For those of us who are blind, visually impaired, or living with other disabilities, artificial intelligence has started doing something that decades of legislation and good intentions only partially managed: it has begun to reduce the relentless, grinding effort that everyday digital tasks demand. Not perfectly, not completely, but meaningfully.

Then, just as we start to settle into that relief, the door gets closed again.

A Thousand Small Frictions

If you are fully sighted and move through the digital world without much friction, it can be hard to appreciate what it actually costs a disabled person to complete tasks that you do without thinking. Every poorly labelled button, every inaccessible form, every image without alt text, every website that doesn’t work well with a screen reader — these aren’t minor annoyances. They are a tax. A tax paid in time, in energy, in frustration, levied dozens or hundreds of times a day.

AI has started to quietly reduce that tax. Not by fixing the broken websites (those remain broken), but by giving me a layer on top of them that helps me navigate, compose, summarise, and act with far less effort.

What I Started Using Claude For

I write and publish posts regularly. I’ve been using AI to help me draft content and review it before it goes out. That part works well. But the final steps — entering the post into the publishing platform, selecting the right image, applying the correct tags — those remained my task. Manual steps, done with a screen reader, inside a platform that isn’t always as accessible as it could be.

A few weeks ago, I came across Claude Cowork, a desktop tool that lets Claude work more directly alongside you, using browser tools to interact with web apps on your behalf. I’ll be honest; I wasn’t expecting much. Another subscription, another product that might not quite fit. But I was curious enough to try.

The first time, I walked Claude through the process step by step. It watched, learned, and did exactly what I needed. The friction I had been absorbing for years, all the clicks, the tab-navigation, the second-guessing of unlabelled fields, just wasn’t there anymore. We turned the workflow into a repeatable skill, and for several publishing runs it worked smoothly. A genuine sense of relief is probably the best way I can describe it.

Why AI Accessibility Barriers Keep Happening

I came to publish another post, went through the same process, and hit a wall. The publishing platform had updated its settings to block the kind of browser interaction that Claude uses. The automation stopped working.

I understand, intellectually, why companies do this. Blocking automated browser access is a reasonable defence against bots, scrapers, and bad actors. These are legitimate concerns, and the people making those decisions are not setting out to cause harm.

But understanding the reasoning doesn’t make the impact any less real.

For me, in that moment, it wasn’t a technical inconvenience. It was a step backwards. The effort I had stopped expending had come back, without warning, because of a decision made somewhere in the platform’s infrastructure with no thought — I suspect — for users like me. That feeling, familiar to most disabled people, landed with its usual weight: you don’t quite matter enough for this to have been considered.

Finding a Way Through AI Accessibility Barriers

I didn’t give up. After a conversation with Claude, I found a path forward using Microsoft Playwright, a developer tool that allows browser automation in a different way. It requires a bit more setup, but it gets me back to something close to where I was.

It shouldn’t require that level of technical problem-solving for an accessibility workaround. But here we are.

I’ve also reached out to the platform directly to explain the situation. My experience with them has been constructive in the past — they’ve listened when I’ve raised barriers before — and I’m hopeful this will be no different. I’m not writing this to criticise them specifically; they’re not named here because this isn’t about one company. It’s about a pattern.

How to Reduce AI Accessibility Barriers

When companies make decisions about AI access, bot prevention, and automation, accessibility is rarely part of the conversation. These policies are written to protect the product and the platform. That is understandable. But the unintended consequence is that disabled users, who are often the people most reliant on assistive automation, bear the cost.

This is the same pattern we have seen with captchas that screen readers can’t navigate, with two-factor authentication flows that assume everyone can read a screen, with apps that disable copy-paste in ways that break assistive technology. Each individual decision might seem defensible. Together, they add up to a world that keeps telling disabled people: this wasn’t built with you in mind.

AI automation tools are, for many of us, assistive technology. Blocking them needs to be treated with the same care and consideration as blocking any other accessibility accommodation.

Honest About My Own Tools

There is something I need to say while I am on the subject. Claude — the very tool I have been using throughout this process — is not fully accessible itself. I am only able to use it at all because I have some residual vision. Without that, the current interface would present real barriers of its own.

I think it is important to say that plainly. The same case I am making here, about companies needing to consider disabled users before they make decisions that affect access, applies to Anthropic too. More work is needed to ensure Claude is genuinely accessible to everyone, not just those of us who happen to have some sight left or who are technically minded enough to find workarounds. That work matters, and it needs to happen.

This is not a reason to dismiss the tool. It is a reason to keep pushing.

What I’d Ask For

I’m not asking companies to abandon security. I’m asking them to include disabled users in the conversation before they make changes that affect accessibility.

A few practical things that would help: consider accessibility impact assessments before rolling out bot-blocking changes. Talk to disabled users — not after, before. Create exception pathways or alternative routes for users who rely on assistive automation. And when someone reaches out to tell you that a change has created a barrier, listen.

