AI-generated alt text only works reliably if the image's purpose and context, rather than just its content, determine the text alternative. Let the AI create a draft, have it professionally reviewed, and only publish the alt text after human approval. This sequence prevents fabricated statements, unnecessary descriptions, and texts overloaded with keywords.
In my work with small businesses and editorial teams, I frequently encounter two problems: images lack alt text, or an automated AI-generated image description is adopted without verification. Image recognition can capture visible patterns and text. However, without additional information, the AI knows neither the image's purpose on the specific page nor the information a screen reader is meant to convey.
A good alt text doesn't describe as much as possible. A good alt text conveys exactly what the image does in the given context.
Creating alt text with AI: five steps
- Determine the purpose of the image: Is the image informative, functional, or decorative?
- Provide image context: On which page does the image appear, and what is immediately before or after it?
- Create an AI design: The AI may only use recognizable and editorially confirmed facts.
- Check alt text: Do the content, function, language, and necessary details match?
- Human release: A responsible person decides whether the text is published or whether the alt attribute is deliberately left blank.
This process can also be implemented in small teams without a dedicated department. A clear approval process is crucial. A robustly designed editorial system for SMEs can integrate the creation and review of alt texts using AI directly into media management.
What WCAG 2.2 requires for images
WCAG 2.2 success criterion 1.1.1 requires that non-text content has a text alternative that serves an equivalent purpose. Purely decorative content, visual formatting, and content that is not displayed must be implemented in such a way that assistive technologies can ignore them. This is described by the W3C in the success criterion for non-text content.
The Web Accessibility Initiative's alt decision tree distinguishes between informative images, functional images, and decorative images, among others. Additionally, the decision tree addresses complex graphics and images with text. These categories form the basis for deciding on the alt attribute. The W3C Web Accessibility Initiative's alt decision tree provides practical guidance.
Accessibility doesn't mean translating every visible detail into words. The goal is equal access to the content or function. A person who cannot see the image should be able to understand and use the page without having to listen to unnecessary descriptions.
The testing scheme: informative, functional, or decorative?
1. Informative images convey relevant content
An informative image conveys information that is not already fully contained in the surrounding text. This could be a product photo, a team photo, a picture of a completed project, or a simple explanatory graphic.
The text alternative answers a clear question in this case: What information would be missing if the image could not be loaded or perceived?
- Product photo: List the visible characteristics that are important for selection or understanding.
- Team photo: Only mention people if their identity is known, relevant, and editorially confirmed.
- Project image: Describe the relevant result instead of every tool and every object in the background.
- Text in image: Copy the main text, unless it is already present as actual text on the page.
2. Functional images trigger an action.
Functional images include, for example, linked logos, graphic buttons, or icons. For usability, the function or the link target is what matters , not the image's appearance.
If an image is the only understandable element in a link or button, its alt text can provide the accessible name. For example, a magnifying glass icon would then need "Search website" instead of "black magnifying glass". A linked company logo that leads to the homepage could have "Berger+Team - Homepage".
If the function is already displayed as visible text next to the icon, or is technically provided via another accessible name, the icon should not be read aloud twice. In this case, the image itself can be treated decoratively.
3. Decorative images retain alt=""
Decorative images complement the design but convey no additional information and have no function of their own. In this case, an empty alt attribute is correct.
alt=""
A deliberately empty alt attribute signals to assistive technologies that the image should be skipped. A missing alt attribute is not the same. If the attribute is completely missing, screen readers may, depending on the software, browser, and integration, provide technical information such as the file name.
Typical candidates for alt="" are decorative graphics, separator images, repeated mood images without informational value, or icons whose meaning is already explained directly next to them in text. The decision depends on the specific use case: The same photo can be decorative on one page and informative on another.
4. Complex images need more than a short alt text.
Diagrams, maps, and infographics often contain too much information for a single alt text. Use a concise alternative text that states the topic and key message, and supplement the full values or relationships as a visible long description within the page content.
A useful alt text could read: "Bar chart showing the development of queries from January to June; significant increase from April onwards." The specific monthly values should be included in a clearly visible explanation. This benefits both screen reader users and people who prefer to read data as text or view the graphic on a small screen.
Seven typical examples from everyday editorial work
Product photo
A photo on a product page is usually informative. Visible features that differentiate the product are relevant: material, color, design, or included components. Interchangeable phrases like "high-quality product photo," on the other hand, convey no concrete information.
team photo
A team photo can convey information about individuals, roles, or the number of team members depicted. Names should only be mentioned if they are known for certain and clearly identifiable within the image. Assumptions about age, origin, relationships, personality, or mood should not be included in accessible image captions.
diagram
A chart requires a brief description in the alt attribute and a detailed explanation in the visible content. The core message must be understandable even without the graphic. For business-critical data, the underlying values should also be made readily available.
Linked logo
A linked logo is functional. The accessible name identifies the company and, if applicable, the link destination. A description of colors, symbols, or typography is only necessary if this design element itself is part of the content.
