DEV Community: Raymond

The Ultimate Guide to Generative Engine Optimization (GEO): SEO for the AI Era

Raymond — Thu, 24 Sep 2026 09:37:36 +0000

If you've been making websites or producing content over the past ten years, then you'll be familiar with the process: choose a keyword, write a long-form article, get some backlinks, and then hope that the Google algorithm grants your site a position among the top 'ten blue links'.

However, the internet is experiencing a major change in its structure since people are now going around the traditional search engines and are instead asking questions directly of AI tools such as ChatGPT, Perplexity, Google Gemini, and Microsoft Copilot.

They are not traditional search engines; instead, they are Answer Engines, and in order to optimize for them a totally new approach known as Generative Engine Optimization (GEO) or Answer Engine Optimization (AEO) is needed.

If you want ChatGPT to recommend your web development clients (or your own TechMitten projects) rather than your competitors, then you need to understand how these AI systems browse the web. Let's look at the workings of AI search and examine how you can design your site for the AI age.

How AI Answer Engines Actually Work

If you want an AI to recommend your business, then you need to understand the way in which it obtains information. AI models have two methods of "knowing" things:

Parametric Memory: This makes up the AI's internal brain; it is the static data which the AI acquired during its initial training (for example, from sources such as Wikipedia and through large-scale data scraping). In the case of a brand that was not well known before the AI's training cut-off date, it will not know who you are.
Retrieval-Augmented Generation (RAG): This is the method by which modern AI links with the live internet. For example, if a user asks ChatGPT to name the best web developers in Michigan, the AI does not depend on its own memory. Instead, it stops, carries out a live web search, reads through the top search results, and then produces a tailored response.

In order for your website to be cited by an AI, you have to win the RAG pipeline.

The AI Speed Limit: Why Web Performance Is Critical

When an AI bot (for example, OpenAI's search crawler) is looking for answers, it assesses around 40 to 50 URLs and then attempts to retrieve the raw HTML from the most promising ones.

The catch is that the AI allows only two seconds per page.

The AI will drop your page before it has even read one word if your origin server is slow, if you have complicated redirect chains, or if your host is slow. Moreover, bots such as OAI-SearchBot parse the raw HTML; they don't execute client-side JavaScript.

Lesson for web development: When you create Single-Page Applications (SPAs) using React, Angular, or Vue without incorporating hydration, the AI bot will simply see a blank page. Server-Side Rendering (SSR) or Static Site Generation (SSG) is not merely a performance advantage any longer; it has become a necessary requirement for achieving AI visibility.

Writing for AI: The 128-Token Rule

Generative models do not read web pages in the way that humans do, nor do they rank whole pages as Google does; instead, they break your document down into small sections (typically about 128 tokens) and then assess those particular sections for relevance.

A recent academic study (known as the "GEO-bench" study) looked at 10,000 queries and discovered that if you make intentional changes to the structure of your content it is possible to increase your visibility in AI responses by as much as 40%.

Here is how to optimize your content for AI bots:

1. Apply the Inverted Pyramid

Don't hide the answer at the bottom of the page since research has found that 44.2% of all AI citations are taken directly from the first 30% of a web page.

Under each of your main headings, provide a direct and concise answer (about 40 to 60 words). This effectively packs your answer into one highly scannable 128-token segment.

2. Introduce Specific Data and Statistics

AI models prefer verifiable facts and, if you include empirical figures, percentages, and financial metrics once every 100 words in your informative copy, your AI visibility can increase by as much as 41%.

3. Make Use of Quoted Material and Attributable Sources

Statements in the field such as "We are the best in the business" are ignored. AI systems require authoritative evidence. Visibility can be increased by 40% by including inline source attribution (that is, linking your claims to primary research), and an additional 30% in visibility can be achieved by incorporating direct quotes from named subject matter experts.

The Technical Setup: Directing the Bots

As a web developer you have complete control over the way AI bots interact with your server, but it's easy to remove your site from AI search because of a simple error.

The robots.txt Pitfall

Since many businesses are frightened that AI will take their intellectual property, they block AI bots in their robots.txt file; but it is important to make a distinction between training bots and search bots.