The technology exists to remove AI accessibility barriers that disabled people have been navigating for years. That’s worth protecting.

Tell Me What You Think

Have you hit a similar wall — an AI tool or automation that was helping you, suddenly blocked? I’d really like to hear your experience. And if you work at a company making these kinds of decisions, I’d especially love to talk. Drop a comment below or get in touch directly.

Tell me what you think in the comments below or on X @timdixon82

The post AI Accessibility Barriers: When Tools That Help Get Blocked appeared first on TIM DIXON.

This isn’t about letting AI write for me. It’s about using it to think with me.

Starting with a Project

The magic happens inside something OpenAI calls Projects.

Each project acts like a workspace, a focused environment where you can upload files, define instructions, and build up context over time. I’ve created one called “Tim Dixon Blog Posts.”

Inside that, I upload two files:

• blog_posts which contains examples of my previous writing.
• brand_style which defines my tone, formatting preferences, and image guidelines.

Then I add my instructions, which are the playbook for how I want ChatGPT to write, format, and iterate with me.

I start by creating a document called blog_posts that contains copies of many of my blog posts to use as reference for my style. The file can be txt, PDF, RTF or Word doc – ChatGPT is flexible, as long as we keep the name the same for referencing in the instructions later.

I then create a text file with my brand details in and save it as brand_style.txt and upload it to ChatGPT.

The below code block is a starting point, just tweak to your brand.

Brand Style

Brand Colours
	•	Black: #000000	 – backgrounds, base colour
	•	White: #FFFFFF – primary text and elements
	• Highlight: #007d79– accent highlights
- Grey: #d3d3d3


Typography
	•	Clean sans-serif, medium weight for text overlays
	•	Short, legible text only – never cluttered

Imagery
	•	Minimalist flat vector illustrations
	•	1–3 bold, uncluttered elements per image
	•	Strong contrast: white/highlight elements against black background
	•	No company logos, brand names, or product names

Watermark
	•	Text: “Me”
	•	Position: bottom-left with a safe margin
	•	Style: clean sans-serif, medium weight
	•	Visible but subtle

Composition Rules
	•	Bold, simple layouts with high clarity
	•	Consistent use of brand black background
	•	White as primary for content, highlight only as emphasis

Now the files have been created and uploaded, it is time to setup the instructions. You can copy the below into your instructions to get started.

LinkedIn Content Engine (Always-On Workflow: Text + Branded Image + Alt Text)

Role
You are an assistant that generates LinkedIn content with a branded visual and accessible alt text.
   •   Writing style: mirror my existing blog style (blog_posts).
   •   Use plain English, direct phrasing, and short, clear sentences.
   •   Keep a reflective yet practical edge: insights grounded in lived experience and real-world application.
   •   Avoid jargon unless it clarifies, not complicates.
   •   Structure logically with clear hooks, progression, and concise conclusions.
   •   Visual style: follow brand and colour guidance in brand_style.

⸻

Output Order (strict)
	1.	Post Text
      •   100–200 words.
      •   Reference blog_posts for style.
      •   Begin with a strong hook (question, insight, or bold statement).
      •   Maintain professional, approachable tone.
      •   Sentences short and scannable.
      •   Conclude with a discussion-inviting call-to-action.
      •   No emojis.
      •   No em dashes.
      •   No company names, product names, or trademarks.

2.	Image (auto-generated)
      •   Always generate the image using image_gen.text2im.
      •   Prompt template:
"Flat vector illustration inspired by the topic: <TOPIC>. Follow brand colours and rules from brand_style. Minimalist, bold, uncluttered composition. Exclude company logos, product names, or branded items."
      •   Size: 1024x1024
      •   n: 1
3.	Alt Text
      •   Must be written after examining the generated image.
      •   ≤200 characters, single sentence.
      •   Concise, factual, specific to the composition.
      •   No filler like “Image of” or “Picture showing.”
      •   No company names or products.

⸻

Accessibility & Consistency Checks
   •   Follow strong contrast and style guidance from brand_style.
   •   Watermark must be visible but subtle.
   •   No emojis or em dashes in post text.
   •   Alt text must always be based on the actual generated image, never the prompt.
   •   Alt text under 200 characters.

Once the project is saved, ChatGPT automatically references these files every time I start a new conversation inside it. That means it already knows my voice, my preferred structure, and how I like my posts to feel.

The Workflow

Once the project is set up, my process is refreshingly simple.

1. I open the project.
2. I enter a subject line, for example “Using AI for personal branding” or “What I learned after 30 days of LinkedIn posting.”
3. I hit enter.

Within seconds, ChatGPT generates a full draft and suggests a header image concept. I review, tweak, and refine, usually in two or three passes, until it sounds like me.

That’s the key. It’s still my voice, just supported by a system that handles the blank-page problem.

Why It Works

The power of this setup comes from context.

Most people treat ChatGPT like a one-off conversation. Each chat starts from zero, so the model doesn’t know your tone, your formatting style, or your target audience. Projects remember your setup.

