Success using Gemini to master Gmail Web with NVDA

[Mister Kayne]

Hello everyone,

I’m writing to share a recent experience that I think might be helpful, especially following some previous skepticism regarding AI-generated advice on this list.

For the first time today, I had to use Gmail on the web to create a specific filter. Being a long-time Outlook user, I wasn't familiar with the web interface, so I decided to use Gemini to guide me. I asked for the specific settings an NVDA user needs to enable or disable for the best experience, and it provided a perfect step-by-step list—covering things like enabling keyboard shortcuts, setting density to compact, and simplifying the inbox view.

Once the settings were dialed in, I asked Gemini for instructions on creating a filter. The guidance was spot on and worked the first time. I even spent some time "playing" with the interface—using keyboard shortcuts to compose, reply, forward, and navigate the inbox. In fact, I am composing this very email using Gmail on the web instead of my usual Outlook client!

I’m sharing this because I know some members were critical when a Gemini prompt was suggested to help with a challenge previously. While I understand the value of human experience, I found that using the AI saved me a significant amount of time compared to posting a question to the list and waiting for a response.

If someone suggests a prompt or an AI-generated solution in the future, I’d encourage us not to be so quick to criticize. For me, it was a highly effective tool that got me up and running independently in minutes.

--

Sincerely;

Mujtaba Merchant
Bangalore, INDIA
Sent from Gmail Web Mail Services

[Gene Asner]

I've looked for quite a bit of how to information using AI bots. I find
that a lot of the time, perhaps as much as 75 or 80 percent, the
information is either wrong or has errors which make it either unusable
or which I have to figure out in trying to apply it. But there are
times when you do get good information and when, even though the
information has serious problems, you can play around and figure out
what the bot should have told you and then apply it. I say to be
skeptical of the advice you get from an AI bot about how do you in
computer-related topics but at times it is useful.

While I don't advocate people rushing to send AI bot information of this
type to lists, I do think that after a good deal of time has passed with
no information or not accurate information being sent, that sending bot
information may be helpful. Also, I have found that if someone asks a
question and there is no response for a good while, say twelve or more
hours, sending information may stimulate discussion. So even if the bot
information is wrong, it may stimulate others to send accurate information.

Gene

> --
> ***
> Please note: the NVDA project has a Citizen and Contributor Code of Conduct.
> NV Access expects that all community members will read and abide by the
> rules set out in this document while participating in this group.
> https://github.com/nvaccess/nvda/blob/master/CODE_OF_CONDUCT.md
> <https://github.com/nvaccess/nvda/blob/master/CODE_OF_CONDUCT.md>
>
> You can contact the group owners and moderators via
> nvda-user...@nvaccess.org.
> ---
> You received this message because you are subscribed to the Google
> Groups "NVDA Screen Reader Discussion" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to nvda-users+...@nvaccess.org
> <mailto:nvda-users+...@nvaccess.org>.
> To view this discussion visit
> https://groups.google.com/a/nvaccess.org/d/msgid/nvda-users/CALR5w2dvLzRh%2BV7Hg7xCVrOQXDfBsUAvF59mG%3DukP_Qes4KHmg%40mail.gmail.com
> <https://groups.google.com/a/nvaccess.org/d/msgid/nvda-users/CALR5w2dvLzRh%2BV7Hg7xCVrOQXDfBsUAvF59mG%3DukP_Qes4KHmg%40mail.gmail.com?utm_medium=email&utm_source=footer>.

[Sean Randall]

on the other hand, the questions you put to Google's AI were about one of Google's services. there would have been plenty of training data available to the model. The real concern for training is, the less accurate data a model has about a subject, the more confident it will sound about its proposed offering, even when we know it's wrong. 

I'm glad you had success in this instance and can see in some circumstances these tools can be very useful. Yesterday's case was not one such: a list user without access to the software the questioner was struggling with copy and pasted the post into a model, sending its unvetted reply back with the hope of being helpful.  That and what you've done today are not comparable scenarios from which to base a comparison nor form a judgment as to an AI's suitability  to help on a list like this. 

  

Thanks

Sean

On 22 Feb 2026, at 05:25, Mister Kayne <mujt...@gmail.com> wrote:



--
***
Please note: the NVDA project has a Citizen and Contributor Code of Conduct.
NV Access expects that all community members will read and abide by the rules set out in this document while participating in this group.
https://github.com/nvaccess/nvda/blob/master/CODE_OF_CONDUCT.md
 

You can contact the group owners and moderators via nvda-user...@nvaccess.org.
---
You received this message because you are subscribed to the Google Groups "NVDA Screen Reader Discussion" group.