Bots used for training (for example, GPTBot): These work by scraping the web in order to train future AI models; you can safely block them if you want to protect your IP.
Bots such as OAI-SearchBot and PerplexityBot: These are the RAG bots which search for up-to-date answers to user queries. If you block them, you'll not show up in the search results on ChatGPT or Perplexity.

Tip: When you specifically permit OAI-SearchBot, be sure to also repeat your disallow rules for sensitive directories (for example, /admin/), since the bot will disregard the global wildcard rules the moment it comes across its own specific name.

Enter llms.txt

Just as there is a robots.txt file for giving permissions and a sitemap.xml file for URLs, so too has the AI era brought about an llms.txt file.

It's a clean, machine-parsable Markdown file that is located at the root of your domain (for example, at techmitten.com/llms.txt). The file gets rid of all the navigational clutter, popups, and layout formatting, thus giving AI reasoning models a clear text outline of what your business does, who it serves, and where your main documentation can be found. So add it to your site now.

Entity Grounding: Proving You Exist

AI does not reason in terms of keywords; instead it works with Entities (which are distinct people, places, organisations, or concepts). If the AI is unable to confirm that your business is a real and legitimate entity then it will automatically suggest your competitors.

To prove you are real, you need to establish a solid "knowledge graph" presence:

Use Schema.org JSON-LD. Place the Organization schema in the <head> section of your homepage and make sure that the @id URI is included, using the sameAs array to link your website directly to your official Wikidata item, Crunchbase, LinkedIn and your social profiles.
Declare your expertise. Make it clear to the AI which topics and technical fields you are qualified to discuss by using the knowsAbout property in your schema.
Keep your NAP consistent. For local businesses, ChatGPT makes heavy use of Yelp, whereas Microsoft Copilot depends on Bing Places and Facebook. The name, address, and phone number (NAP) must be exactly the same on all these platforms. The AI doesn't make guesses; if the address of your business differs on Yelp and Google, then the AI will reject you as a possible match.

Off-Page SEO: Mentions Over Backlinks

In traditional SEO the hyperlink held the position of king, but in Generative Engine Optimization unlinked brand mentions are now taking over that role.

Evaluation based on 75,000 brands reveals that unlinked references to a brand are related three times as strongly to AI visibility as traditional backlinks. Moreover, 82% of all AI citations are from earned third-party media (for example, digital public relations, industry news, and discussions on Reddit), whereas just 6% come from the company's own website.

The AI will not believe a major claim that you make on your website unless other credible sites are also talking about it; therefore, your public relations should centre on getting your brand name cited in reputable industry publications, even if they don't provide you with a hyperlink.

How to Track AI Traffic in GA4

A major source of annoyance with AI search is the fact that it functions as a "zero-click" environment: users obtain their answers without having to go to your website. Furthermore, when they do click through, mobile apps tend to remove the referral data, which causes Google Analytics 4 (GA4) to treat AI traffic as general "Direct" traffic.

In order to fix this, you have to create a Custom Channel Group in GA4.

In the GA4 Admin section, go to Data Display and set up a rule based on a regular expression, positioning it above the standard Referral channel. Use this precise RE2-compliant regular expression to capture traffic from AI:

^.*(chatgpt\.com|chat\.openai\.com|openai\.com|claude\.ai|anthropic\.com|perplexity\.ai|gemini\.google\.com|bard\.google\.com|copilot\.microsoft\.com|bing\.com\/chat|grok\.com|x\.ai|grok\.x\.com).*

You'll probably be amazed when you filter this traffic since, even though the amount of AI search is less than that of Google organic search, the conversion rate is usually much higher because the AI has already pre-qualified the user's intent.

Your 90-Day GEO Action Plan

If you'd like to optimize your site for ChatGPT and Perplexity, then follow this simple roadmap:

Days 1–30: The Tech Fix

Check your robots.txt file to make sure that OAI-SearchBot, Claude-SearchBot, and PerplexityBot are permitted.
Audit your server response times (keep them below 2 seconds!).
Implement Server-Side Rendering if it's necessary.
Create a file called llms.txt.
Set up your GA4 regex filter.

