt starts with the best available source. Repeatedly compressing an already degraded copy cannot restore detail.
On Windows, right-click the PDF, open Properties, and read Size, including the byte count. Size on disk measures storage allocation and is not the upload size. On a Mac, select the PDF and choose File > Get Info. If you prefer a command, these return the file length in bytes:
# macOS Terminal
stat -f%z 'report-compressed-v1.pdf'
# Windows PowerShell
(Get-Item -LiteralPath '.\report-compressed-v1.pdf').LengthWrite down the original page count and a few details that must remain clear. For an invoice, that might be the invoice number, total, decimal point, and payment link. For a portfolio, it might be image detail and caption text. This gives you a concrete comparison after compression without turning a simple upload into a large testing project.
Ask AI for a plan when the choice is unclear
You can describe the document without uploading it. Include its size, page count, main content types, destination, and available tool. Those facts are usually more useful for choosing a first attempt than a vague request to "compress this PDF without losing quality."
I need a PDF below 2,000,000 bytes and would prefer below 1,800,000.
It has 18 pages and is currently 8,600,000 bytes.
Most pages contain selectable text. Pages 12 to 18 are scanned receipts.
The smallest important details are totals and reference numbers.
It is for screen reading, not professional printing.
I can use a compressor with Low, Medium, and High presets.
There are no required form fields or digital signatures.
Suggest a first preset and the pages I should inspect afterward.
Explain what to try if the output is too large or the receipts become blurry.
Use only the controls I described. Ask about missing facts that affect the choice.
Do not predict an exact output size or claim that any result has passed.The figures above are an illustrative prompt, not a compression benchmark. Substitute your own facts. If the assistant recommends a control your tool does not have, give it the visible labels and request a revised plan.
If you can share the document with an AI assistant, page thumbnails can help it identify scans, charts, and dense tables. Check small numbers on the full page yourself; thumbnails hide fine detail. For a confidential PDF, describe the page types and settings without sending document images or text.
After a failed attempt, give the assistant the output size and the exact detail that became unreadable. Here is an illustrative follow-up for a mixed document:
Medium produced 2,600,000 bytes. High produced 1,700,000 bytes,
but reference numbers on scanned pages 12 and 15 became unclear.
The other pages still look good. I have the original Word document
with the scans and photographs inserted separately.
How can I reduce the photographs while preserving those scans?
Give me a next attempt I can make from the original.This points the assistant toward a useful change in the source file. If you have only the PDF, say so; a tool with separate image controls may help, but you still have to inspect the result. The assistant cannot know whether the file passed from your description alone.
Make a first attempt with an online compressor
For a file you are allowed to send to a cloud service, Adobe Acrobat's online compressor provides Low, Medium, and High compression. These are compression strengths: Low generally prioritizes image quality, while High prioritizes a smaller file. Basic use is limited; do not assume that an unlimited series of attempts or batch processing is free.
- Select the original PDF in the compressor.
- Start with Medium for an ordinary mixed document, or Low if fine image detail matters and the size reduction needed is small.
- Read any file-specific estimates. If the first estimate is far above the limit, decide whether trying stronger compression is appropriate for your pages.
- Compress and download a new copy.
- Measure the downloaded file and compare its most demanding pages with the original.
- If it is still too large, make the next attempt from the original. If it is already small enough and readable, stop reducing its quality.
Measure the download rather than relying on an estimate. In the dated example below, Adobe displayed 4.0 MB before compression, but the downloaded file was 179,153 bytes. Your file may behave differently.
On a phone or tablet, download the compressed PDF, then open it in your Files or Downloads app. Check that copy before selecting it in the upload form. A small thumbnail can hide damage; use a larger screen if you cannot comfortably read dense tables on the phone.
Adjust resolution and image quality separately
Two controls often determine the tradeoff. Downsampling reduces how many pixels an image contains at its displayed size. A quality control changes how the image is encoded. Reducing either can make a file smaller, but they damage detail in different ways. If your tool exposes both, change one at a time so that you can tell which change caused the problem.
The PDF24 online compressor exposes DPI and image-quality controls, plus options that can remove or alter PDF features. It processes uploads on its servers. For an ordinary document, begin with its current defaults, then adjust only the image control you need. Leave options that flatten forms or strip annotations, attachments, outlines, output intents, or structural information off when those elements are required. They can affect much more than the appearance of a photograph.
For scanned text and phone photographs
In a scan, the letters are part of the image. A large heading may survive while a small account number becomes ambiguous. Inspect the smallest necessary lettering first, including punctuation, minus signs, and the distinction between similar digits.