To unsubscribe from this group and stop receiving emails from it, send an email to nvda-users+...@nvaccess.org.
To view this discussion visit https://groups.google.com/a/nvaccess.org/d/msgid/nvda-users/CALR5w2dvLzRh%2BV7Hg7xCVrOQXDfBsUAvF59mG%3DukP_Qes4KHmg%40mail.gmail.com.

[Mujtaba Merchant]

Hi Gene,

Thanks for your thoughtful response. I completely understand your apprehension regarding AI hallucinations; it’s definitely a valid concern, and I agree that a healthy dose of skepticism is necessary when dealing with AI-generated technical advice.

My logic in this specific instance was quite simple: since both Gmail and Gemini are Google products, I figured Gemini would have the most direct and up-to-date "internal" knowledge of Gmail’s settings and accessibility hooks. It felt like asking the manufacturer for the manual rather than a third party.

I also appreciate your point about using AI to stimulate discussion. Even if a bot gets the details slightly wrong, having a starting point can often prompt a "good soul" on the list to jump in with a correction or a more refined method. It gets the gears turning much faster than a silent thread.
In this case, the efficiency of getting an immediate, working answer was a huge win for my productivity. I'm glad we can have this balanced conversation about where AI fits into our toolkit—not as a total replacement for human expertise, but as a useful first line of defense when used carefully.

To unsubscribe from this group and stop receiving emails from it, send an email to nvda-users+...@nvaccess.org.
To view this discussion visit https://groups.google.com/a/nvaccess.org/d/msgid/nvda-users/e860c8b8-d228-947d-e4c5-9467fe45de27%40gmail.com.

[Gene Asner]

That is not correct and no such inferences or assumptions should be
made. I won't respond further in this thread because such discussion as
you raised and I am responding to should be on a chat list or a general
blindness technology list. but I'll comment this time. I got the best
answer to the unread question from Copilot. I used it as a test in
different bots.

Also, bots change in quality, sometimes dramatically over time. Some
get better, some get worse, and they may change again over time. I use
a bot that is the best at the moment and if I find it declining, I try
and compare others. Sticking with one bot just because you like it at
the moment is not a good way to proceed if you see it declining. Gemini
has declined significantly over the last few weeks and I don't use it
any more. I may later if it improves and others get worse. But if you
ask bots about whether Gemini is declining and if they have information
about it, the better ones will tell you that there is discussion in user
forms about the decline.

Copilot, when I asked the question gave much better information.

It is still wrong a lot in such questions but it is right at least a
little more.

Gene

[Steve Nutt]

Wow, my experience is quite different to yours.
I often get right answers with Gemini and am generally happy with it, but it is not my only tool, and I do try to verify what I read.
All the best

Steve

--
Computer Room Services
77 Exeter Close, Stevenage, Hertfordshire, SG1 4PW
T: +44(0)1438-742286, M: +44(0)7956-334938
E: st...@comproom.co.uk, W: https://www.comproom.co.uk

To view this discussion visit https://groups.google.com/a/nvaccess.org/d/msgid/nvda-users/3bacbe35-9467-763b-cf02-6d275a6384c7%40gmail.com.

[Gene Asner]

Bots are better at different things and we don't know what kinds of
things you ask.

But here is a Reddit thread demonstrating what I said. Gemini has
gotten dramatically worse.
https://www.reddit.com/r/GeminiAI/comments/1r5jdwr/gemini_decline/

Gene
Gene

[Steve Nutt]

Let's keep this balanced though eh?
One of the comments on that very Reddit thread that you provided was the following:-

I’ve been seeing a lot of these ‘regression’ posts lately, but I have to say, my experience has been the exact opposite.

I’ve found that the results I get are strictly tied to the partnership I’ve built with the model. Instead of treating it like a magic search bar that should read my mind, I treat it like a high-level collaborator.

I’ve invested the time to provide direct, clear instructions (in the settings) with specific protocols for how I want it to handle my inputs.

When I’m clear about the constraints and the 'Master Rules' for our interaction, it doesn't regress; it actually gets sharper because the context is grounded. If you're getting 'lazy' or 'backwards' results, try being more explicit about your expectations and building a persistent instruction set.

For me, it’s been a game changer.

** End Of Reddit Comment **

I am getting the same result. It's how you frame your question.

To view this discussion visit https://groups.google.com/a/nvaccess.org/d/msgid/nvda-users/3830e74f-dcca-5e3f-7f21-507fedb62b60%40gmail.com.