Days 31–60: Content Re-Architecture

Rewrite the top 30% of your service pages by including a direct summary of about 50 words right at the beginning.
Add some hard statistics and check your claims against reliable sources.
Incorporate comprehensive Organization schema into your headers.

Days 61–90: Off-Page and PR Efforts

Audit the listings of your local directory to make sure the NAP information is consistent.
Begin carrying out digital PR campaigns with the aim of getting your brand name mentioned on industry blogs and forums such as Reddit.

The move to Answer Engines has already taken place, and by using these Generative Engine Optimization techniques you can make it so that when the AI is asked to provide the best web solutions it will directly point to you.

Welcome to TechRay.dev: Why I'm Writing About AI, Agents, and the Future of Building Software

Raymond — Thu, 24 Sep 2026 00:05:56 +0000

Hi, I'm Raymond Busuttil, and this is the first post on my new corner of the internet.

I've wanted a place like this for a while. Tech moves fast, and the last couple of years have felt like a decade of change compressed into a few seasons. I spend a lot of time reading, tinkering, and building with the latest tools, and I kept catching myself explaining the same things to friends and colleagues over and over. A personal site felt like the obvious next step: a place to think out loud, share what I learn, and hopefully save you some time along the way.

What TechRay.dev is about

Three things will show up here more than anything else:

Trending tech. I'll cover the tools, platforms, and shifts that are worth paying attention to, and I'll be upfront about the ones I think are overhyped.

AI, practically speaking. Not the breathless "everything is about to change" version, and not the dismissive "it's just autocomplete" version either. I want to look at what these systems are genuinely good at, where they fall over, and how to use them well in real work.

Agentic coding. This is the one I'm most excited about. We've moved from AI that suggests the next line of code to AI that can plan a task, read a codebase, run commands, fix its own errors, and hand back a working result. That changes how software gets built, and it changes what it means to be a developer.

Opinion, with receipts

Expect two flavors of content on this blog.

The first is opinion pieces. I have views, sometimes strong ones, and I'll share them. But I'll try to hold them loosely. If I change my mind about something, I'll say so, because in a field that reshapes itself every few months, the people who never update their opinions are the ones who get left behind.

The second is helpful, hands-on articles: how-tos, workflows, lessons learned, and the occasional "here's the mistake I made so you don't have to." If a post doesn't leave you with something you can use or think about, I haven't done my job.

A quick take on agentic coding

Since it's going to be a recurring theme, here's where my head is at right now.

The biggest shift isn't that AI can write code. It's that AI can now work on code. There's a real difference. Autocomplete helps you type faster. An agent can take a goal like "add pagination to this endpoint and update the tests," explore your project, make changes across several files, run the test suite, notice something broke, and try again.

That's powerful, and it's also where people get into trouble. A few things I've come to believe:

Your judgment matters more, not less. When code is cheap to generate, the scarce skill becomes knowing what to build, spotting what's subtly wrong, and deciding what "good" looks like.
Review is the new bottleneck. Agents can produce changes faster than a human can carefully read them. Learning to review efficiently, and to structure work so it's reviewable, is a real skill.
Clear instructions beat clever prompts. The people getting the best results aren't using magic phrases. They're giving agents clear context, tight scope, and a way to verify their own work, like tests or a build step.
Small, verifiable steps win. Asking an agent to rebuild your whole app in one go is a gamble. Asking it to do one well-defined piece and confirming it works is how you make steady progress.
Fundamentals still pay off. Understanding how systems work, how to debug, and how to read code makes you a far better collaborator with an AI than someone who's just accepting whatever comes back.

I don't think agents replace developers. I do think developers who learn to work with them will have a serious edge, and I want to help make that transition less intimidating.

The healthy skepticism clause

I'm optimistic about this technology, but optimism isn't the same as cheerleading. AI tools make mistakes, sometimes confidently. They can introduce security problems, invent APIs that don't exist, and produce code that looks right but isn't. Hype cycles also tend to skip over the boring stuff: cost, reliability, maintenance, and the impact on people's jobs and careers.

I'll try to cover the full picture here, the exciting parts and the uncomfortable ones. If something is impressive, I'll say so. If it's a shiny distraction, I'll say that too.

