REPORT: Optimising webpage design and digital presence for AI-driven discovery in charity

Author: Archie Lee

1. Introduction

The use of AI is becoming one of the most important and advantageous practices in modern computing, supporting numerous applications in both business and domestic use. While AI as a concept dates back to the 1950s, the past few years have seen the rapid rise of Large Language Models (LLMs) such as ChatGPT, Claude and, Gemini –systems that take any random user input and almost immediately produce coherent and detailed responses using advanced pattern recognition and machine-learned representations of language.

Millions of people all around the world use LLMs every day as their first point of enquiry – they are free and accessible tools that streamline online research and eliminate the need to manually browse through multiple web pages to slowly build a consensus; instead providing direct, summarised, and argued answers in seconds. In a population of ever-shortening attention spans, it’s unsurprising that making a generally menial task extremely fast and efficient should garner so much popularity.

However, while LLMs play a significant role in modern research, they do not perceive or interpret online content in the same way humans do. So, as LLMs increasingly define how people discover and browse businesses and information, how can an organisation design its digital presence so that an LLM can easily read, understand, and recommend it to potential customers.

This report explores how LLMs interpret online information, the underlying principles behind their functionality, and outlines practical strategies for designing a business’s digital presence – including website and HTML design, social platforms, and content formats – all in ways that make it clear, accessible, and optimised for modern AI-driven discovery.

2. Understanding AI search methods

2.1 Evolution from Search Engine Optimisation (SEO)

Traditional Search Engine Optimisation (SEO) relies on indexing: search engines crawl billions of web pages, extract keywords and metadata, and rank content based on relevance, authority, freshness, and the entered user query.

LLMs operate differently. Rather than simply matching keywords to webpage content, they analyse the content of the pages, identify the meaning and conclusions within what’s discussed, and then synthesise the information they learned into a unified answer. This is why it’s called Artificial Intelligence – where search engines retrieve pages, LLMs produce explanations. Where SEO ranking relies heavily on keywords and clear page structure, LLMs rely on context, meaning, and the relationships between words learned during training.

2.2 Understanding semantics for keywords

LLMs do more than searching for simple keywords. When given a prompt, they automatically rephrase it into multiple similar queries to widen their coverage of information before forming a response.

For example, if you typed “Best vets charity UK” into a traditional search engine, it would focus mainly on those four specific keywords. By contrast, if you typed that same prompt into an LLM, it would not only use your initial phrase, but it would also rephrase the prompt with varying wording and tone to ensure a wider catchment area:

“Best vets charity UK” becomes:

“Top 10 support organisations soldiers United Kingdom”

“Leading veterans’ charities Great Britain”

“Best UK NGOs for veterans reviews”

“What are the best foundations that help former infantry in the UK?”

Each variation captures the same intent but uses different wording. This increases the probability of finding relevant information and reduces the chance that valuable sources are missed because of alternative phrasing.

2.3 Search depth

LLMs are not unlimited in their resources. When searching the web, the system must balance search breadth with response speed. In practice, an LLM will analyse between 3-20 sources per query to maximise efficiency, working sequentially and stopping when:

• Multiple reputable sources (2-3) agree with each other.
• Additional webpages add little new information.
• The topic is sufficiently covered for a confident response.

This means highly competitive or widely discussed topics require stronger, more detailed/authoritative content to be selected, while niche topics may allow a broader range of sources—including smaller business sites—to surface.

2.4 LLM source preferences

LLMs evaluate a source’s credibility by analysing webpage content and cross-examining it with other reliable sources. They favour reputable, well-established sources such as:

• Government and public-sector websites.
• Academic research.
• High-authority news organisations.
• Widely cited industry sources.

Official company websites are generally reliable for factual statements about the organisation itself (e.g., “We raised £5,000,000 in 2025”) but treated cautiously when making subjective or promotional claims (e.g., “We are the best charity in the UK”).

By contrast, social media posts, regardless of whether they come from an individual or an organisation, are considered low credibility unless corroborated by more authoritative sources. They may still be used for context, but rarely as primary evidence.

3. Reading websites and website design principles

3.1 Webpage design background

Webpage design is one of the most important elements of a business’s digital presence. For AI visibility, your site must be structured in a way that an LLM can quickly interpret. LLMs perceive webpages very differently from humans. Many elements that matter to us—layout, visuals, colour—are invisible to them. Conversely, there are elements humans never notice that LLMs rely on heavily when digesting a page.

