Quick answer: audit AI-written alt text by looking at each image and asking four questions. Does the text describe what is actually shown? Does it add meaning for someone who cannot see the image? Is it concise, without keyword stuffing? And is it the right choice for this image, including an empty alt for purely decorative ones? Google says to review alt text and other metadata written by AI, because errors there can appear in search results.
Alt text is a small field that gets a lot of automation. Image tools, plugins and chat assistants will all offer to write it for you in bulk. The results are often fine and sometimes wrong in ways that are easy to miss, such as naming an object that is not in the picture. This guide explains what alt text is for, using Google and W3C documentation, and gives you an audit routine for AI-written text. This post is one part of the guide to AI tools for bloggers and creators.
The wider picture of image file names, compression and placement is covered in the earlier post on image SEO for bloggers. This post focuses on the audit step.
What Alt Text Is For
Google’s image SEO documentation calls alt text the most important attribute for providing metadata about an image. The text is described as useful for accessibility, because it helps people who cannot see images, including screen reader users and people on low-bandwidth connections. Google also says it uses alt text along with computer vision algorithms and the contents of the page to understand the image’s subject, and that alt text is useful as anchor text when an image is used as a link.
So alt text serves two readers: a person using assistive technology and a search engine. The person comes first. Text that reads naturally to a screen reader user and conveys what the image adds to the page usually serves search well too.
Google’s advice on writing it is short. Focus on useful, information-rich text that uses keywords appropriately and fits the context of the page, and avoid filling alt attributes with keywords, which is keyword stuffing and can make the site look like spam. The documentation illustrates the range with examples labeled missing, stuffed, better and best, and the best examples are specific without being long.
What the W3C Adds: Not Every Image Needs a Description
The W3C Web Accessibility Initiative offers an alt decision tree that is a handy audit tool. In summary: if an image contains text that also appears nearby as real text, or is purely decorative, use an empty alt attribute. An image in a link or button, where the destination would be unclear without it, needs alt text that communicates where the link goes or what the action does. An image that contributes meaning to the page and is a simple graphic or photograph gets a brief description that conveys that meaning. A chart or complex graphic needs its information included elsewhere on the page.
AI tools often skip this thinking. They describe every image, including decorative ones, which can add noise for screen reader users. A divider banner or a purely atmospheric picture may be best left with an empty alt. Deciding that is a human judgment, so build it into the audit.
The Four-Part Audit
For each image, work through these checks in order.
- Accuracy: open the image and compare it with the text. Are the objects, colors, number of people, actions and setting correct? Remove anything that is not visible.
- Purpose: does the image carry meaning, serve as a link, or merely decorate? Choose a description, a function statement or an empty alt accordingly.
- Length: is it as short as it can be while still conveying the point? Cut filler such as “image of” or “picture showing,” since screen readers already announce an image.
- Keywords: is any keyword there because it fits, and not because it was inserted? Rewrite any phrase that reads strangely aloud.
Google does not set a character limit in the documentation reviewed for this post, so any specific number you see elsewhere is a convention and not a rule. A practical test works better: read the alt text aloud. If it is a sentence or a short phrase a person could say in one breath, and it tells a listener what they would otherwise miss, it is probably the right length.
Common Ways AI Gets Alt Text Wrong
Several failures show up repeatedly in AI-written alt text:
- Hallucinated details: naming objects, brands or text that are not in the image.
- Wrong counts or relationships: “two children” when there are three, or the wrong person doing the action.
- Generic text: “a person using a laptop” for an image whose point is a specific screen or setting.
- Keyword stuffing: a repeated target phrase tacked on to every image on the page.
- Describing the prompt, not the image: for AI-generated images, the text may repeat what you asked for instead of what the generator produced.
- Over-length: paragraph-long descriptions where a short phrase would do.
That fifth item matters for sites that generate images with AI. The generator may not follow the prompt exactly, so an alt text copied from the prompt can be wrong. Always describe the final image. Review each generated image for oddities before it is uploaded, such as distorted hands, garbled lettering or extra objects, and describe only what is truly there.
Before-and-After Examples
These examples are invented to show the editing, not to describe any real image. Suppose a post about meal prep has a photo of five glass containers filled with chicken, rice and broccoli, lined up on a kitchen counter.
- AI draft: “meal prep meal prep ideas healthy meal prep containers best meal prep for weight loss.” This is stuffed, and it makes a health claim the image does not support.
- Edit: “Five glass containers of chicken, rice and broccoli lined up on a kitchen counter.” This is accurate and specific.
- Better in context: “Five glass meal prep containers filled with chicken, rice and broccoli on a kitchen counter.” The topic appears because it fits.