For a screen-reading trial with adjustable resolution, you can compare a copy at 200 DPI with one at 150 DPI, keeping image quality otherwise unchanged. These are starting experiments, not guaranteed settings. Dense receipts, small type, and faint stamps may need more resolution. A scan with a high nominal DPI can still be poor if the source photograph is out of focus.
At the same physical page size, changing an image from 300 to 150 pixels per inch halves each dimension and leaves one quarter as many pixels. The PDF will not necessarily become one quarter of its size because encoding, text, and other objects still contribute bytes. This also explains why a large reduction in resolution can erase tiny strokes.
Try grayscale only if color is irrelevant to the document's purpose. Keep color where it distinguishes a stamp, highlighted instruction, chart series, or photographic evidence. Pure black-and-white conversion can erase pale writing or shaded information, so compare those areas before accepting it.
If the original is blurred, tilted, or badly lit, a better scan may help more than another compression preset. Capture the page evenly, keep its edges visible, and check the smallest text before creating the PDF. Avoid asking generative AI to reconstruct a number or letter that you cannot verify against the source.
For photographs, charts, and digital text
If your tool has a percentage quality scale, an illustrative comparison is 80% versus 70% at the same resolution. Read those percentages as that tool's settings, not as a measurement of how much quality remains. Scales are not interchangeable across compressors. Keep the higher-quality result if both fit.
Inspect faces, product labels, fine textures, and smooth gradients. Block patterns, smeared edges, and visible bands are signs that the tradeoff has become objectionable for the intended use. A casual screen handout and a print portfolio can have different acceptance criteria.
Keep digital text and charts as text and vector artwork when exporting. Turning every page into a picture makes small lettering dependent on image resolution and can remove search and link behavior. For artwork, Adobe's PDF optimizer settings distinguish JPEG-style compression for gradual color transitions from ZIP compression for solid-color illustrations. The same documentation explains that removing embedded fonts can trigger font substitution on other computers.
For a mixed PDF, inspect the scan before choosing a setting based on how good the digital pages look. If a global preset damages two scans while the photographs remain excellent, return to the source document and reduce its oversized photographs separately. Re-export, then recheck the complete file. This gives you more control than repeatedly applying stronger compression to every page.
Use local compression when uploads are unsuitable
Windows: PDF24 Creator
PDF24 Creator is a free Windows application with offline PDF compression. Its desktop processing keeps files on your PC; the similarly named online tools use a different processing route.
- Open PDF24 Creator and choose its PDF compression tool.
- Add a copy of your original PDF.
- Use the current default settings for the first attempt. Where the tool exposes resolution and quality, record the values you used.
- Save the result under a new name, measure it, and inspect the pages with the smallest text or most detailed images.
If the result is still too large, adjust one setting and process the original again. Follow the same scan and photograph checks used for an online result. A local tool addresses where processing happens; it does not make an aggressive image setting harmless.
Mac: Preview
Apple documents a built-in route in Preview: open the PDF, choose File > Export, select Reduce File Size in the Quartz Filter menu, and save a new copy. Use Export, then check the file on disk. Apple warns that the compressed result may be lower quality.
This is a convenient first attempt for an ordinary PDF. If the filter blurs a scan or small image labels, reject that copy and return to the original. Preview's filter is less useful when you need to tune different page types separately. Use an application with adjustable image settings or improve the source export.
Acrobat Pro: inspect what occupies the space
For more control, Adobe's advanced optimization route is Menu > Save as Other > Optimized PDF on Windows or File > Save as Other > Optimized PDF on Mac. Start with Audit space usage before choosing changes.
If images dominate, work in the Images settings. If fonts dominate, review embedding or subsetting with care. Keep the fonts needed for the recipient to see the document correctly. Review each removal setting separately: options in the Discard panels can remove document data or flatten fields. Save a new copy and test the features you need.
For a print deliverable, keep the printer's required resolution, color, and PDF standard. A screen-oriented preset may fit the upload limit while being unsuitable for production. If those requirements and the size cap conflict, ask for another transfer method.
What the ten-page practice example shows
The August 25, 2026 demonstration used a synthetic application pack. It contains a 7.2-point table, photographs, fine chart labels, an image-only scan, three PDF hyperlinks, and a QR code. You can download the original PDF and the compressed practice copy to inspect the tradeoff yourself.
| File or check | August 2026 result | What it tells you |
|---|---|---|
| Original PDF | 24,488,944 bytes; 10 pages | The starting file is available for comparison |
| Downloaded PDF using High | 179,153 bytes; 99.27% smaller | This file met the 1,800,000-byte target; that reduction is not a promise for other files |
| Recorded checks | 10 pages, 20 searchable text items, 3 links, the expected QR destination, and 8 expected text items found by OCR on the scan | These checks cover selected details, not every visible character |
The compressed scan is visibly softer. It was accepted for this practice task, but choose a clearer copy if it fits your own limit. Digital text staying sharp does not tell you whether a scanned receipt is still readable.
