Alt attributes provide a text alternative for images, used by screen readers when a user cannot see the image and by search engines and multimodal AI models to understand what the image depicts. Fixing this issue means writing concise, descriptive alt text for every meaningful image on the page, and using empty alt for purely decorative images so assistive technology can skip them. The same alt text helps multimodal LLMs and image search tools surface and cite your visuals.

Example

<img src="/images/marathon-shoes.jpg"
     alt="Side view of a lightweight neon-green marathon running shoe">

<img src="/images/divider.svg" alt="">

Where to make the change

• Raw HTML: add the alt attribute to every <img> tag. For decorative images, use alt="" so screen readers skip them.
• WordPress: the media library has an Alt text field on every uploaded image. Fill it once and the value is reused everywhere the image is inserted.
• Shopify: in the product or page editor, click an image and use the Edit alt text option. Theme code that renders product images should pull image.alt.
• Wix or Squarespace: click the image in the editor and use the Alt text or Image description field in the side panel.
• Headless or framework sites: require an alt prop on every image component and lint for missing values during build.

Common causes and how to resolve them

• Attribute missing: add alt="" at minimum so the page validates. Replace with descriptive text on any image that conveys information.
• Alt text is the filename: "IMG_4837.jpg" tells the user nothing. Replace with a description of what is in the image.
• Alt text duplicates the visible caption: screen readers will read both, doubling the user's listening time. Differentiate the alt from the caption or empty the alt.
• Decorative images get descriptive alt text: background patterns and dividers should use alt="", otherwise screen readers narrate noise.
• Bulk imports left alt blank: a one-off audit of the media library or a CMS-side script can fill missing alt values from product names, captions, or asset descriptions.

Best practices

• Describe the function, not just the appearance: for a chart, describe what the chart shows. For a button image, describe what the button does.
• Keep it concise: aim for under 125 characters. Screen readers do not pause mid-sentence, so long alt text is exhausting to listen to.
• Skip "image of" or "picture of": the screen reader already announces it as an image.
• Include the focus keyword when natural: alt text is a minor on-page signal, but stuffing keywords in unnatural alt text harms both accessibility and ranking.

This test fails when the raw HTML document weighs significantly more than necessary, even before counting images, CSS, or JavaScript. Bloated HTML increases time to first byte, slows parsing, and stalls rendering on mobile networks and lower-powered devices. Fixing this issue means trimming inline data, removing legacy markup, and keeping the document focused on what the user needs.

Where to make the change

• Application code or templates: identify which sections produce the most bytes (often a large inline JSON blob, an inline base64 image, or a giant comment block) and refactor them to load externally or be removed.
• WordPress: check for plugins or page builders that emit large inline configurations. Disable unused plugins and prefer themes that ship clean markup.
• Shopify: theme code may inline large product or section JSON. Move heavy data to a fetch-on-demand endpoint when possible.
• Headless or framework sites: review the rendered output for serialized state (for example __NEXT_DATA__ in Next.js) and trim what is sent on first paint by streaming or lazy-loading non-critical chunks.

Common causes and how to resolve them

• Inline base64 images or fonts: move them to external files so the browser can cache them across requests.
• Excessive inline CSS or JavaScript: link to external stylesheets and scripts so they cache. Inline only the truly critical CSS for above-the-fold content.
• Large server-side JSON payloads: ship only the data needed for first render. Fetch the rest on demand.
• Repeating boilerplate or commented-out code: strip comments and dead code in the production build.
• HTML compression disabled: Gzip or Brotli compression often shrinks the response by half or more. Confirm the server sends the appropriate Content-Encoding header.

Best practices

• Aim for under 100 KB of HTML: most pages can comfortably stay under this. Anything beyond a few hundred KB usually indicates structural bloat worth refactoring.
• Enable compression: Gzip is the safe minimum, Brotli is even better for HTML and other text formats.
• Defer non-critical content: rendering only what is needed for the initial viewport keeps both bytes and parse time down.
• Audit periodically: add HTML size to your performance budget so regressions get caught in CI.

This test fails when the page builds a Document Object Model with too many nodes or excessive nesting depth. Large DOMs increase memory usage, lengthen style and layout calculations, and slow rendering, particularly on mobile devices. Fixing this issue means simplifying the markup so the browser has less work to do on every interaction.

Where to make the change

• Application code or templates: identify the heaviest sections (long lists, repeating components, deep grid wrappers) and either paginate them, virtualize them, or flatten the markup.
• WordPress: heavy page-builder themes often emit deeply nested wrapper divs. Consider switching templates or using a theme that emits cleaner markup.
• Headless or framework sites: use a virtualized list component for long item lists so only visible rows are rendered into the DOM.

