A good image does three jobs at once. It helps a sighted reader grasp a point faster. It gives a blind reader the same facts in words. And it loads fast enough that no one waits for it. This guide shows a simple workflow for all three: give each image a clear job, write alt text from that job, pick a sane format and size, place the image near the text it supports, then check the live page.

What should alt text actually say?

Alt text is a stand in for the image. It is what a screen reader reads out loud when a person cannot see the picture. It is also what shows if the image fails to load. So the test is simple. Read the alt text on its own. Does the reader still get the point of the image? If yes, the alt text works. If it only names the file or repeats the heading, it does not.

Start with the question the nearby text asks. Then say what the image adds to the answer. A diagram next to a paragraph about checking images might get alt text like “Six step workflow from image source to final check on the live page.” That tells the listener what the diagram is for. A bare label like “workflow” does not. A full inventory of every box, arrow, and color would take too long to listen to, and it would bury the point.

Keep it short. One or two plain sentences cover most images. End with a full stop, since many screen readers pause at punctuation and that pause helps the listener. Skip openers like “image of” or “photo of”. The listener already knows it is an image. That is what the screen reader announced before it read your text.

Do not stuff keywords into alt text. A string of search terms is hard to listen to, and it tells the reader nothing new. If the honest description of the image happens to include a phrase people search for, that is fine. The phrase is there because it is true, not because it was forced in. The same rule applies to filenames and captions. Write for the person first, and the search engine can use the same honest words.

One more habit helps. Write the alt text when you place the image, not later in a bulk pass. You know the job of the image best at the moment you choose it. A week later, you will guess. A bulk pass across fifty images turns into copy and paste, and copy and paste turns into the same vague phrase on every image. The WebAIM guide to alt text walks through the same idea with clear examples for each kind of image.

Six stage workflow connecting a verified image source, visual purpose, nearby explanatory copy, appropriate text alternative, durable public delivery and final validation
Figure 1. A useful image follows a chain from source and purpose to nearby text, an honest alternative, durable delivery, and a final check on the live page.

How do you decide what kind of image this is?

Not every image needs the same treatment. The right alt text depends on what the image does on the page. Ask one question: if the image vanished, what would the reader lose? The answer sorts the image into one of four kinds. Each kind has its own rule.

An informative image carries facts the text does not. A photo of a product, a chart, a map, or a diagram of a process all count. Give it a short alt text that states the essential fact. Put the fine detail in the body text or the caption, where every reader can reach it. The alt text is the summary. The page is the full story.

A decorative image adds mood or shape and no facts. A soft wave between sections, a texture, or a background flourish all count. Give it an empty alt attribute, written as alt=””. That is not the same as leaving alt out. A missing alt makes many screen readers read the filename out loud, which is worse than silence. An empty alt tells the screen reader to skip it. If the page still makes full sense with the image gone, the image is decorative.

A functional image is a control. It is a link or a button, often an icon. Name the action, not the picture. A chat icon that opens a chat gets alt text like “Open chat”. A printer icon gets “Print this page”. A logo that links home gets the site name or “Home”. Describing the shape of the icon helps no one. The listener needs to know what happens when they activate it.

A complex image holds too much for one or two sentences. Think of a chart with twelve data points, or a map with many stops. Keep the alt text short and name the kind of information it holds, such as “Bar chart comparing load time before and after image compression.” Then put the full numbers in a table or in the body text next to it. That way the data is open to everyone, including people who can see the chart but want exact figures. The W3C Images Tutorial sorts images into these same kinds and gives examples for each.

Images of text deserve a warning of their own. If words only exist inside a picture, a screen reader cannot read them, a phone user cannot tap to copy them, and a small screen may shrink them past reading. Logos are the fair exception. For everything else, put the words in real text on the page and let the image support them. A step in a process, a price, a condition, or a warning should never live only in pixels.

Four column comparison of informative, decorative, functional and complex image roles, showing the matching text alternative approach for each
Figure 2. Image role decides the alt text: explain information, skip decoration, name the action, or pair a complex visual with full detail in text.

What makes a good filename?

A filename is a small label for humans first. It helps the next editor find the right file in the media library, and it helps you avoid uploading the same diagram twice under two names. Pick a short name that says what the picture shows. “image-seo-workflow-check.png” is good. “IMG_0042.png” says nothing. “image1.jpg” says nothing. A long string of keywords is worse than either, since it reads like spam and tells the editor nothing about which version is current.