The test record shows a limitation of OCR checks: all eight expected text items were found, yet one occurrence of REF-8042 was read as REF-6042. Another occurrence was recognized correctly. Check each important number on the page to distinguish an OCR mistake from damage to the visible text.
The screenshots document this dated example. Current controls and results may differ.
Check the exact file you will upload
Open the original and compressed copy side by side at the same zoom. Begin at a normal reading size, then enlarge the smallest important details. Compare the same region on both pages. Matching zoom prevents a fit-to-width view from making one version appear artificially sharper.
- Confirm the file length is below the destination limit and your chosen buffer.
- Check that all required pages are present, in order, with the right orientation. Inspect margins for missing content.
- Read the smallest necessary text in the compressed copy. Check names, reference numbers, dates, totals, decimal points, and negations against the original.
- Search for text that was searchable before. For an image-only scan, read the image; a failed search does not by itself indicate compression damage.
- Check the actual destinations of required hyperlinks. Visible blue or underlined text does not prove that a clickable link survived.
- If a QR code is required, scan it and compare the decoded value with the original. For a URL, include the path and query parameters. Check a non-URL code against its expected text or identifier.
- For important submissions, open the compressed copy in a second viewer and check any required form or accessibility behavior.
Inspect every page of a short document. For a long report, review page order and look through all thumbnails for major changes, then examine each page type at full size. Sampling cannot establish that every number is correct. If the consequences of a missed detail are serious, check every relevant page.
In a scan with an OCR layer, compare both the visible image and the searchable text when downstream extraction matters. The picture can be readable while the OCR text contains a wrong amount, and an OCR match can coexist with a blurred image. Search, visual inspection, link checks, and file-size checks answer different questions.
Once a copy is small enough and survives these checks, give it a clear final filename and upload it. If the portal supplies a preview or downloadable submitted version, check that too, especially if it processes documents after upload. Keep the original until the recipient has accepted the submission.
When the result is still too large or loses detail
| Problem | Check first | Next useful attempt |
|---|---|---|
| A mostly text PDF barely shrinks | Whether images actually occupy much space | Audit fonts and embedded objects, try structural optimization, or export a fresh PDF from the original office file |
| A scan fits the limit but its numbers are unclear | The same numbers at matching zoom in the original | Use more resolution or less aggressive encoding; improve a poor original scan |
| Photos look bad while text is clear | Block artifacts, smeared edges, and gradients | Raise image quality, then reduce oversized source images more selectively |
| The output becomes larger | Whether OCR, format conversion, or a PDF rewrite changed the file | Keep the smaller valid original; try a different method with fewer changes |
| The file appears to be 2 MB but the portal rejects it | Exact bytes, file type, password protection, and any other portal requirements | Try below 2,000,000 bytes with a buffer; verify that you downloaded a PDF rather than a web page |
| A QR code no longer scans | Whether the original code scans and whether its border was cropped | Use less image reduction and preserve its clear border; verify the decoded value again |
| Links or fields disappear | Flattening, rasterizing, or removal settings | Return to the original and disable the option that removed them, or use another tool |
| One viewer renders blank or broken pages | The same copy in another viewer and the original | Discard that copy and re-export or recompress from a clean source |
| High compression still exceeds the limit | Oversized source images, required page count, and destination alternatives | Rebuild from the source, use selective optimization, or request a permitted split or larger upload |
| AI recommends repeated stronger compression | Whether it knows which pages already failed | Report the exact failed details and ask for a method that preserves them; use another submission route if no readable copy fits |
OCR is for recognizing text and making scans searchable. It is not automatically a size-reduction step. OCRmyPDF's documentation explains that its output can grow because recognition adds information and the new representation may be less compact. If searchability is required, budget for it; if it is unnecessary, do not add OCR just because compression failed.
If the destination allows separate files, split at sensible document boundaries and confirm the rules apply to each file, not the combined submission. Some portals require one complete PDF or impose a total attachment limit. Do not remove required pages to make a file fit.
If you have the Word source, reduce selected images before exporting
This helps when photographs consume space but nearby scanned receipts must stay clear. Work on a copy of the editable document. For Word desktop on Windows:
- Select an oversized photograph, then choose Picture Format > Compress Pictures. If the tab is missing, select the picture first; a narrow window may show only the button's icon.