Common causes and how to resolve them

• Long lists rendered in full: show a paginated subset or virtualize the list so only on-screen items exist in the DOM.
• Excessive wrapper divs: a stack of layout wrappers ten deep can usually be flattened to two or three with modern CSS layouts (Grid, Flexbox).
• Hidden modals and dropdowns kept in the DOM: render them on demand instead of always emitting them with display: none.
• Third-party widget injecting many nodes: a chat widget or carousel may add hundreds of nodes per page. Lazy-load it after main content renders.

Best practices

• Aim for under 1500 nodes total and a maximum depth of 32: Lighthouse uses these as warning thresholds and they correlate with smooth rendering on mid-tier mobile devices.
• Render only what is visible: long lists, accordions, and modals belong outside the initial DOM until needed.
• Prefer flat semantic markup: a single <article> with a few children is faster than a deeply nested <div> tree.
• Audit with Lighthouse: the DOM size audit reports the worst-offending elements, which often points at a single template that needs rework.

This test fails when the browser spends excessive time parsing, compiling, and executing JavaScript while the page loads. Heavy JavaScript work blocks the main thread and is the primary cause of poor Interaction to Next Paint (INP), the Core Web Vital that replaced First Input Delay in 2024 and that Google uses as part of its ranking algorithms. Fixing this issue means shipping less JavaScript, splitting what remains, and deferring everything that is not needed for the initial interaction.

Where to make the change

• Application code: audit your bundle with a tool such as Bundle Analyzer or Source Map Explorer. Identify the largest dependencies and either remove, replace, or lazy-load them.
• Build configuration: enable code splitting so each route loads only its own chunk. Tree-shake unused exports from libraries.
• Third-party scripts: the biggest INP wins often come from removing or deferring analytics, advertising, and tag-management scripts.

Common causes and how to resolve them

• Single large bundle: split the bundle by route or feature so the user only downloads and executes the code for the current page.
• Heavy third-party tags: load analytics, ads, and chat widgets after main content with async or defer, or move them into Google Tag Manager so they can be controlled centrally.
• Unused JavaScript: audit with the Coverage tab in DevTools to find code that loads but never runs on the page. Remove or split it out.
• Long tasks blocking the main thread: break work into smaller chunks with setTimeout, requestIdleCallback, or scheduler.yield(). Move heavy computation to a Web Worker.
• Polyfills shipped to modern browsers: use modern-only bundles for ES2017+ and serve legacy polyfills only when needed via differential serving.

Best practices

• Target sub-200 millisecond INP at the 75th percentile: this is the official "good" threshold for the Core Web Vital.
• Ship less code: the cheapest JavaScript to execute is the JavaScript you never sent. Audit dependencies for lighter alternatives.
• Defer non-critical scripts: use defer for scripts that need the DOM, async for fully independent ones.
• Move CPU-intensive work off the main thread: Web Workers can handle large data transformations without freezing the UI.

This test fails when one or more images carry disproportionately large embedded metadata such as EXIF data, thumbnails, color profiles, geolocation, and camera information. This metadata is rarely useful on the web and adds bytes to every request without any rendering benefit. Fixing this issue means stripping the metadata as part of your image processing pipeline.

Where to make the change

• Image build pipeline: use a tool such as ImageMagick (mogrify -strip image.jpg), jpegoptim --strip-all, or oxipng --strip all to remove metadata at build time.
• WordPress: a media optimization plugin will strip metadata automatically on upload.
• Shopify, Wix, Squarespace: the platforms strip most metadata server-side. Failures usually point to images served from external sources you control.
• Image CDN: Cloudflare Images, Cloudinary, and Imgix strip non-essential metadata by default during transformation.

Common causes and how to resolve them

• Original camera files uploaded as-is: camera files contain large EXIF blocks (lens, GPS, embedded JPEG previews). Strip during upload or build.
• Color profile larger than the image: embed a small sRGB profile or strip the profile entirely if the image is meant for the web.
• Stock photo metadata: stock images often carry copyright notices, IPTC blocks, and watermarks larger than the image data itself.
• Geolocation in personal photos: beyond bytes, EXIF GPS can leak the photographer's location. Strip for privacy as well as performance.

Best practices

• Strip metadata on upload: make it part of the asset pipeline so no human action is required.
• Preserve only essentials: a small ICC color profile is occasionally worth keeping for accurate color rendering; everything else can usually go.
• Verify with a quick check: tools like exiftool image.jpg show what is left after stripping, so you can confirm only what you want remains.