Keep the name honest. A filename must never claim the picture shows a real client, a real office, a real result, or a real place when it does not. A file called “brampton-seo-results-chart.png” is not proof of results. It is a label on a drawing. If the chart shows sample data, say so in the caption and the body text. The filename is not the place to make claims the page cannot back up.

Match the extension to the real file type. A PNG saved with a .jpg ending confuses tools and editors, and it can break delivery in some setups. When you export a new version of a diagram, give it a clear version or date in the name, such as “image-seo-workflow-check-v2.png”, and delete or archive the old one. A media library full of near copies named “final”, “final2”, and “final-new” is how the wrong diagram ends up on a live page.

One more point. A filename is a light hint about the subject, not a ranking lever. It will not rescue an image that sits on an unrelated page, has no alt text, and adds nothing to the reader. The page around the image does the heavy work: the heading, the paragraph before it, the caption under it. Get those right and the filename is a small bonus. Get them wrong and no filename will fix it.

Which file format should you pick?

Pick the format by what the picture is, not by what is trendy. Each format has a job it does well, and a job it does badly. Match the format to the picture and you get small files with clean edges. Force one format on everything and you get blurry diagrams or giant photos.

Use JPEG for photos. It compresses smooth tones well, which is what a photo is made of. Use PNG for diagrams, screenshots, and anything with sharp lines or flat color, since it keeps edges crisp. Use SVG for logos, icons, and simple line art, since it scales to any size without blur and stays tiny. Use WebP or AVIF for photos and diagrams when your site and your audience’s browsers support them, since they often beat JPEG and PNG on size at the same quality. Keep a JPEG or PNG fallback in the markup when you serve the newer formats, so nothing breaks on an older browser.

Keep GIFs for short simple animation only. A GIF is a poor choice for a photo and a poor choice for a large diagram. If you need motion, a short video file is often smaller and sharper than a GIF of the same clip. The images on this page are PNG diagrams, which fits their job: flat color, sharp lines, and text labels that must stay readable.

Whatever you pick, check the result with your eyes. Open the exported file at the size it will show on the page. Read the smallest label. If the text in a diagram is fuzzy, the format or the export size is wrong. No amount of alt text fixes a diagram nobody can read. The web.dev image course explains how each format behaves and when to reach for it.

Responsive image delivery path showing optional source candidates leading to a standard HTML image element with a stable source URL, meaningful alternative text and a loaded public asset
Figure 3. Responsive sources can improve delivery, but an informative image still needs a plain HTML image element, a fallback source, useful alt text, and a check that it loads.

How big should the image file be?

Big images are the most common reason a page feels slow. A photo straight from a camera can be several megabytes. A page that shows it at 800 pixels wide does not need the 4000 pixel original. Resize the file to near the size it will display, then compress it. Most photos look the same to the eye at a sensible compression level, and the file can shrink by half or more.

Work out the largest size the image will ever show. A full width hero on desktop might need 1600 pixels wide. A diagram inside an article column might need 800. An icon might need 64. Export at that width, plus a little headroom for sharp screens, and no more. Uploading a 4000 pixel file and letting the browser shrink it wastes every visitor’s data and time.

Serve responsive versions when the same image shows at very different sizes. The srcset attribute lets the browser pick a small file on a phone and a larger one on a desktop. Always keep a plain src as the fallback, so the image still loads where srcset is not supported. The MDN reference for the img element shows how src, srcset, and sizes work together.

Set width and height on every image. When the browser knows the shape of the image before it loads, it can hold the right amount of space. Without those numbers, the page text jumps down the screen when the image pops in, and readers lose their place. That jump is a layout shift, and it is one of the Core Web Vitals measures. A stable page is easier to read and scores better.

Compress with a tool, not by eye alone. A good export preset or a compression tool will cut file size while keeping the picture clean. Then look at the result. If edges smear or text blurs, ease off the compression. If it still looks clean, you may be able to push further. The aim is the smallest file that still looks right at its display size, not a magic number of kilobytes.

When should an image load lazily?

Lazy loading tells the browser to wait before fetching an image until the reader scrolls near it. That is a win for images far down the page, since many readers never scroll that far and the browser never has to fetch files nobody sees. Add loading=”lazy” to images below the fold and the first screen loads faster.