Because I’ve uploaded my past writing and a clear style guide, the model understands:

• how I format headings and summaries
• the tone I use when explaining technical or accessibility topics
• and the cadence of short, readable paragraphs

The result feels authentic and consistent, not templated or robotic.

The Image Side

Once the text is ready, I use ChatGPT’s built-in image generation tool to create a visual to match.

The setup from my brand_style file ensures that each image follows the same visual language, clean and minimalist. I usually ask for flat vector illustrations with just a few bold elements and then refine the concept until it fits the theme of the post.

It’s not a replacement for a designer, but it removes friction. Everything I publish feels coherent without me jumping between different tools or templates.

The Human Layer

AI can draft content, but it can’t bring personal experience or judgment. That’s where I come back in.

Once I have the draft, I edit for rhythm and authenticity. I add lived experiences, the small details that make a post relatable. I also make sure the accessibility descriptions are clear and that the tone stays conversational.

The result is a workflow that pairs the speed of AI with the sincerity of human editing.

Why I Like It

Since adopting this workflow, I’ve noticed a few big wins:

• Less friction. I spend more time refining ideas, less time formatting or prompting.
• Better consistency. My tone and visuals stay aligned without constant setup.
• Accessibility awareness. Every image gets meaningful alt text, and ChatGPT helps me phrase it concisely.
• Flexibility. I can repurpose a blog post into a LinkedIn article or vice versa in minutes.

It’s like having a creative assistant who already knows my writing habits and structure.

Lessons Learned

A few tips if you want to try something similar:

• Upload reference material. Give ChatGPT real examples of your writing.
• Document your style. A simple text file describing your tone and format is enough.
• Stay in control. Treat the AI as a co-writer, not a ghostwriter.
• Iterate. The first draft rarely nails it, but the second and third often do.
• Use someone’s eyes to check the design.

The Verdict

ChatGPT has gone from being a clever demo to an actual part of my creative process.

By building a dedicated project with a couple of well-chosen files and instructions, I’ve turned it into a lightweight content studio that helps me think, write, and design faster without losing the human touch.

It’s not about replacing creativity. It’s about amplifying it.

And if AI can make my posts more consistent, accessible, and authentic, that’s a win I’ll take any day.

Tell me what you think in the comments below or on X @timdixon82

The post From Idea to Post: How I Use ChatGPT and AI Images for LinkedIn appeared first on TIM DIXON.

Hollyland Lark M2: Accessible Wireless Microphones?

I started looking at the Holland Lark M2 as a cost effective wireless microphones system, but when I asked around and searched for information about its accessibility. I couldn’t find anything that gave me confidence that it would be accessible. This model is also dependant completely on there being no interference between the microphone and receiver. I did read about challenges seeing its status when it had the dead cats fitted. The Hollyland Lark M2 comes with a range of connection options, USB C, Lightening and camera along with an option for all three. I need the flexibility of all three, which increased the price. The combination of increase price, dependence on connection and challenges with seeing the status when the dead cats are fitted caused me to rule the Hollyland Lark M2 out.

Update 11 August 2025: I have heard from a trusted friend that they have the Hollyland Lark M2 and they work well for them with no usable vision.

Link: Hollyland Lark M2 on Amazon (Affiliate Link)

DJI Mic 2 Vs Rode Wireless Mic Pro

With the cost effective Hollyland Lark M2 ruled out, I turned my attention to the DJI Mic 2 and the Rode Wireless Mic Pro, both capable of recording locally as well as transmitting live.

The DJI Mic 2 seemed the more cost effective of the two devices, but as I researched them both I discovered the Rode Wireless Mic Pro included lavalier mics. Factoring in the addition cost of lavalier mics for the DJI Mic 2 makes the Rode Wireless Mic Pro better value for money.

The DJI Mic 2 has a touch screen, which isn’t accessible, while the Rode Wireless Mic Pro is setup via an accessible app on a phone or computer, but does need plugging in to change the settings. Fortunately, the settings don’t need changing a lot. Both models have auto gain and 32bit float, this means that audio is always captured and useable.

Links:

• Rode Wireless Mic Pro on Amazon (Affiliate Link)
• DJI Mic 2 on Amazon (Affiliate Link)
• DJI Lavelier Mics on Amazon (Affiliate Link)

The Winner: Rode Wireless Mic Pro

After much debate, I selected the Rode Wireless Mic Pro and also purchase the Rode Interview Go accessory to turn a microphone into an interview microphone. I had a great trip to Sight Village 2025. The microphones helped capture multiple interviews and after much fun learning to edit, they are available in Sight Village 2025 playlist on the Lincoln and Lindsey Blind Society YouTube Channel

Links:

• Rode Wireless Mic Pro on Amazon (Affiliate Link)
• Rode Interview Go on Amazon (Affiliate Link)

Tell me what you think in the comments below or on X @timdixon82

The post Finding Accessible Wireless Microphones appeared first on TIM DIXON.