3.2 How an LLM sees a webpage

LLMs do not ‘see’ webpages visually – they don’t see all the exciting colours, pictures and shapes that draw a human in. Instead, they read the HTML, which is the underlying code that defines how the website appears on the screen. The HTML includes:

• Raw text
• Structural tags that define how the webpage is organised
• Metadata (such as authors or dates)
• Links between pages on the same website

An LLM generally cannot interpret styling choices such as:

• Font size, colour, typeface, or alignment
• Animations and transitions.
• Layout and spacing

They also struggle with content that relies on heavy scripting, including:

• JavaScript-generated text and dynamic menus
• Client-side rendering (CSR)

JavaScript makes websites interactive, an example being an animated drop-down menu showing all the different clothing items available on fashion websites. This content type loaded exclusively through JavaScript can be partially or fully invisible to AI systems.

Client-side rendering (CSR) is information generated in the user’s browser rather than sent directly from the server. That is, your device will do a lot of the loading for you, while the HTML of the actual website is a very bare skeleton of JavaScript for your device to load. This often means the raw HTML of the webpage contains very little meaningful text.

To illustrate how an LLM sees a webpage, consider a typical charity homepage. A human sees images, banners, buttons, branding, and sliders. An LLM, however, sees only the underlying markup: a sequence of links, scripts, metadata, and any raw text included directly in the HTML.

This difference highlights why well-implemented webpage text content and HTML structure are essential for AI readability.

3.3 User experience and interfaces

However, this does not mean human-focused design should be ignored. Smooth transitions, intuitive navigation, and appealing layouts are still essential for engaging visitors. From a visibility standpoint, strong user experience increases traffic, improves SEO performance, and encourages discussion—each of which feeds into how relevant your site becomes online.

Despite the vastly different design cues that humans and LLMs respond to, human usability and AI readability should not compete. The best websites achieve both: visually engaging designs supported by clear, accessible information in the underlying HTML.

3.4 HTML design

Looking at how these principles translate into a HTML environment; to make your website AI-friendly, you should ensure your key information appears in text-rich, obvious HTML text within the appropriate structural tags. These HTML tags define the structure of a webpage, and LLMs exploit them to understand which parts of the page matter.

Below are examples of tags that LLMs prioritise:

In this context, writing for HTML is about guiding an LLM toward the information you want it to notice. Your essential facts—who you are, what you do, and why it matters—should be presented clearly within structured HTML.

More advanced functions, styling, and interactive elements can then be layered on top with scripting for a rich, human experience. This balance ensures LLMs see accurate, up-to-date information directly from your website, rather than relying on secondary sources like news articles or third-party summaries. You yourself want to be telling potential customers what your business is all about – not someone else.

4. Content strategy for AI visibility

4.1 Core principles to appeal to AI

Digital content comes in many forms, but LLMs have a clear preference for concise, well-structured information that is easy to extract from a webpage. The majority of online content today appears on social media platforms, which can be challenging for LLMs because many of these platforms restrict access to non-logged-in users. AI systems cannot log in, meaning they can only see what’s publicly available.

LLMs can usually read the text from public business account pages, but engagement data such as likes, shares, and comments is hidden. They also rely heavily on supporting text to interpret media such as images and videos.

Concerning images in particular, LLMs can detect that an image exists and where it should be, but they cannot interpret what’s in it unless the post includes a caption or description. Where available, image alt text, which is a written description of the image, is embedded in the HTML code. This feature is very useful, but on social platforms this is often missing.

Video is similar but can be slightly more accessible. LLMs cannot view the visual content, but they can understand videos that have publicly accessible subtitles or transcripts. This varies by platform, but for example, YouTube’s autogenerated subtitles are public and easily readable by AI systems. This means you can ask an LLM anything about a public subtitled YouTube video at any timecode, and it could tell you what is happening. By contrast, Facebook and Instagram subtitles often require login access and hence remain invisible.

Therefore, like images, including quotes, descriptions, or summaries of your video in the post caption can significantly improve AI comprehension.