Never lazy load the image a reader sees first. The hero image, or the first diagram at the top of an article, should load right away. Delaying it makes the top of the page look empty, and it slows the largest paint on the page, which is another Core Web Vitals measure. For that first image, use fetchpriority=”high” instead, and skip the lazy attribute.

Add decoding=”async” to images that are not urgent. It lets the browser decode the picture without blocking the rest of the page. It is a small setting, and it costs nothing. The five diagrams on this page all use loading=”lazy” and decoding=”async” because they sit inside article sections, below the first screen.

Watch out for one trap. Lazy loading can hide an image from a quick check if you only look at the top of the page. Scroll the full page after you publish, and confirm every image actually appears. A lazy image that never loads because of a script clash looks exactly like a broken image to the reader who scrolls to it. The MDN guide to lazy loading covers the attribute, its limits, and when not to use it.

How do captions and nearby text help?

An image works best when the reader meets it after the question it answers. Say what the reader is about to see in a heading and a short paragraph. Show the image. Then use the caption to name the figure and state the takeaway in one line. Then move on to the next point. That order lets a reader scan the page and still follow the argument, even if the image is blocked or slow to load.

Captions and alt text are partners, not copies of each other. The caption is visible to everyone. It can explain why the diagram matters or point at the part to notice. The alt text is a stand in for the picture itself. If the caption repeats the alt text word for word, the screen reader user hears the same sentence twice. Write them to do different jobs: alt text says what the image is, the caption says why it matters here.

Keep captions honest. A caption should never make a claim the article cannot support. “Figure 2. Four alt text decisions for image roles” is a clear label. “Figure 2. The proven way to rank image results” is a promise no caption can keep. If the picture shows a sample or a made up example, say so with a plain label like “example workflow” or “illustrative decision”. A reader should never mistake a sketch for a client story.

Place images near the text they support. A diagram about image checks belongs next to the paragraph about image checks, not three sections later. Search engines read the words around an image to work out what it shows, and readers do the same. When the picture and the paragraph drift apart, both get weaker. Our on-page SEO checklist covers this wider point: a page reads well when each part supports the heading above it. Images also pass value through links, so the way pages connect matters too. Our internal linking guide for small business sites shows how to link related pages so readers and crawlers can move between them.

Think about narrow screens as well. A wide diagram with small labels can look tidy on a desktop and turn unreadable on a phone. Cut the labels down, put the detail in the body text, or stack the steps in a vertical layout. A table that scrolls sideways is another honest option for dense data. The aim is not to squeeze a desktop picture onto a phone. The aim is to keep the reading order, the purpose, and the caption intact when the space shrinks.

How do you keep images honest and current?

Most image failures are maintenance failures. A process diagram still shows last year’s steps. A photo still shows a person who left. A product shot still shows a model you no longer sell. An alt text still describes a diagram you redesigned. Nobody meant to mislead anyone. Nobody owned the file, so nobody checked it.

Keep a small ledger for the images that matter. One row per image is enough. Note the page and section, the public URL, the source and rights note, the image kind, the alt text decision, the owner, and the date you last checked it. Also note what should trigger a fresh look: a service change, a redesign, a campaign ending, a content refresh, or a reader reporting a problem. When one of those triggers fires, the ledger tells you which images to review instead of leaving you to guess.

Name a real owner for each image. An owner is a person or a role who can answer for the file. An editor can own the caption. A service lead can check a process diagram for facts. A developer can own the markup and the delivery. “The marketing folder” is a storage place, not an owner. If no one can answer a correction request, the image has no owner.

Be careful with stock photos. A licensed stock image is fine for a concept, a mood, or a generic illustration, as long as the page presents it honestly. It is not fine as proof. A stock photo of an office is not your office. A stock photo of a team is not your team. A stock chart is not your results. If the page needs proof, use real evidence or say plainly that the image is illustrative. Readers trust pages that do not pretend.

Review images when you refresh the page around them. A content refresh can change the argument so the old figure answers the wrong question. Our content refresh workflow shows how to decide whether a page should be kept, revised, combined, or set aside. Apply the same call to its images. Keep a diagram that still matches the current process. Revise one whose labels changed. Replace one whose source no longer supports its use. Remove one that has become noise or that repeats the paragraph above it. And when a redesign touches templates, check the images again, since a template change can strip captions or resize pictures without warning. Our website redesign SEO checklist covers the wider checks a redesign needs.