Icon
An icon without a visible label must convey its action via an accessible name, such as "Open menu," "Download file," or "Start search." If the same text is visible next to the icon, the image can be treated decoratively to prevent the screen reader from outputting the label twice.
Mood image
An image can be decorative if it merely complements the design. If the same image is used to depict a specific place, achievement, or particular situation, it is informative. The page's purpose is the deciding factor.
Text in image
Important information should generally be displayed as actual text on the page. If relevant text is contained solely within the image, the alternative text must convey the necessary wording. If the visible page content already contains the same information, repeating it in the alt text is usually redundant.
Checking alt text with AI: five before-and-after examples
Example 1: Product photo of a wooden table
Poor AI design: "A beautiful, sustainable designer wooden table made from local oak, handcrafted by a traditional South Tyrolean family business."
Corrected version: "Rectangular dining table made of light oak with black metal legs."
Rationale: The AI draft invents origin, production, tradition, and sustainability. The corrected version is limited to visible characteristics relevant for product selection.
Example 2: Team photo
Poor AI design: "The motivated sales team is delighted with the successful year-end closing."
Corrected version: "Five members of the sales team are standing in front of the store in Bolzano."
Rationale: Motivation, enjoyment, and occasion cannot be reliably deduced from a photograph. The location may only be mentioned if the image context or confirmed metadata clearly establishes the location.
Example 3: Linked shopping cart icon
Poor AI design: "Black shopping cart with two wheels."
Corrected version: "Open shopping cart."
Reasoning: The icon is functional. For usability, what matters is the action it triggers, not the appearance of the symbol.
Example 4: Diagram
Poor AI design: "A diagram with multiple blue bars."
Corrected version: "Bar chart of monthly website requests: increase from 42 in January to 96 in June."
Reasoning: Chart type and colors are not sufficient. The content is what matters. The values mentioned must be clearly legible in the chart or editorially confirmed.
Example 5: Decorative mood picture
Poor AI design: "Sunbeams fall through green leaves and symbolize growth and hope."
Corrected version: alt=""
Rationale: If the image serves only as visual embellishment, a poetic description creates additional distraction. The editors must classify the image as decorative and leave the alt attribute empty.
What information does AI need for good alt text?
Automatic image recognition identifies visual patterns and any text they may contain. For AI-generated alt text to be truly useful, editorial information is also required.
- Image type: Photo, logo, icon, diagram, infographic or image with text.
- Page context: Product page, team page, project report, instructions or homepage.
- Image function: inform, trigger an action, or simply design.
- Visible facts: Confirmed people, products, places, values and characteristics.
- Surrounding text: Information that should not be repeated unnecessarily.
- Output language: the language of the respective page.
- Output format: Short alt text, alt="" or indication of a necessary long description.
This logic is also relevant in btlabs Core. The system includes a built-in function for alt text drafts. This function speeds up creation and review, but does not replace editorial approval. You can find a classification in the overview of website functions in btlabs Core.
This is how I understand meaningful AI integration into existing processes : The technology reduces routine work, while context, verification and responsibility remain with humans.
Creating alt text with AI: copyable prompt
Task: Create a suitable text alternative for an image on a website.
Image type: [Photo, product image, team photo, logo, icon, diagram, infographic or text in image]
Page context: [Describe the topic and purpose of the page.]
Function of the image: [informative, functional, unclear or possibly decorative]
Visible and confirmed facts: [List only reliable information.]
Directly adjacent text: [Insert heading, caption, or paragraph.]
Preferred language: [German, Italian or English]
Instructions: First, decide whether the image is informative, functional, or decorative. Do not add names, emotions, origins, intentions, diagnoses, relationships, locations, or details that are neither visible nor confirmed. Avoid repeating adjacent text. For a decorative image, only display alt="". For a functional image, describe the action or link destination. For a complex image, recommend adding a visible, longer description.
Output: Category, suggested alt text and a one-sentence explanation.
The prompt improves the AI design because it considers not only the image itself but also its function and context. However, the result remains a suggestion. Names, numbers, locations, and technical statements must always be verified against the original information.
Typical errors in AI image description
- Invented identities: The AI assigns a name or role to a person.
- Interpreted emotions: A facial expression is described as joy, fear, or motivation.
- Suspected origin: People, products, or places are associated with a region or culture.
- Assumed intentions: The AI claims what a person thinks, plans, or wants to achieve.
- Health-related attributions: Physical or psychological diagnoses are derived from the image.
- Incorrect details: Small labels, numbers, models or materials are supposedly recognized.
- Advertising language: Terms like "high-quality", "unique" or "sustainable" replace an objective description.
- Keyword filler: Search terms are written into the alt text without any substantive reason.
- Redundancy: Image caption, title, and alt text convey the same information multiple times.
- Incorrect category: A decorative image receives a long text, or a functional icon is only described visually.
Based on over 20 years of experience in photography, web design, branding, and technical implementation, I draw a clear conclusion: Good visual communication begins with the image's purpose, not its description. Without defining the image's function, reliable alt text generation using AI is impossible.
