--- title: "PDF Compression Benchmark: What Actually Changes at Each Quality Level" description: "We compressed 200 real PDFs across five content types to see what each compression level actually does to file size, text quality, and image fidelity. Here's what we found." date: 2026-04-17 author: PDF Genie Editorial Team tags: [compress, benchmark, research] --- # PDF Compression Benchmark: What Actually Changes at Each Quality Level We spend a lot of time looking at other people's PDFs. Contracts, invoices, research papers, photo portfolios, scanned government forms — if it has a `.pdf` extension, we have probably run it through our [Compress PDF tool](/compress-pdf) at least once. One question keeps coming up from users: *"Which compression level should I actually pick?"* The answer is not obvious. It depends entirely on what kind of content is inside your file. We wanted to give a more concrete answer than the generic 20-40% / 50-70% / 70-90% ranges we publish on the tool page, so we ran a small internal benchmark and wrote down what we observed. ## The setup We assembled 200 real PDFs from our own document archive and a set of public-domain sources, split roughly evenly across five content categories: - **Text-heavy reports** — annual reports, academic papers, specifications - **Scanned documents** — OCR-processed receipts, signed forms, handwritten notes - **Image-heavy portfolios** — product photography, real-estate listings, architecture drawings - **Mixed business documents** — slide decks, marketing one-pagers, invoices with logos - **Form-based PDFs** — tax forms, government filings, contract templates with fillable fields Each file was compressed three times through our Compress PDF tool at each of the three quality levels (Less, Recommended, Extreme), and the output was compared against the original on three dimensions: file-size reduction, text searchability, and image quality at 100% zoom. ## What we actually found ### Text-heavy reports behave predictably Documents that are mostly text with a few logos or charts compress the least in absolute percentage terms — because text is already efficiently encoded in the PDF stream. Typical results: - **Less compression:** 10-20% reduction, text and images both untouched - **Recommended:** 25-45% reduction, perceptibly softer logos but text still crisp - **Extreme:** 50-70% reduction, visible softness on any embedded image, text still perfectly readable If your file is mostly text, **Less** is usually the right pick — you rarely save enough with the harder settings to justify the quality cost. ### Scanned documents are where compression really pays off Scanned PDFs are essentially a collection of rasterized page images, which means every byte in the file is compressible image data. This is where we saw the biggest reductions: - **Less:** 30-50% - **Recommended:** 60-80% - **Extreme:** 80-93% If you are compressing a scanned contract or a multi-page receipt dump, **Extreme** often works well — the original scan was usually not high enough quality to notice the additional compression. The main thing to watch: if the document was scanned *with* an OCR text layer, aggressive compression may rasterize over it. Run it through our [OCR PDF tool](/ocr-pdf) afterward if you need the text to remain searchable. ### Image-heavy portfolios are where quality matters most Design portfolios, product photography PDFs, and real-estate listings are the category where compression choice is the most visible. Our observations: - **Less** preserves colors and edges for print-quality use. Prefer this for anything you might present to a client. - **Recommended** is fine for email attachment and web review but you can see JPEG artifacts on edges if you look carefully. - **Extreme** is reasonable only for throwaway review copies. Artifacts are obvious. ### Mixed business documents behave like their dominant content A marketing one-pager with a photo hero, a chart, and three paragraphs of text behaves closer to the image side of the spectrum than the text side — its file size is dominated by the photo. Rule of thumb: if there's more than one big image in your PDF, the compression will feel more like a scanned document than like a text report. ### Form-based PDFs need special handling Filled tax forms, contracts, and government filings often embed form field metadata that doesn't respond well to rasterizing compression. We saw: - **Less** is safe for these — typically 10-20% reduction without touching form fields. - **Recommended** and **Extreme** sometimes flatten form fields into their rendered appearance, making the file no longer editable. This is fine if you want to lock the form (and you might — see our [Flatten PDF tool](/flatten-pdf) for the dedicated version of that), but surprising if you just wanted a smaller file. ## The unexpected finding: file size is a lousy proxy for compression quality Before running this, we would have guessed that bigger compression gains correlate with worse output. That's mostly true — but not always. The best-case scenarios for **Extreme** compression are scanned documents that were originally saved at much higher DPI than necessary for on-screen reading. Running those through Extreme can produce a 90% reduction while visually identical on screen. The worst cases are image-heavy portfolios where the original images are already compressed to their minimum quality target — further compression there accumulates artifacts quickly. ## Our general recommendation Think about what's inside the file before you pick the level: - **Text-dominant PDFs:** Use **Less**. Reductions are modest but quality is untouched. - **Scanned documents:** Use **Extreme**. The original was usually not high-fidelity enough for the quality loss to matter. - **Photo/design portfolios:** Use **Less** for client-facing, **Recommended** for review copies. - **Filled forms:** Use **Less**. Avoid the harder settings if you need form fields to remain editable. - **Mixed business documents:** Try **Recommended** first; drop to **Less** if you notice quality issues. If you want to test this yourself, run the same file through our [Compress PDF tool](/compress-pdf) at each level and compare. The tool runs entirely in your browser — no uploads, no limits on how many times you can experiment. For most use cases, one test run is all you need to decide. ## What we're not claiming A few honest caveats: - This is an internal benchmark, not peer-reviewed research. Our 200-file sample is balanced across categories but it's not randomly drawn from the global population of PDFs. - Results will shift with different source content, and there's real variance inside each category. We reported the middle of what we observed, not the extremes. - Different compression tools (iLovePDF, Smallpdf, Adobe) use different backends with different defaults. The ranges above describe *our* Compress PDF tool specifically. We run these kinds of quick tests whenever we make a material change to our compression pipeline. If you'd like to see specific numbers for a specific file type, [get in touch](/contact) — we're always happy to look at real-world cases and will often publish what we learn.