Image asset ledger with columns for public purpose, source or rights note, page location, image role, alt text decision, accountable owner, review trigger and public validation status
Figure 4. A small ledger ties each image to a purpose, a source, an owner, a review trigger, and a check on the live page.

How do you check the finished page?

An upload confirmation proves nothing about what visitors see. The editor preview can look fine while the live page shows a broken image, an old cached copy, a missing caption, or alt text the template rewrote. Check the public page after every publish. Open the real URL, in a normal browser, and scroll the whole page.

Work through a short list for each image. Does it load? Does it look sharp at its display size? Does the caption sit under it? Does the alt text match the image kind? Does a narrow screen keep the reading order? Does a keyboard user reach the link when the image is a button? If any answer is no, fix it before you move on. These checks sit inside the wider review in our technical SEO checklist, which covers crawl, rendering, and page foundations together.

  • Name the reader question and the visual job before you source or design the image
  • Record a truthful source or rights note, and never present a generic visual as local proof, client work, or results
  • Sort the image as informative, decorative, functional, or complex before you choose the alt text
  • Place the image near the text it supports, and use a caption to state the visible takeaway where it helps
  • Write short alt text from the image job, and use empty alt only for true decoration or repeated content
  • Keep required facts, steps, and claims in real page text, never only inside image pixels
  • Use a durable public URL and a plain HTML image element for informative pictures, with a fallback source in responsive markup
  • Resize and compress to the display size, set width and height, and lazy load only what sits below the first screen
  • Check load, layout, alt text, captions, and narrow screen order on the live page after release
  • Record the owner and the review trigger, and report only what you observed, not a promise about search results

One last point about promises. Good image work helps people and helps search engines understand the page. It does not promise a spot in image search, a preview, traffic, or a ranking. Those outcomes sit with the search engine. What you can promise is a page where every image has a job, every job has words to match, and every file loads cleanly. That is a solid result on its own, and it is one you can check with your own eyes.

Five stage public image validation ladder covering visual relevance, accessible text equivalent, successful asset load, rendered HTML inspection and an explicit boundary that no search outcome is inferred
Figure 5. A live page check proves delivery and access facts. It never proves a future spot in image search or a ranking.

Common questions

Does every image need alt text that describes every detail?

No. Alt text should give the essential fact or function in context. Decorative images take an empty alt attribute. Complex images take a short alt plus the full detail in nearby text or a table. A long list of every color and shape in the picture is hard to listen to and hides the point.

Can an important diagram be a CSS background image?

No. A background image has no alt attribute, so a screen reader user gets nothing from it. It also sits outside the normal content flow, so it is easy for it to vanish on some devices or in some tools. Use a plain img element and put the explanation in real text beside it.

Will good alt text and filenames get my images into image search?

They help search engines understand the image, but they do not promise a listing, a preview, traffic, or a ranking. Image search weighs the page, the image, the words around it, and many other signals. Treat alt text and filenames as honest inputs, not as a ticket to a result.

Should I lazy load every image?

No. Lazy load images that sit below the first screen. Load the first image right away, since delaying it leaves the top of the page empty and slows the largest paint. After any change to loading, scroll the full live page and confirm every image appears.

Can I use a stock photo to show a real client or office?

No. A stock photo can illustrate a concept if the page presents it honestly. It must never be labeled or described as a real client, a real office, a real team member, or real results. If the page needs proof, use real evidence or label the image as an illustration.

Want images that work as hard as your words?

Get a free audit: a 120-point crawl, a competitor gap map, and a 90-day roadmap, delivered within 48 hours. Free, with no obligation. It covers image delivery and page speed along with the rest of your site, and it shows which fixes matter most. See what we do on our services page, or start below.

Start with a free audit

Get a free SEO audit of your website

A 120 point crawl, a competitor gap map and a 90 day roadmap, ready in 48 hours. Free, with no obligation.

Get my free audit

SEOelinks Team
Written by
SEOelinks Team

SEOelinks is a Brampton-based SEO agency helping service businesses across Canada and worldwide win organic search. Our strategists publish the same field-tested playbooks on technical SEO, local visibility and AI search that we run for clients every day.

More about SEOelinks →

Leave a Reply

Your email address will not be published. Required fields are marked *