Now a decorative banner behind a heading, a plain gradient with no content. The AI draft says “colorful abstract background with soft shapes.” Under the W3C decision tree, this is purely decorative, so an empty alt attribute is the better choice. And a screenshot of a settings menu: “screenshot of settings” is too vague. “Settings menu with the Notifications toggle turned on” conveys what the reader needs to see.
Auditing a Whole Site Without Losing Your Mind
On a site with many posts, auditing each image one by one can feel impossible. Prioritize. Start with your most visited posts and posts that rank for image-heavy searches. Then work through newer posts, since those are the ones most likely to have been written with automation.
In WordPress, alt text is stored in the media library and in each post’s image block. Many themes and plugins let you see images without alt text in a single list, which is a useful first filter. For the images that do have text, a quick review pass is still worth doing for anything an AI tool wrote. Keep a simple spreadsheet with the post, the image name, what the alt said, what you changed and the date.
Do not change a published post’s content beyond the alt text during an audit, and do not delete images. The audit is meant to improve accuracy, not to rewrite posts. Correct the text of an inaccurate image, or replace the image, through your normal editing process.
Set Up a Prompt That Makes Review Easier
If you use an AI assistant to draft alt text, a better prompt makes the audit faster. OpenAI’s guidance recommends clear, specific instructions and iterative refinement. For alt text, that might look like: “Describe this image in one short sentence for a screen reader user. Mention only what is clearly visible. Do not guess brands or text. Do not start with image of. If the image is decorative, say DECORATIVE.” You still check the result, but the draft starts closer to what you need.
Share the page topic with the assistant too, so it can describe the image in context. A photo of a laptop means different things in a post about home offices and a post about repairing screens. Context is part of what Google says makes alt text useful.
Remember that Google’s generative AI guidance says review applies to metadata including alt text, as well as titles and descriptions. The broader accuracy process is in the guide to fact-checking AI-written drafts, and alt text is one more place where that discipline applies.
A Printable Alt Text Checklist
- The text describes what is actually in the final image.
- The image is informative, functional or decorative, and the alt text matches that role.
- Decorative images have an empty alt attribute.
- No “image of” or “picture of” at the start.
- No repeated keyword phrases.
- Short enough to read aloud in one breath.
- Brand names and visible text are included only if they are actually in the image.
- For a linked image, the text says where the link goes.
Alt Text for Different Image Types
Different images need different approaches, and an audit should treat them separately.
- Photographs: describe the subject and setting in a short phrase, such as who or what is shown and what is happening, skipping details that do not matter to the page.
- Screenshots: describe what the reader needs to see, such as the menu name and the setting highlighted. If the screenshot contains important text, that text should be available on the page too.
- Charts and diagrams: give a short summary of the point in the alt text, and put the full data in the surrounding text or a table, as the W3C decision tree recommends for complex images.
- Logos and icons: use the name of the organization, or the function if the image is a link or button.
- AI-generated illustrations: describe what the image shows, which may differ from the prompt, and leave an empty alt if it is purely decorative.
- Banners above headings: usually decorative if the heading already says what they illustrate, but give a description when the banner adds meaning.
Alt Text, Titles and Captions: What Each Field Does
WordPress gives each image several text fields, and people often fill them with the same words. They do different jobs. The alt text is for people who cannot see the image and for search engines. The title attribute is short extra information that some browsers show when hovering. The caption is visible text under the image for everyone. The description is mostly internal.
Google’s image documentation notes that it extracts information about an image from the page content, including captions and image titles, and recommends placing images near relevant text. So the fields can support each other, but they should not be copies. A caption that adds context for all readers, a short title and a plain alt description cover the ground without repetition.
Finish the audit by viewing the published page the way a screen reader user might: read the page top to bottom and listen to the image descriptions in context. Do they add information, or do they interrupt the text? Descriptions that repeat the surrounding paragraph word for word can be shortened or removed, and images that clearly illustrate a heading may only need a brief phrase.
Frequently Asked Questions
How long should alt text be?
Google’s documentation sets no character limit. Aim for a short phrase or sentence that a screen reader user could take in at once, and cut anything that does not add meaning.
Should every image have alt text?
Every informative image should have a useful description. Purely decorative images should have an empty alt attribute, according to the W3C decision tree.
Can I put my target keyword in alt text?
Only where it fits naturally in an accurate description. Google warns against keyword stuffing in alt attributes.
Is AI-written alt text bad?
No, but it needs review. AI can describe things that are not there, so compare each description with the image itself.
Do I need to audit old posts?
Prioritize your most visited and newest posts first, and work through the rest over time.