What's coming next

To give you a taste of what's ahead, here are some topics I'm planning to dig into:

How I set up my workflow for building with AI coding agents
Common mistakes people make when delegating work to an agent (and how to avoid them)
What to look for when choosing between AI tools and models
How to keep your own skills sharp while leaning on AI
Honest reviews of new tools and releases as they land

If there's something specific you'd like me to cover, I'd genuinely love to hear it. This site will be better if it's shaped by the people reading it.

Thanks for being here

Whether you're a seasoned engineer, a curious beginner, or just someone trying to make sense of where technology is heading, I'm glad you stopped by. I'm figuring this out as I go, and I'd like to do it in public, with you along for the ride.

Grab a coffee, look around, and check back soon. There's a lot to explore.

Until next time,
Raymond Busuttil
TechRay.dev

AppBlips: The Simplest Way to Turn an Idea Into a Working App or Website

Raymond — Sun, 20 Sep 2026 05:10:25 +0000

Almost everyone has had the thought, "I wish there were a simple app for that." Maybe it's a tracker for your daily routine, a quiz for your study group, or a small website for your bakery, your portfolio, or a side project. The idea is the easy part. Getting it built is where most people stop.

These days, AI can write code for you, which sounds like the problem is solved. But plenty of AI builders still feel like they were designed for professional developers. You open one and find multi-file projects, build pipelines, code editors, and settings to understand before you've made a single thing. If you're not a developer, that's a steep hill to climb just to test an idea.

AppBlips was made for the opposite kind of person: someone with an idea, not a computer science degree. It's a free AI app and website builder with one guiding goal: to make the whole process as simple as it can possibly be.

So, what is AppBlips?

In one sentence: you describe what you want in plain English, and AppBlips builds a working app or website for you in seconds.

That's really the whole idea. You don't write code, you don't set up a project, and you don't run a build step before you can see the result. You type, you watch, and you use what you made. The finished product is a single, self-contained file that you can keep, share, or publish.

AppBlips is open source under the MIT license, and it comes in two flavors: a hosted version at appblips.com where you sign in and start building, and a self-hosted version you can run yourself. More on both below.

Simple, in four ways

Lots of tools claim to be easy. Here's what "simple" actually means in AppBlips.

1. One prompt box. Instead of a wall of panels and menus, the main thing you see is a box where you type. Describe your idea the way you'd explain it to a friend, like "a habit tracker with a list of habits, daily checkboxes, and a progress bar," then press Enter or click Build. If you'd rather work from a design you like, drag in a screenshot or reference picture and ask AppBlips to build from it.

2. One live preview. Your app appears in a live preview right beside the prompt, and you can watch it take shape as it's built. The moment it finishes, you can click, type, and test it. There's no separate deploy step just to see what you made.

3. One file. Every app or website you create is a single, self-contained HTML file. There are no folders full of code, no dependencies to install, and no build step. Download it and you can host it or share it just about anywhere, and what you saw in the preview is what you get in the exported file.

4. Zero setup. On the hosted version, you sign in and start typing. There's nothing to install, nothing to configure, and no API key to track down.

Your first app in about a minute

Here's what a first session looks like:

Go to appblips.com and sign in. You can create an account right there.
Type a prompt, for example: "Build a simple habit tracker with a list of habits, checkboxes to mark them done each day, and a progress bar showing how many I've completed."
Press Enter or click Build, then watch your app appear in the preview.
Not quite right? Type a follow-up like "Add a motivational quote at the top that changes each day."
Happy with it? Export the file or deploy it to its own link.

That's it. There are no tutorials to finish first.

Tips for getting great results

You don't need any special "prompt engineering" skills, but a few habits help a lot:

Be specific about the main action. "A timer that counts down from 25 minutes with start, pause, and reset buttons" gives AppBlips far more to work with than "a timer app."
Say what it keeps track of. Tell it what people add, edit, or track, for example "each habit has a name and a daily checkbox."
Mention the look and feel. Something like "clean, minimal, dark theme with rounded cards" sets the tone from the start and helps you avoid a generic look.
Start rough, then refine. You don't have to get everything right in one go. Build a first version, then improve one detail at a time with follow-up prompts.
Show, don't just tell. If you've seen a design you like, attach a screenshot and ask AppBlips to match or adapt it.