Alt text, image caption, file name and long description
- Alt text: The text alternative in the alt attribute conveys the purpose or relevant image information when the image is not perceived.
- Caption: The visible caption provides context for all readers, such as name, location, time, or source.
- File name: The filename serves technical organizational purposes and can describe the image objectively. It does not replace alt text.
- Title attribute: A title attribute is not a reliable substitute for the text alternative and should not contain any essential information.
- Long description: A visible long description explains complex diagrams, maps, or infographics in detail.
Alt text and captions can complement each other. If the caption already contains all the necessary information, the alt text should not repeat the same content. Instead, check whether a brief addition is needed or whether alt="" improves readability.
Multilingual alt texts for South Tyrolean websites
Multilingual alt text belongs to the respective language version of the page. German alt text on an Italian page can disrupt the reading flow of a screen reader and be mispronounced. Therefore, do not just translate individual words, but adapt the text alternative to the image context and the surrounding content of the respective language version.
Three rules apply to a German-Italian or trilingual website:
- Informative images: The text alternative is formulated in a way that is understandable and meaningful in each language of the page.
- Functional images: The action or link target is named in the respective language.
- Decorative images: alt="" remains empty in all language versions.
For names, product names, and place names, a literal translation is not advisable. Check which official name applies in the respective language. Small teams benefit from centralized maintenance where each language version is transparently approved. For more extensive content, we, as a collective of freelancers, can provide support with texts and multilingual editing.
Approval check for small editorial teams
Before publishing alt text using AI, a responsible person should answer seven questions:
- Need: Does the image need a full text alternative, or is alt="" correct?
- Purpose: Does the text convey the function of the image within the specific context of the page?
- Accuracy: Have all names, places, numbers, materials and visible details been confirmed?
- Redundancy: Does the alt text only repeat the title or the image caption?
- Language: Does the wording fit the language and tone of the page?
- Link function: Does a functional image describe the action or the goal rather than its appearance?
- Exam without image: Is the page understandable and usable if you mentally remove the image?
For diagrams, add an eighth question: Are the key message and relevant data accessible outside the image? If not, even precise alt text is insufficient.
Human approval combines automatic image recognition with editorial responsibility.
Legal context: no automatic compliance
The requirements of Directive (EU) 2019/882 have been applicable to the products and services covered by the Directive since 28 June 2025. Italy transposed the Directive by Legislative Decree No. 82 of 27 May 2022; its relevant provisions have been in force since 28 June 2025. The European Commission documents the Italian transposition and the related guidelines.
An automatically generated alt text does not automatically make a website legally compliant or fully accessible. Accessibility also encompasses structure, keyboard navigation, contrast, forms, media, understandable content, and technical semantics. This article is an editorial guide and not legal advice.
Questions and answers about alt texts
How long should alt text be?
There is no universally applicable character limit for every use case. Write as concisely as possible and as comprehensively as necessary so that the purpose and relevant information are clear without unnecessary details.
Should SEO keywords be included in the alt text?
The alt text primarily serves to convey the image's purpose. A technically relevant term may be included, but search terms without contextual necessity worsen the accessibility of image captions.
Can the filename replace the alt text?
No. A clear filename makes media management easier, but it does not replace a context-sensitive text alternative in the alt attribute.
Does a caption replace the alt text?
Not automatically. A caption is visible and can provide additional context; the alt text fulfills its own semantic function. Check both texts together to avoid unnecessary repetition.
Do logos need alt text?
An informative or linked logo generally requires an easily understandable name that includes the company name and, if applicable, the link destination. A simply repeated logo can be decorative depending on the image context, provided the name and function are already clearly recognizable.
How do I handle icons correctly?
For a button without a visible label, an accessible name must convey its function, such as "Open menu". If an equivalent label is directly next to it, screen readers can skip the icon itself using alt="".
Do AI-generated images need special alt text?
For the text alternative, the purpose of the image is what matters, not its creation. If the artificial creation is relevant for understanding, transparency, or editorial context, this information belongs in the visible image caption rather than automatically in every alt text.
How do multilingual alt texts work?
Each language version receives an alternative text in the language of the page. Multilingual alt text should be localized to reflect its meaning and checked within its respective context; decorative images retain alt="".
Can I fully automate all alt text?
AI can create drafts, highlight missing information, and accelerate reviews. However, the classification of the image's purpose, the verification of sensitive data, and final human approval should remain mandatory.
How do I check alt text without a screen reader?
Mentally remove the image and read only the page, caption, and alt text. If the purpose, information, and user interaction remain clear, the foundation is sound. A follow-up test with a real screen reader will then show whether the reading flow is smooth.
Conclusion: Creating and checking alt text with AI
AI-generated alt text saves small teams time if the process is clearly structured. The image's purpose comes before the description, confirmed facts before assumptions, and human approval before publication.
Technology takes over repetitive tasks. Context, judgment, and responsibility remain with humans. This results in accessible image captions that convey the content precisely and don't burden anyone with unnecessary or fabricated information.