Thanks Sunny, appreciate the clarification and the more detailed response.

However, I still think the central concern remains unresolved.

Even if alt text is only generated when an image is placed or replaced, the issue is still that AI-generated text is being added to production documents by default. That is the part many of us are objecting to.

Improving the quality is welcome, but quality improvement does not solve the consent/workflow problem. A better guess is still a guess. In professional workflows, alt text is not harmless background metadata. It is document content. It can be exported, read by assistive technology, reviewed by clients, included in accessibility audits, translated, archived, and carried through regulated or approval-sensitive workflows.

That means it needs the same level of user control as any other content placed into a document.

I also want to push back slightly on the idea that improved handling of icons and graphics “directly addresses” the inaccurate descriptions reported. It may help, but it does not eliminate the risk. We have already seen examples where a keyline/technical element was described as something like a tall blue building, and other users have reported wildly inappropriate or NSFW-style descriptions for graphics. That is not accessibility. That is inaccurate content being inserted under the banner of accessibility.

Making documents more accessible is a good goal. Nobody is arguing against that. But making documents appear accessible by automatically inserting unreliable descriptions is not the same thing. In fact, it can make things worse, because users may believe the accessibility work has been handled when the descriptions are actually wrong, misleading, irrelevant, or inappropriate.

The enterprise admin control is useful, but it does not address individual users, freelancers, small studios, education users, charities, smaller publishers, or corporate users without centralised admin control. Regulated and approval-sensitive work does not only happen inside large enterprise deployments.

The safer design is still:

Gen Alt Text off by default.

Offer to generate alt text when needed.

Allow generation for selected images or all images.

Require user review/approval before generated descriptions become part of the file.

Provide a reliable way to find, audit, edit, or remove generated alt text already added.

I appreciate that 21.4 improves the feature, but this UserVoice request was not only about accuracy. It was about default behaviour, consent, review, reversibility, and trust.

Until AI-generated content is opt-in and reviewable before insertion, I do not think this should be considered resolved.

Related thread:
https://community.adobe.com/questions-671/indesign-generative-ai-options-on-by-default-1561085?tid=1561085&postid=7672456#post7672456

Thank you for the detailed feedback everyone. A few specifics worth clarifying:
Alt text is generated when an image is placed or replaced - not when an existing document is opened. If alt text is appearing in a file you only opened without placing anything, that's a bug we'd want to investigate directly.

Updates in 21.4:
- Improved alt text quality across image types, including better handling of icons and graphics - this directly addresses the inaccurate descriptions reported.
- Fixed an issue where cropped image alt text wasn't generating correctly
- Enterprise admins can now enable or disable this feature from the admin console, built for regulated and approval-sensitive environments.

We know that quality and trust in the output are what make this feature worth using. The 21.4 improvements are a step, and raising accuracy will remain a priority.

Another aspect being overlooked here is document integrity.

Traditionally, opening an InDesign file should not materially alter its content unless the user intentionally edits something. With this feature enabled by default, simply opening legacy documents can now result in machine-generated text being added into production assets.

That changes the relationship users have with the software itself.

For regulated or archived workflows, “opening a document” and “modifying a document” are very different things.

From the home page: "This is a democratized platform and even if you may not have any suggestion, your vote counts." I don't feel like our vote is counting.

This issue is not resolved for any user. A one time CTA alert for a business threatening feature no one asked to have turned on by default is just as effective as a post it note already thrown away in the trash can.

A bad design is a bad design is a bad design. Intentionally designed or not.

As users have clearly stated and Eugene shared from the beginning, this is a massive red flag to have this turned on by default. Especially with the horrifying results already reported by many users in the community forum.

Definitely not resolved.

This feature is poorly designed and even more poorly thought out; there is no process of informing the user when AI text is added to every image in an old document they opened. There is also no process to revert that.

What an expected yet still disappointing response. The entire point of this thread is that feature that is "as designed" sucks and can harm users and businesses.

I agree, this is not resolved.
Adobe should NEVER change customer data by default.

This is not resolved.

Thanks for this great example how to not communicate with the community, especially with power users who are wholehearted fans of the software.

Even the category »Resolved for user« is a punch in the face.

The feature needs to be turned OFF.
The reasons for that have already been mentioned by Eugene Tyson.

I agree with all the comments: The Ravi's response is disrespectful to a well argued suggestion, and disregards the fact that this automatically generated material is generated on older documents as they are opened, before any image is placed.

The response ignores some of the undeniable problems with this approach and
makes the term "uservoice" sound ironic.

That’s a tone deaf reply. The issue isn’t whether there was a checkbox or a CTA once during onboarding. The issue is that AI-generated content is being automatically injected into professional production documents by default — including regulated, government, accessibility, legal, and corporate workflows — and it has already been shown generating NSFW and wildly inaccurate descriptions.

“Users were informed once” is not really an adequate defence when the feature can create reputational or compliance risks months later inside files people may never manually review line by line.

AI alt text as an assistive starting point? Fine. AI silently writing production content by default? That’s the part people are objecting to.

Yes. As designed. That is exactly the problem: BAD designed! Turn it off by default should be the design 😡😡

As somebody who does research reports and just discovered that Adobe is inserting completely inaccurate alt text to describe our figures (without our consent), I could not agree more with this. What a terrible user experience, terrible use of AI, terrible for accessibility. Please, please, please make this all Opt-In. The worst part is the place where I'm supposedly able to disable it doesn't exist for me in InDesign. UGGGGGG

100% agree. Text should never be changed without agreement and opt in. Default should be opt out.

Dave

Unwanted feature turned on by default is a critical miss by Adobe. Do better Adobe. There are those who have no need for these AI features.

Fully agreed, I am in a corporate design position (non-legal) and unwanted AI alt-tags would cause big issues in our approval process and even bigger if they got published unknowingly.

I can see the benefit here, but it is only positive when prompted by the user/toggled off by default. Otherwise it feels like a rogue tool with potential to add hours to my workflow due to fixing its' mistakes.

100% agree!

(There is no need by the way for any of this AI slop, but at least it should be off by default!)