Websites that are well optimised for search engines have a significant head start. After all, search engines also need machine-readable code.

People don’t notice if the hierarchy of headings is incorrect or if a text has the wrong content attribute. A machine needs this information to be able to process the content. 

If a website is not easily readable by AI agents, the information can either not be captured at all or only partially. Websites that are difficult to read are unlikely to feature in the AI agents’ responses. 

Present content in such a way that AI agents can make sense of it

AI systems need clear, concise statements. Clarity takes precedence over style: AI systems treat text as a data set, not as a story.

The post Checklists for AI optimisation of websites has first been published on start the loop_.

In future, however, it will not only be humans but also AI agents that browse websites. The person who sent the agent into action only comes into the picture later – once the AI agent has examined the website and deemed it relevant. Only once this hurdle has been cleared does the website have a chance of communicating with a human.

Why fewer people will visit your website

AI systems are the interface through which we will communicate with software in the future. Whereas five years ago you still had to understand the rules by which Photoshop operates in order to edit an image, today you simply come up with a prompt and the software delivers a result.  

The same applies to searching online. In the past, you would type in a few keywords, let the search engines look for them, and then work your way through the list of search results. Nowadays, you type a question into the AI window in your browser. The AI agent comes back with an answer.

When it lists the sources from which it has obtained his information, it is a very short list. This is because the AI agent has filtered out the majority of its search results during its research. What we get is a heavily pre-filtered summary. And we just have to hope that we aren’t missing anything important.

If you want to be found, you have to make yourself visible 

Many websites are not well readable by AI agents. Information may either not be captured at all or only partially. These websites remain invisible; they will not appear in the AI agents’ responses. 

People are forgiving of structural shortcomings; they don’t notice if the hierarchy of headings is incorrect or if a text has the wrong content attribute. Machines need machine-readable code; they cannot make sense of unstructured content. AI agents filter out anything they cannot read. 

Don’t forget the content

However, it is not just the technical optimisation that matters; content also plays a role. This has always been the case, but now that AI agents can actually read, this aspect has become even more important for a website’s ranking and visibility.

People scan two or three pages and read around 400 words; AI bots read EVERYTHING. People might patiently work their way through complex sentences, whilst AI bots expect clear, concise statements and filter out texts that do not get to the point. Easily digestible content formats such as definitions, comparison tables and instructions are more important than ever.

We need both: people and AI agents

Once again, people are at the heart of this journey. They may not have found their own way here – the AI agent led them here – but they are the ones who make decisions and interact. So we’re still building websites for people. 

The post An AI-readable website – why is this important? has first been published on start the loop_.

A website which performs well for search engines and whose structure and technical aspects are sound has a clear head start in the age of AI. The fact that websites must be machine-readable is nothing new. Seen in this light, GEO is not a replacement for traditional search engine optimisation; it is more of an extension. There is a great deal of overlap, as AI systems also require technically accessible, indexable content in order to read and evaluate websites.

Texts Take on a Different Meaning

Since AI systems “read” entire texts, it is no longer enough to simply place the right keywords strategically. Contextual connections and the structure of the text are becoming more important.

Headings should contain specific statements. It must be clear what the following text is about. Clear statements are better than flowery phrasing.

AI systems look for definitions and key takeaways. The goal here should be to write concisely, precisely, and in a way that’s easy to quote. Consistent terminology is also important. The product, service, or topic should always be referred to by the same name.

Whenever possible, use numbers and content formats that are easy to quote, such as definitions, comparisons, FAQs, or step-by-step instructions.

The SEO rules still apply, but with AI systems, the focus is shifting

Where to start: Check your SEO status

Checking the state of a website’s SEO optimisation is a good place to start. In my experience, Yoast SEO Checker is a good tool for this. It quickly provides an overview of a website’s SEO status. Google’s PageSpeed Insights also assesses SEO status by assigning a score.

The Yoast SEO Checker provides detailed guidance on areas that could be improved. Common issues include HTML errors, insufficient internal linking or missing meta information.

The post GEO and SEO – what’s the difference? has first been published on start the loop_.

It is impressive when an AI tool produces a complete blog article from a brief text instruction. Or a photo that looks completely realistic. The system did not invent the image, though. All parts of this image were created by someone at some point. In the form of a photo or an illustration. Midjourney, DALL-E & Co. only reassemble them. In this sense, the systems do not create anything new; they merely combine existing elements in new ways.

This article is about the concept of artificial intelligence. This article goes into more detail: Do I need AI for my website?