- Keep Apply only to this picture checked so that you do not also reduce the scans. Choose a resolution suited to the intended reading or printing use, then select OK. Repeat for other photographs that can tolerate less detail.
- For a saved document, choose File > Save a Copy, then Browse and PDF from the file-type list. Give the PDF a new name and save it.
- Open that PDF and check its bytes, page breaks, links, and the scans you preserved. If it still exceeds the limit, use it as the starting point for a moderate compression attempt.
Microsoft documents selected-picture compression and PDF export. The picture controls described here are desktop features. Preserve needed bookmarks and accessibility tags in the PDF export options. If you have only the PDF, skip this route.
Handle forms, signatures, and special PDF requirements
A file can look correct and still lose the behavior the recipient needs. Decide which of these requirements applies before using a general preset:
| Requirement | What to preserve or verify |
|---|---|
| Fillable fields | Test typing, checkboxes, saving, and reopening in the intended viewer. Avoid flattening while the recipient still needs to enter data. |
| Digital signature | Follow the recipient's signing instructions. A visible signature image does not establish cryptographic validity. Keep the signed original. |
| Password or editing restrictions | Use the owner's authorized method and credentials. If a portal rejects protected files, obtain a permitted submission copy. |
| PDF/A or another required PDF standard | Validate the resulting file against the required standard. A filename or unchanged appearance does not prove conformance. |
| Accessibility | Check required tags, reading order, text alternatives, and keyboard access. Avoid converting all pages to images or stripping structural information. |
| Embedded attachments, comments, or bookmarks | Keep the elements the recipient uses. Treat their removal as a deliberate content decision. |
Adobe documents editing restrictions for digitally signed PDFs. If the recipient needs a revised file, request an unsigned source and compress it before signing. Rewriting a signed document can affect its signature; check signature validation in the recipient's intended application.
Compressing an image of a handwritten signature raises a different issue: it must remain legible enough for the stated purpose. It does not become a certificate-based digital signature because it is visible in a PDF.
Optional: structural optimization and batch work
For someone comfortable with the command line, qpdf can reorganize a PDF and compress its streams without downsampling images. With qpdf installed, this is a structural optimization attempt for an ordinary unsigned PDF:
qpdf --object-streams=generate --recompress-flate --compression-level=9 input.pdf output.pdf
qpdf --check output.pdfThis command does not request JPEG image optimization. It may help a text-heavy file or inefficient PDF structure, but it cannot promise a large reduction for a scan dominated by already compressed images. Object streams require a viewer that supports PDF 1.5 or later. Check warnings and test the output in the destination viewer. A successful structure check does not prove that small text, links, signatures, or accessibility requirements are correct.
For a recurring batch, group files by content type and required features before choosing settings. Keep originals in a separate folder, write outputs to a new folder, and record input and output bytes. Reject files over the limit and flag feature or readability failures for review. Do not overwrite the whole batch with the strongest preset.
A SHA-256 record identifies the exact copy you reviewed or handed over. It can help track batches and prevent file mix-ups, but it does not measure quality. The original review kit contains the demonstration's scripts and evidence if you want to examine its more extensive checks.
Other size and submission questions
Can I get the file under 500 KB?
Change the byte target, then follow the same process. A one-page text PDF and a long scan have very different prospects. The 179,153-byte practice result does not establish that your file can reach 500 KB. If readable required content will not fit, request another submission route instead of accepting unclear text.
Will ZIP or cropping solve the size problem?
A ZIP file changes the upload format. It does not make the PDF itself smaller, and it is unsuitable when the portal requires a PDF. Adobe explains that its PDF crop operation hides content without removing the underlying data or reducing file size. For a photographed document with large unnecessary surroundings, crop the source image before rebuilding the PDF, while preserving all required page content.
Should I use Print to PDF or screenshots?
Use a PDF optimization or export route when you need to preserve document features. Printing creates a new representation, and taking screenshots makes page content image-based. Either route can change links, forms, tags, or other required behavior. If you deliberately use one for a static reading copy, test that copy against its actual requirements.
Can compression guarantee no quality loss?
Lossless stream compression and lossy image processing are different operations. If images must retain their exact pixels, use a method that preserves them and accept that the reduction may be modest. If some image degradation is acceptable, choose the clearest copy that fits. Check the file itself before accepting an AI recommendation or a "high quality" preset label.
Use the practice materials with your own checks
The printable practice checklist and practice prompts and checklist record the demonstration's more extensive tests. They include a deliberately damaged PDF for testing the checks themselves; creating one is unnecessary for an ordinary upload. Use these materials to reproduce the example, replacing its page counts, selected text, and size thresholds with your own.