Changing things is just as easy as making them

You ask for changes the same way you asked for the app. "Make the background dark." "Add a button that resets everything." "Put a daily quote at the top." AppBlips edits the app for you, with no code to read and none to write.

There's also a safety net. Every generation and every edit is saved as a version, with undo and redo ready whenever you change your mind. That means you can experiment freely without worrying about breaking your work. You can also switch between mobile and desktop views, with adjustable zoom, to check how your creation looks on different screen sizes.

Two studios, so you never feel overwhelmed

Rather than one giant tool that tries to do everything, AppBlips gives you two focused studios you can switch between from the header.

App Studio is for interactive things: to-do lists, budget trackers, habit trackers, quizzes, mini games, unit converters, calculators, planners, timers, and dashboards. Apps built here save their data in your browser and are designed to feel at home on a phone, so the preview starts in mobile view.

Website Studio is for content-first pages: landing pages, portfolios, restaurant or business sites, and blogs. It builds the parts you'd expect from a real website (navigation, a hero section, content sections, and a footer), and it's designed desktop-first so it reflows nicely down to phones. Websites are static and informational by default, and there's an extra convenience: you can click text, images, and links right in the preview to edit them directly. For bigger changes, you can ask through chat instead.

Each studio comes with its own starter ideas, so you're never staring at a blank box wondering what to type.

Sharing your creation

When you're happy with what you've built, you have two options.

Export. Download your app or website as a single HTML file. Because it's self-contained, you can host it or share it anywhere you like.

Deploy (hosted version). Publish your creation to its own public link with one click. Visitors can open it in any browser or even install it on their phone like a regular app, since it works as a Progressive Web App (PWA). If you'd like to keep it private, you can optionally add a password.

Bonus: put AI inside the apps you build

This one's optional, but it's fun. Before you build, you can switch on the AI toggle. Then you can ask for features that use AI themselves. For example: "Build a joke-writing app. When someone picks a topic and presses Generate, write one short, family-friendly joke, show a loading state, and display a friendly retry message if something goes wrong." AppBlips takes care of teaching the generated app how to talk to an AI model, so all you do is describe the feature in ordinary language.

How this works depends on where you run AppBlips. On the hosted version, AI features are powered by the platform, so you don't need a separate API key. On a default self-hosted setup, the first time someone uses an AI feature in the finished app, a small dialog asks them to connect their own AI provider, and that key stays in their own browser.

Hosted or self-hosted: which is right for you?

Hosted (appblips.com) is the easiest path and the best fit for most beginners. Sign in, build, and go. Your projects are saved to your account and sync across devices, and you get the deploy-to-a-public-link feature. It's currently a free cloud beta.

Self-hosted is for people who want full control. You run AppBlips on your own computer or server, with no account and no cloud sync, just you and the app. You'll need Node.js and an API key from an OpenAI-compatible AI provider (the docs mention OpenAI and local models through Ollama as options). You can start it with a few commands, or use Docker for a longer-lasting setup. Self-hosting doesn't include the public deploy feature, but exporting to HTML is always available. Step-by-step instructions are in the self-hosting guide in the docs.

One helpful note: a self-hosted setup is designed for a single local user, so keep it on your own machine unless you add your own access control.

Built with safety in mind

Every app you generate runs inside a sandboxed preview, which is a kind of walled-off space. The app can't reach into the rest of the AppBlips page, and the page can't reach into the app. It's one of the ways AppBlips keeps AI-generated code contained while you experiment.

Who is AppBlips for?

Anyone with an idea who would rather not learn a developer tool first. Students can build study aids, designers can prototype a concept, small business owners can put up a simple website, and hobbyists can make the exact little tool they've always wanted.

Developers are welcome too. The code is on GitHub, and issues and pull requests are always welcome.

Quick answers to common questions

Do I need to know how to code? No. You describe what you want in plain English and ask for changes the same way.

Is it free? AppBlips is open source under the MIT license, and the hosted version is currently a free cloud beta. If you self-host, you'll connect your own AI provider, which may have its own usage costs.