The overwhelmed Stone Age brain

I only have a very vague idea of what AI is. I find it difficult to grasp the abstract mathematical and statistical concepts behind neural networks. AI is not a topic that can be assessed intuitively. Our brains are not designed for that. AI works very well technically, but hardly anyone understands why.

At the same time, the fascination is immense. Here is something you can ask questions of, something you can have a conversation with. Something that has an answer for everything and seems to know everything.

Who invented it?

One unresolved and, in my opinion, very controversial issue is copyright. The training data comes from somewhere; someone wrote, said, coded, photographed, or drew it all at some point. At the moment, there are no regulations governing what data a company is allowed to access for its products. The issue of privacy is also unresolved. The big players read every social media post and every message.

Search engines list the sources that match a search term. With AI, there are no longer any sources. There is only training data, but it is not accessible.

Hallucinations and fakes

To enable the models to calculate this, the training data is divided into countless small pieces (tokens). The systems compile their content from this data scrap and a lot of statistics. The model simply calculates until something suitable emerges. The main criterion here is probability.

This can result in texts that sound credible but have no basis in reality. For example, the AI tool may cite sources that do not exist. These errors are known; they are a logical consequence of the probability methodology. The systems could correct these errors, but that would cost valuable time. At the moment, the errors are being tolerated; fast results are more important to providers than correct results.

AI is also great for manipulating and falsifying information. The algorithms used by social media platforms are the perfect breeding ground for this. Highly idealized images and videos have always been produced in the lifestyle and beauty sectors. AI makes producing such content many times easier. Authentic images are disappearing. Entire platforms such as Pinterest are now flooded with AI-generated photos.

The fact that our brains react particularly strongly to negative news plays a problematic role. At best, AI-generated videos are funny, but they are often provocative and frequently malicious (keyword: deep fakes). In the battle for attention, something controversial will always trump a positive message. Posts that generate negative emotions get the most likes.

The distinction between what is real and what is fake is becoming increasingly difficult. Many people consider AI-generated content to be authentic. This is a massive social and political problem.

Conclusion

AI is just a tool, but one with enormous potential. Potential that is difficult to assess. And that has an impact in a social and political context.

There are many useful applications, such as in medical technology. But the influence of technology on our society is so massive that we need to find political rules for it. All areas of life are affected. We cannot leave critical issues such as political opinion-forming, mental health, or weapons technology to large corporations that are solely focused on their profits.

The post Reflections on AI has first been published on start the loop_.

LLMs (large language models) are commonly used on websites. This is a (sub)form of AI that can handle language. This means that someone asks a question and the LLM system searches for something suitable from the information it has been trained with and uses it to generate a response.
Many routine processes can be made more efficient with LLMs.

• Automatically generate product descriptions in a shop
• Automatically generate ALT text for images or subtitles for videos
• Optimize content for specific search engine keywords
• Have texts read aloud
• Translate content
• Write content
• Create images

Search engines will change

When you type a search query into Google, there still appears a list of websites. But at the very top, now an AI summary is displayed. For many queries, this snippet of text already answers all questions, which means that people are less likely to click on one of the links in the list below.
This has already led to complaints as traffic to many websites has declined. Since the first page of search results usually mostly consists of paid ads, this is naturally a somewhat delicate problem for Google.

It is unclear what impact AI summaries will have in general. In any case, search results have long since ceased to be selected based on relevance, despite Google’s claims to the contrary. A search query would have to be very niche for this principle to apply. I suspect that AI summaries will only bring about a gradual change in this regard.

It is certainly interesting for search engine optimization (SEO) that people are increasingly phrasing their search queries as questions. Two years ago, if you simply entered the keywords “turf, Munich” into the search box, today’s search query might be “Who can lay turf for me in Munich? Find me the contact details.” With a certain degree of probability, the website will then appear in the AI summary that has already pre-formulated this exact question in its content.

New WordPress-Team

Incidentally, a new team was formed within the WordPress project in June 2025 to deal with the topic of AI. Here is Rae Morey’s assessment in The Repository: The Repository: WordPress Has a New AI Team. Here’s What They’re Planning to Do.

The announcements are still quite vague and there is no roadmap yet, but I think the goal will be to tap into the potential of the open source WordPress system for AI. Open source systems have an open and easily accessible structure; they communicate with all conceivable external data sources via interfaces. For companies, this would mean, for example, that systems that handle customer data, marketing, and sales could actively work together. CRM, analytics, email, marketing, and websites would network with each other and carry out tasks and actions independently.

It doesn’t happen all by itself

A funny image effect can be created quickly, and ChatGPT generates some kind of text from every prompt. But the very quick results are not really useful in a business environment. If you want good results, you have to invest time.