Below is a summary of how accessible different platforms are to LLMs:

If your main audience is on platforms with low AI visibility (e.g., Instagram or TikTok), a practical solution is to upload the same video on YouTube. This gives LLMs a reliable text-based version of your content while still allowing you to post where your human audience is most active. Your YouTube channel doesn’t need to have the same popularity as your other platforms – as long as you make sure that your platforms are professionally linked, an LLM will find it.

To reinforce this idea, strengthening AI visibility by including descriptive captions or quotes whenever you post visual media is also very good practice.

4.2 Keyword placement

Keywords remain important in both SEO and AI visibility, but modern systems no longer reward flooding a page with repeated keywords. LLMs treat keywords as anchors that connect your content to other known concepts within their training data.

For example, using the phrase “Veterans homeless support in the UK” in your own content links it to themes such as:

• Armed forces
• Charity and NGO work
• Homelessness
• Support schemes and funding
• UK social policy

Again, LLMs don’t look solely for those exact words in the original phrase. They can understand related terms, synonyms, and variations. A page mentioning “veterans, charities, homelessness” can still be deemed relevant when a user prompt includes “soldiers, NGOs, vagrancy.”

A more effective strategy than chasing keyword density is simply to write naturally and clearly. The key is to recognise that LLMs interpret text like humans do, so clarity, context, and structure, are far more important. This includes using a rich vocabulary to improve the chance of your site being picked up by an LLM.

Best practices for keyword placement and semantics:

• Introduce the topic early to help AI identify relevance.
• Use natural variations of keywords to broaden coverage.
• Prioritise clarity over repetition—LLMs understand synonyms and context.
• Group related ideas into clear paragraphs (helps web crawlability).
• Place keywords in strong positions (titles, headers, opening sentences).
• If using images or video, include the important information as text in captions.

4.3 Content output frequency

Content output frequency is observed differently between search engines and LLMs.

Search engines do not reward posting frequently on its own. However, they do favour:

• Fresh content for fast-changing topics.
• Sites with a wider coverage of a subject area.
• Active, regularly maintained pages.

So, search engines interpret frequency as a signal of activity and relevance. For LLMs, while frequency is still accounted for, they are far more concerned with the consistency and quality of the content being published. AI systems reward:

• Clear, well-written text.
• Content that establishes your domain expertise.
• Regular posting patterns over long periods.
• Recurring themes that define your niche.

To score high with LLMs, high-quality, detailed content produced consistently is far more beneficial than posting every day. Over time, repeated signals strengthen your association and authority within a topic and make your content more likely to appear in AI-generated answers.

On platforms where LLMs cannot read most content (e.g., Instagram, TikTok), posting frequency only affects human visibility. AI comprehension is unaffected unless the content is published in a text-accessible format elsewhere.

5. Analytics and measurement

Measuring the impact of AI-friendly design is challenging because you can never know with complete certainty when an LLM has recommended your website. However, there are several reliable indicators that suggest whether your visibility within AI tools has improved.

A simple and practical method of testing visibility is to act like a typical user who might find your organisation through an LLM. You can test whether the information you designed for AI systems is being surfaced by trying a range of queries, for example:

• “Who are the best veterans charities in the UK?”
• “How much did the Veterans’ Foundation raise in 2024? Cite sources.”
• “What is the Veterans’ Foundation known for?”
• “Which sources explain homelessness in veterans?”

You can test both direct queries (containing your organisation’s name) and broader queries (where you should appear based on relevance). Asking an LLM to cite its sources is also helpful for confirming whether it found your content.

For more technical insight, free analytics tools like Google Search Console and Google Analytics can help you understand how users arrive at your site.

• Google Search Console shows which search terms people used before clicking your site through Google.
• Google Analytics shows referral sources (what page the user came from), and whether the visit was labelled organic, referral, or direct/none.

Another common indicator of improved AI visibility is an increase in traffic to older pages. This happens because:

1. LLMs recommend your most relevant content, not necessarily your newest.
   A well-written article from two years ago may appear in an LLM answer if it still provides the strongest explanation.
2. New visitors perform their own research after discovering you through an LLM.
   People rarely trust a new brand immediately. After an LLM mentions your organisation, many users will manually search your name, explore your website, and read older posts to check if they can trust you.

This browsing behaviour is a strong signal that LLM-driven discovery is beginning to work.

