Hi all,
Has anyone else noticed significant accessibility regressions in ChatGPT since the introduction of GPT-5.6?
On the ChatGPT website, I can no longer reliably access features such as Deep Research or Agent Mode using NVDA. The options appear under “Add files and more”, but they are announced more like static text than actionable controls.
Pressing Enter or Space on options such as Deep Research often activates the microphone instead of selecting the chosen feature. The same problem appears to affect other items in the menu, including Web Search, Create Image and connected services.
Typing commands such as /DeepResearch or /Agent into the message field does not appear to activate the actual dedicated mode. ChatGPT may respond to the wording, but it is not clear that the underlying tool or mode has been enabled. These capabilities may have been slowly depricated with the new "work" option. They are still referenced in Open AI'S help material, however.
I have tested this with NVDA in Chrome and am also trying Microsoft Edge. Standard screen-reader interaction methods, including Enter, Space and Insert+Space, have not resolved the problem.
There also appears to be an accessibility issue in the ChatGPT iOS app. The camera control or camera toggle is not reliably accessible with VoiceOver.
These problems make major ChatGPT features effectively unavailable to blind users, even though sighted users appear able to select them using a mouse. I have a plus account.
I would be interested to know:
1.
Whether other VoiceOver, NVDA or JAWS users are experiencing the same issues.
2.
Whether anyone has found a reliable keyboard or screen-reader workaround.
3.
Whether these regressions have already been reported to OpenAI.
4.
Whether Deep Research and Agent Mode have recently changed or been incorporated into another interface.
My current setup is Windows 11 with NVDA and Chrome, as well as the ChatGPT iOS app with VoiceOver.
By Darren12, 12 July, 2026
Forum
Windows
Comments
Have you tried simulating a mouse click?
Not sure about the desktop keyboard layout, but the NVDA laptop keyboard layout allows for a mouse click simulation by pressing the NBDA key plus left square bracket for a left mouse click, and NVDA plus right square bracket for a right mouse click. I have not tried this myself, but give it a try and see if simulating a mouse click will allow you to access those controls.
HTH.
Yes
I noticed this too. I use NVDA mouse click emulation to activate Deep Search for example.
I can confirm this
I can definitely confirm this issue.
While in browse mode NVDA can read all the items just fine but pressing space or enter on the item that NVDA has focus does not activate the focused item. For example my NVDA is focused on the upload file item in browse mode but pressing enter or space either does nothing, or activates a different item entirely.
On the new ChatGPT desktop app for Windows the menu that pops up when the add file button is activated is completely inaccessible with NVDA, even in browse mode or using object navigation.
Mouse simulation
Just tried simulating a mouse click, and that definitely is a workaround for the time being. It does indeed activate deep research, at least. Incredibly helpful. Thanks.
Chat GPT app.
Yes, the menu does seem invisible from within the Windows app. You can, however, upload a file if you hit insert space on the add files and more button and then tab, the upload file dialogue pops up. Other Options can't be accessed though. Probably best to stick to the web if you want to use deep research or the other features, using a simulated mouse click, for the meantime, until this is addressed. Disappointing. I've filed a bug report through the chat with Open AI, but unsure how effective that will be.