A chatbot designed to answer questions about a product must be well implemented and trained. Otherwise, no one will enjoy using it. We all know how frustrating it is to interact with a disoriented chatbot.
Creating high-quality texts requires well-thought-out prompts and numerous rounds of revision to refine and elaborate on the results. Plus, you should thoroughly check all the facts at the end to make sure the texts don’t contain any nonsense.
When you have a layout built, you should define the requirements precisely beforehand. Otherwise, you’ll end up with a design that looks fancy but leads to a dead end after a few clicks.

AI tools can accomplish a great deal. However, the quality of the results is not always satisfactory. A text may not get to the point, an image may contain irritating errors, and a code snippet may generate a cascade of error messages and paralyze the entire system.
One does not need a great deal of prior knowledge to place an order. However, one needs all the more when it comes to checking and correcting the output of AI tools.

What AI cannot do

The AI model is at a loss when it cannot find anything in its data that matches the task. In the best case scenario, the system has access to a search engine and can query the internet. If this is not the case, though, the model simply continues to calculate and piece together puzzle pieces. This can result in texts that sound credible but have no connection to reality.

For example, the AI tool may recommend books that do not exist or cite sources that are completely made up. This is referred to as hallucination. The term is not actually appropriate: since AI systems have no reality, they cannot hallucinate.

AI models follow clear rules to deal with unimaginably large amounts of data. They calculate probabilities. Thinking creatively outside the box and breaking new ground is therefore not the strength of these systems. Critically evaluating things is not their forte either.

Nor can they provide answers to the “big” questions. What really interests my readers? How can I win new customers for my specific product? What would I have to do differently to increase my visitor numbers? What don’t my customers like about my website? The AI model has no answers to these questions other than general marketing tips.

Who needs AI for their website?

I can think of three scenarios in which I believe work processes will change:

• Websites that publish a lot of content (text and images) on fairly general topics at short intervals.
• Websites where a lot of text and images need to be optimized for SEO.
• Websites that serve as communication hubs and interact with other marketing tools (analytics, email, social media, CRM).

On extensive websites where the same processes are repeated over and over again and which thrive on publishing similar content in ever-new variations, AI tools can certainly save time. However, you shouldn’t have high expectations when it comes to text quality. And the topics shouldn’t be too specific.

Of course, you can also use AI assistants if you only write a text once every six months or need an image quickly. For me, however, that would fall more under the category of personal preference. AI does not necessarily make a text better, nor does it necessarily make it easier to write if you are not used to do it.

The situation is different for smaller websites with a personal profile that focus on a very specific topic. Websites that do not use large amounts of text and images and do not target a mass market can continue to thrive without AI. Of course, you can make your life easier and get AI support, but FOMO is not the order of the day.

So, Kirsten, how do you use AI tools?

AI tools are very good at summarizing texts. This can actually save time when preparing a text. However, I would not have my text written by AI. I feel constrained by a half-finished text; I find it difficult to break free from the path that has already been trodden. Personally, I find revising tedious and time-consuming.

I would never use ChatGPT & Co. as a search engine. AI tools are not suitable for researching facts. Due to the nature of the system, at least 35% of the results are incorrect (so-called “hallucinations”), and sources cannot be identified. I hope that this will change in the future, but I am rather skeptical. LLMs do not look at sources; they more or less blindly assemble texts according to probability rules. This makes it impossible to trace where the information comes from. It remains unclear who is serving me information and with what ulterior motives. This is a massive problem; chatbots consume clicks and sources.

The huge amounts of garbage and disinformation currently flooding the internet worry me. That, combined with the market power of a few corporations, makes me wonder how information diversity can survive in the future.

I find the current changes in coding exciting. Many processes that used to be very cumbersome are becoming more accessible thanks to AI tools. On the one hand, this is great because it removes many obstacles. On the other hand, it naturally leads to a flood of dubious results. As in other areas, it is becoming increasingly difficult to distinguish the useful from the useless. Only someone who is familiar with the subject can verify how good and useful a piece of code is.

How can you stand out from all the AI noise?

Texts and images generated by AI are currently flooding the internet. This content often comes across as shallow and lacking in substance. There is a term for this: “AI slop,” which refers, among other things, to the uniformity of articles whose titles begin with phrases such as “The ten best [search term].”
How can you set yourself apart from this?

• Communicate your expertise from your personal perspective.
• Critically evaluate information and provide your customers with guidance.
• Seek personal interaction, whether through forums, groups, comments, or face-to-face encounters.

If you want to grow a relationship between website owner and visitors, content with personality and genuine expertise is essential.

Be unpredictable, be creative.

The post Do I need AI for my website? has first been published on start the loop_.