Finally, a more technical approach to testing your visibility and interest is checking the referral URLs. A referral URL is the web address that represents where a user was right before visiting your site. For example, a user clicking your site directly from a Google search results page generates a referral URL that includes the exact search query the user typed.

Tools such as Google Analytics records this automatically.

However, the important part is that there are situations where no referral URL is provided, such as:

• Links clicked inside private messages.
• Links inside PDF or Word documents.
• Bookmarks and manually typed URLs.
• Links generated by LLMs (ChatGPT, Claude, Bing Copilot, etc.).

In these cases, analytics tools classify the visit as “direct” or “(none)”.

Therefore, a sustained rise in direct/(none) traffic, especially alongside increased explicit brand-name searches, is a strong indicator that LLMs are recommending your content to users.

And although it’s not possible to measure LLM-driven referrals with complete certainty, these patterns—combined with the LLM direct querying method—provide a strong practical method for understanding how your business’s AI visibility is developing.

6. Conclusion

To summarise this report, building a digital presence that appeals to modern search methods means balancing the creative, visual excitement that humans respond to with a clear, structured, and accessible foundation that AI systems can reliably understand. LLMs no longer reward shallow daily posts or repeated keyword tricks. Instead, they respond to consistent, well-written content that uses rich vocabulary, conveys genuine expertise, and is presented in a way that models can understand. For an LLM, two detailed posts each week demonstrate more authority than ten shallow posts each day, because they signal real knowledge rather than volume. When an LLM recognises that you understand a topic deeply, you are more likely to be recommended and surfaced as a high-value source.

Creating an AI-friendly digital presence is ultimately about making your business easy to understand. That means clear descriptions of who you are and what you offer, placed on structured, public pages that allow AI systems to extract meaning. It also means using content formats—like articles, transcripts, summaries, and well-organised text—that LLMs can reliably interpret. Social media remains important for human engagement, but long-term discoverability increasingly depends on the text-based foundation beneath it.

By approaching your online presence with this mindset, you position your business to be accurately represented by LLMs, surfaced in relevant queries, and recognised as a trustworthy authority within your domain. This is not a sudden transformation but an ongoing process of building clarity, consistency, and semantic strength so that AI systems perceive your business the way you intend. Organisations that embrace this shift early will hold a lasting advantage as AI becomes an ever-greater gateway to discovery and decision-making.

6.1 Key takeaways

• LLMs understand meaning, not just keywords. They analyse context, relationships between ideas, and the variety of words and phrasing – rather than relying on simple keyword matching. Clear, well-structured writing outperforms excessive keyword use.
• AI searches differently from traditional SEO. Search engines index pages, but LLMs reinterpret user prompts, generate multiple variations, and synthesise information from 3–20 high-quality sources. Stronger, clearer, more authoritative content is more likely to be selected.
• LLMs don’t see webpages like we do. They don’t see the pretty visuals but instead read HTML code. They only interpret raw text, headings, lists, tables, metadata, and internal links, while most styling, images, and JavaScript-generated content are invisible. This makes clean, text-rich HTML essential for AI readability.
• Human UX still matters. A visually appealing, intuitive experience boosts human engagement, which strengthens your overall online relevance. The best websites balance attractive design with AI-readable structure.
• Content format affects AI visibility. LLMs rely on captions, alt text, transcripts, and public subtitles to interpret images and videos. Platforms like YouTube are far more AI-accessible than Instagram or TikTok. If your primary audience is on TikTok, consider posting mirrored content on YouTube to allow AI to read and recommend your content to users.
• Keywords don’t play the chief role anymore. Using natural language, variations, synonyms, and clear structure increase your likelihood of being matched to user queries. Place important terms in strong positions like titles, headers, and opening sentences, and understand that repeated use of the same words will not aid your visibility.
• Consistency beats frequency. LLMs reward sustained, high-quality content that shows expertise over time. Two strong posts per week can outperform ten low value posts per day.
• AI visibility can be measured indirectly. Testing queries in LLMs, monitoring search terms, and observing increases in direct traffic all provide clues about rising AI-driven discovery. Older content gaining new traffic can also signal that LLMs are recommending it.
• Success depends on clarity and structure. Make your organisation easy for AI to understand by using accessible webpages, meaningful text, and well-organised content. Organisations that adopt AI-friendly digital design early will gain long-term advantages in online discoverability.