Image SEO Audit Tool - Alt Text, Image Weight & Format Checker

Give it a URL. The page is opened in a real browser, every image it actually renders is collected — <img> tags and CSS background-image — and each one comes back with its weight, format, HTTP status, displayed size, natural size, alt text and the problems it has.

Output is JSON: one record per page, plus one summary record for the whole run.

What a real run returns

https://www.smashingmagazine.com, single page, default settings: 28 images, 12 issues — 9 images displayed far smaller than the file actually served, 3 images with no alt text. Formats found on that page: 10 SVG, 6 PNG, 12 undetermined.

What it checks

Issue code	Raised when
ALT_MISSING	The image has no alt attribute, or an empty one
HEAVY_WARN	The file is over 200 KB and up to 500 KB
HEAVY_CRITICAL	The file is over 500 KB
NON_NEXT_GEN	A JPEG or PNG heavier than 50 KB, where WebP or AVIF would be lighter
UPSCALED	The image is displayed larger than its own resolution, so it looks blurry
DOWNSCALED	The image is displayed at under 25 % of its resolution, so most of the downloaded bytes are wasted

Each family of checks can be switched off: checkAlt, checkWeight (weight and format), checkDimensions.

Input

Field	Type	Default	What it does
startUrl	string	—	Required. The page to analyze, or the page the crawl starts from
crawlUrls	boolean	false	Follow internal links and analyze several pages instead of one
maxPages	integer	10	Upper bound on the number of pages analyzed when crawlUrls is on
checkAlt	boolean	true	Report images with no alt text
checkWeight	boolean	true	Report heavy images and outdated formats
checkDimensions	boolean	true	Report images displayed at the wrong size
timeout	integer (ms)	15000	Time budget for loading a page, and for each image request
userAgent	string	Mozilla/5.0 (compatible; SEO-Image-Analyzer/1.0)	User agent sent to the site

Only links on the same domain are followed, and anchors, mailto: and tel: links are ignored. There is no sitemap reading: the crawl only sees links present on the pages it visits.

Output

One record per page (type: "page"), with its images and its own counters:

{
  "type": "page",
  "pageUrl": "https://www.smashingmagazine.com",
  "title": "Smashing Magazine — For Web Designers And Developers",
  "httpStatus": 200,
  "imagesCount": 28,
  "issuesCount": 12,
  "images": [
    {
      "imageUrl": "https://www.smashingmagazine.com/images/logo.svg",
      "alt": "Smashing Magazine",
      "displayed": { "w": 52.06, "h": 69.42 },
      "natural": { "w": 113, "h": 150 },
      "bytes": 2735,
      "bytesKB": 2.67,
      "format": "svg+xml",
      "mime": "image/svg+xml",
      "httpStatus": 206,
      "cacheControl": "public,max-age=31536000,immutable",
      "issues": ["DOWNSCALED"]
    }
  ],
  "summary": {
    "altMissing": 3,
    "heavyWarn": 0,
    "heavyCritical": 0,
    "nonNextGen": 0,
    "upscaled": 0,
    "downscaled": 9,
    "byFormat": { "svg+xml": 9, "png": 6, "unknown": 11, "svg": 1, "octet-stream": 1 }
  }
}

One summary record for the run (type: "site-summary"), always the last one:

{
  "type": "site-summary",
  "pagesCrawled": 1,
  "pagesTurnedAway": 0,
  "imagesTotal": 28,
  "issuesTotal": 12,
  "topIssues": [
    { "type": "DOWNSCALED", "count": 9 },
    { "type": "ALT_MISSING", "count": 3 }
  ],
  "byFormat": { "svg+xml": 9, "png": 6, "unknown": 11, "svg": 1, "octet-stream": 1 },
  "byProblem": { "ALT_MISSING": 3, "HEAVY_WARN": 0, "HEAVY_CRITICAL": 0, "NON_NEXT_GEN": 0, "UPSCALED": 0, "DOWNSCALED": 9 },
  "byStatus": { "200": 8, "206": 8 },
  "thresholds": { "HEAVY_WARN": 204800, "HEAVY_CRITICAL": 512000, "HEAVY_WARN_KB": 200, "HEAVY_CRITICAL_KB": 500 }
}

thresholds is there so the numbers behind the verdicts are never a black box.

What you are charged for

• Page Analyzed: one event per page record actually written to the dataset, never more than maxPages, and never twice for the same page. A page that could not be loaded is reported with httpStatus: 0 and an empty image list, and is not charged. A page the site answers with an HTTP error (a broken link's 404, a server error page) is reported with that status so you can see it, and is not charged either, nor are its images. A page the site turns away — the challenge or block page of an anti-bot protection (Cloudflare, Akamai, AWS WAF, DataDome, Imperva) served in its place, or a 401, 403 or 429 refusal — is reported like a page that could not be loaded, with turnedAwayBy naming what answered ("Cloudflare challenge", "HTTP 403"): it is not analyzed, not charged, takes none of the maxPages places, and its links are not followed. The run's status message says how many pages were turned away and by what.
• Image Analyzed: one event per image analyzed on those pages, after deduplication by URL.

Only HTML pages count as pages. Links to files (images, PDFs, documents, feeds, archives) are not followed, and an address that turns out to serve a file is neither written nor charged: its place goes to a real page. A startUrl that leads to a file ends the run with a message saying so, and nothing is charged.

The amounts are shown in the pricing section of this page.

Limits, stated up front

• Weight is read from HTTP headers, not by downloading the image. A server that returns neither Content-Length nor a byte range leaves bytes at null, and the weight and format checks are then skipped for that image.
• Inline images (data: URIs) are measured from the page itself. Their bytes is the decoded size of the embedded payload, and their format comes from the declared media type, so they get the same weight and format verdicts as hosted images. They have no httpStatus and no cacheControl, since nothing is requested over the network. Note that a base64 payload takes about a third more room inside the HTML document than the size reported here.
• The imageUrl of an inline image is shortened. A payload of several thousand characters is unreadable in a result table, so the field keeps the media type and the first 40 characters, then states how many were dropped: data:image/avif;base64,AAAAHGZ0eXBtaWYxAAAAAG1pZjFhdmlmbWlhZgAA… (+6264 chars). Deduplication still compares the full payload.
• CSS background images have no alt text and no natural size, by construction. They are therefore always reported as ALT_MISSING, and never get an UPSCALED or DOWNSCALED verdict.
• The page is not scrolled. Images that a lazy-loading script only fetches once the visitor scrolls may be missing from the report, or reported without their final source.
• A site behind an anti-bot protection may turn the audit away. The audit does not try to get past a protection: such pages come back with turnedAwayBy and no images, uncharged. If it is your site, let the audit through (for example with an allow rule for the userAgent set in the input) and run it again.
• pagesCrawled counts every page record, including the pages that failed to load and the pages the site turned away; pagesTurnedAway counts the latter.

Typical uses

Pre-migration audit of a site's images, accessibility check on alt text, hunting down the heavy images that slow a page down, and checking that a WebP/AVIF migration actually reached every template.