This test fails when external links opened in a new tab (with target="_blank") lack the rel="noopener" or rel="noreferrer" attributes. Modern browsers (Chrome 88+, Firefox 79+, Safari 12.1+) apply noopener implicitly, so this is largely a defense for older browsers and a way to keep the destination page from running in the same browser process as the original tab. Fixing this issue means adding the attributes explicitly to every external target="_blank" link.

Example

<a href="https://example.com" target="_blank" rel="noopener noreferrer">
  External link
</a>

Where to make the change

• Application code or templates: search for target="_blank" and add rel="noopener noreferrer" to each match. A simple find-and-replace handles most cases.
• WordPress: the block editor adds rel="noopener" automatically on every external link opened in a new tab. Older content may need a one-time bulk update.
• Shopify, Wix, Squarespace: editor-generated links typically include the rel attributes. Manual HTML embeds need them added.
• Headless or framework sites: wrap external links in a component that always sets the correct rel attributes.

Common causes and how to resolve them

• Hand-written external links missing the attribute: add rel="noopener noreferrer" to each one.
• Old content from before the editor added it automatically: a one-time bulk update through the database or CMS export covers it.
• Markdown-rendered content: configure the renderer to add the attributes to every external link by default.

Best practices

• Use noopener noreferrer together: noopener handles the security concern, noreferrer additionally hides the referring URL. Both are safe defaults for most external links.
• Reserve target="_blank" for genuinely external links: internal links should open in the same tab so users can use the back button naturally.
• Add a lint rule: ESLint plugins for React and similar frameworks can enforce the rel attribute automatically.

This test reports on whether the page applies a nofollow directive that tells search engines not to follow outbound links. The nofollow attribute (and the related sponsored and ugc values) controls how link equity flows from your page. Fixing this issue means using the right rel value on each external link based on its purpose so signals are passed only where you intend.

Where to make the change

• Raw HTML: add rel="nofollow", rel="sponsored", or rel="ugc" to specific outbound links rather than blanket-applying them site-wide.
• WordPress: the block editor offers a "mark as nofollow" option per link, and SEO plugins can auto-tag external links by domain.
• Shopify, Wix, Squarespace: editor-generated links allow rel attribute toggles; manual HTML embeds need them set explicitly.
• Headless or framework sites: wrap external links in a component that decides the rel based on the link's classification.

Common causes and how to resolve them

• Site-wide nofollow on every external link: overly aggressive. Reserve nofollow for sponsored, untrusted, or user-generated content.
• Sponsored content not labeled: use rel="sponsored" on paid placements and affiliate links to comply with Google's link spam policy.
• User-generated content unmarked: comments, forum posts, and review user submissions should use rel="ugc".
• Trusted editorial links nofollowed: these should pass equity. Remove the nofollow.

Best practices

• Use the most specific value: sponsored for paid links, ugc for user content, nofollow for everything else you want to disclaim. They can be combined: rel="sponsored noopener".
• Do not nofollow internal links: "PageRank sculpting" with internal nofollow no longer works and just wastes crawl signal.
• Audit periodically: as content ages, sponsored relationships change. Review affiliate and partner links once a year.

This test fails when the page uses a meta refresh tag to automatically reload or redirect after a delay. Meta refresh is largely obsolete: it hurts accessibility by jumping focus and disorienting screen reader users, and it is a weaker signal to search engines than a proper server-side 301 redirect. Fixing this issue means removing the meta refresh and using the appropriate alternative (a 301 redirect for permanent URL changes, JavaScript or CSS for dynamic content updates).

Where to make the change

• Raw HTML: remove <meta http-equiv="refresh" ...> from the <head>.
• Server configuration: replace with a proper 301 redirect at the server, CDN, or platform level.
• WordPress: a redirection plugin handles 301s without touching code. Remove any meta refresh injected by older themes or shortcodes.
• Shopify, Wix, Squarespace: use the platform's URL redirect settings rather than meta refresh.

Common causes and how to resolve them

• Meta refresh used as a redirect: replace with a server-side 301. Search engines treat 301 redirects as authoritative; meta refresh is a fallback at best.
• Meta refresh used to reload data: use fetch() with setInterval or a real-time data layer instead. Reloading the entire page wastes bandwidth and frustrates users.
• Inherited from a legacy template: remove the tag from the template; do not just override it on individual pages.

Best practices

• Use 301 redirects for URL changes: they pass full ranking signal to the new URL and are cached aggressively.
• Use 302 only for genuinely temporary moves: the 302 status indicates a temporary redirect; reserve it for short-lived changes.
• Avoid auto-reloading content: if real-time updates matter, fetch only the changed data rather than reloading the page.