What can I build? Anything that fits inside a single web page, from calculators and games to portfolios and landing pages.

What if I make a change I don't like? Every edit is saved as a version, so you can simply undo it.

Ready to try it?

Head to appblips.com, type the first idea that pops into your head, and see what happens. If you'd like a guided start, the docs have a quickstart that walks you through your first build. And if you make something you're proud of, I'd love to hear about it.

No code, no clutter, just your idea, working.

How to Use LM Studio & Ollama From Your Android Phone in 2026 (Private, Local AI)

Raymond — Sat, 28 Mar 2026 06:27:45 +0000

Running powerful AI models locally on your computer is no longer the "hard part" in 2026. The real challenge used to be access—how do you actually use that power from the device you carry all day?

While there is still no official mobile app for LM Studio, the days of messy reverse proxies and manual IP configs are over. LMSA offers a clean, direct bridge between your Android device and your home server.

With LMSA, you can connect your phone directly to models served via Ollama or LM Studio without the headache. This turns your local AI into something truly usable, portable, and private.

👉 Get started at lmsa.app

Why This Matters: The Local AI Revolution

Local AI has improved dramatically over the past year. Models like Qwen 3.5 9B now deliver performance that rivals paid cloud tools while running entirely on consumer hardware.

The Benefits of Staying Local:

Zero Subscriptions: No monthly fees to access your own hardware.
Total Privacy: No data is sent to external corporate servers.
Zero Latency: High-speed responses over your local network.
Full Control: You choose the model, the temperature, and the system prompt.

Step-by-Step: Connect LMSA to LM Studio

This setup takes less than five minutes and works entirely over your local Wi-Fi.

1. Install LMSA on Your Android Phone

Download and open the LMSA app.
Tap Start Here to initiate the onboarding.

2. Open Server Configuration

Tap Configure next to the Server option. You will enter your computer’s details here shortly.

3. Start Your Local AI Server (Desktop)

On your PC or Mac:

Open LM Studio.
Navigate to the AI Server / Developer tab.
Load your desired model.
Click Start Server.
Crucial: Ensure the following are enabled:
- CORS (Cross-Origin Resource Sharing)
- Serve on Local Network

4. Bridge the Connection

Back on your Android device:

Identify your computer’s local IP address (e.g., 192.168.1.12).
Enter the IP into the LMSA server field.
Tap Apply.

Note: As long as both devices are on the same Wi-Fi network, the connection should be instantaneous.

Using Ollama (Recommended for 2026)

Ollama has streamlined network sharing significantly in recent updates.

Open the Ollama desktop application.
Access Settings: Click the gear icon (top-right) or find it in the system tray.
Network Access: Under the Server tab, toggle on "Expose Ollama to the network".
This automatically binds Ollama to 0.0.0.0, making it visible to your phone.

Why Use LMSA Instead of a Web UI?

While you could use a browser, a dedicated app provides a far superior experience:

Feature	LMSA App	Standard Web UI
Model Management	Switch models instantly in-app	Often requires desktop interaction
Persistence	Structured, saved chat history	Sessions often clear on refresh
UX Design	Mobile-first touch optimization	Squeezed desktop layouts
Security	Encrypted local-only traffic	Variable security

Recommended Models for Local Use

If you’re unsure which "brain" to give your phone, try these:

Qwen 3.5 9B: The gold standard for speed vs. intelligence.
Llama 3 8B: Excellent for creative writing and general chat.
Mistral 7B / Phi-3: Best for older hardware or maximum speed.

The Bigger Picture: Personal AI Ecosystems

Tools like LMSA are part of a shift toward personal AI systems. Instead of relying on centralized, censored cloud services, you are building a private intelligence hub where your devices work in harmony.

Quick Setup Recap

Install LMSA on Android.
Start Server in LM Studio/Ollama on PC.
Enable Network Access & CORS.
Enter PC IP into LMSA.
Load Model and start chatting.

Final Thoughts

Local AI is no longer a hobbyist experiment—it’s a practical daily tool. With LMSA, the gap between desktop power and mobile convenience has vanished.

Your local AI is no longer stuck at your desk. It's in your pocket.

👉 Download LMSA Now