So much of the conversation around AI and the web is geared towards commercial websites that sell things. Every new tool and personalisation feature seems built around one question: how can we get people to buy more?

But what if you aren’t selling anything? What if you are a grant giver, or you rely on donations to keep your services running, or everything you do is about supporting a local community?

When you take the sales pitch out of the equation, the web trends for 2027 look quite different. For foundations and charities, the next year isn’t about chasing flashy tech. It is about building leaner, faster websites that work just as well for an AI agent as they do for a human in need of help.

If you’re a comms directors or digital leads running a purpose-driven website, here’s what you should be focusing on for 2027.

2. Where personalisation fits when you aren’t selling

Because commercial personalisation is obsessed with upselling, it is easy to assume charities don’t need it. In reality, it solves one of the biggest headaches on a non-profit website: trying to talk to two completely different groups of people at once. The people you help and the people who support you.

Most charity websites are stuck asking for money while simultaneously offering support. Privacy-safe personalisation, using simple cues like location or how someone arrived on the site, rather than creepy tracking, helps you subtly separate those journeys. A returning donor sees impact updates and a straightforward way to give; someone searching for urgent support gets straight to the help they need without wading through fundraising appeals.

For foundations, the most useful bit of AI isn’t personalising layouts, it’s using tightly controlled chatbots (a topic for another day) as an eligibility filter. Instead of forcing a small charity to read a massive guidance document or fill in yet another boring form, a simple chat tool grounded in your funding rules lets them ask, “We’re a two-year-old local charity with £100k turnover. Can we apply?” They get a straight answer in seconds, and you don’t waste time reading applications that were never going to qualify.

4. Real voices in a sea of AI copy

When anyone can generate three paragraphs of passable, empathetic-sounding text in seconds, generic charity copy becomes completely invisible. In 2027, audiences will tune out vague lines about “making a lasting difference” and look for proof that real people are doing the work.

That means putting trusted voices front and centre: direct quotes from frontline staff and the people you support, original data from your own research, and honest reflections on what hasn’t worked alongside what has. It also means breaking the habit of locking vital information inside 40-page PDFs (which are awful on mobile, frustrating for screen readers, and harder for AI tools to digest properly…. not to mention they’ve probably been saved as a 100MB file by your print designer). Put your reports and grant guides on proper, well-structured web pages instead.

And whatever you do, keep AI-generated images off your site. These days, AI generated images are sometimes used as a “brand style” and that’s ok in a way but people spot them instantly, and nothing wrecks trust faster than a synthetic photo of a “service user” with uncanny lighting or six fingers (ok… ok…. I get it! It’s not 2025 and AI generated images don’t do six fingers anymore…. but you get the point… please don’t write in with complaints).

A charity client asked us why people who start a donation do not finish it, since their analytics showed where people stopped without showing why.

Some drop-off is normal. M+R Benchmarks 2026 puts donation page conversion at 11 per cent on desktop and 8 per cent on mobile, counting everyone who reaches the page. Fundraise Up’s 2025 data reports 26.7 per cent in the US, counting donors who open the checkout.

Our hypothesis was about layout, so we set out to test a preference between the donation pop-up and the embedded form.

We ran an unmoderated study in Useberry with sixty people who had donated online in the past two years, on mobile and desktop, each completing a real donation with test cards.

The layout preference separated the two formats on desktop and not on mobile, and three of the reasons people gave for stopping had nothing to do with either format. All three created friction, and all three were straightforward to fix; so we now check for them across our work with all our clients.

1. People want to see what the site is about before they decide about cookies

Ten of the sixty users who met a cookie overlay raised it without being asked, and they gave the same reason each time: they were deciding before they could see what they were deciding about.

Nielsen Norman Group found the same resistance, with participants describing large overlays as intrusive and one saying they did not want to accept cookies before knowing what the site was.

Enforcement has moved the same way, with the CNIL, the French data protection authority, fining Google 150 million euros in 2021 and Microsoft 60 million euros in 2022 for making cookies harder to refuse than to accept.

We now recommend keeping the banner small enough that it never covers the page, placed at the top or bottom, with Accept, Reject and Customise as buttons of equal weight and the detail behind Customise, so people answer the question once they can see what and who they are answering it for.

2. One national flag on an international charity

The charity works across many countries, and we recruited users in the United States and the United Kingdom.

Nothing on the site states a nationality, apart from the language selector in the top right, which showed a Union Jack beside the letters EN.

Four of sixty users came away believing the charity served people somewhere other than where they lived.

Some users named the flag directly, and all of those were in the United States.

One of them told us the site felt irrelevant to them because there was a foreign flag, and in the same answer they could not say what the organisation did.

A flag tells a visitor which country an organisation belongs to, which sends the wrong answer when the organisation works in the visitor’s country as well.

We now recommend the language name in its own script, so English, Español, Français, العربية, हिन्दी, 中文 and 日本語, with a globe icon where the control needs a symbol.

Flags belong on content that genuinely applies to one country.

3. Asking twice could cost you more than asking once

Out of sixty users, seven raised the monthly upsell screen without being prompted, and named it as the point where they nearly stopped.

The screen arrives after someone has chosen a one-off amount, offering two monthly options as full width buttons with their existing choice underneath as a small text link phrased as a refusal.

The screen is a Fundraise Up (a platform we often recommend) checkout feature, the pre-conversion upsell, which appears once a one-time amount is chosen and before payment, with the platform’s AI choosing which amounts to show where that is switched on.

The repetition drew strong reactions, with users writing that being asked more than once to be a monthly donor made them feel like giving up.

Users also read the layout as intentional, describing the one-off option as made deliberately difficult to see. 

Two catalogued patterns describe what the screen does, and both carry an ethical charge: visual interference, where an option is present and clearly worded while its styling hides it, and confirmshaming, where declining is written to carry a social cost.

The UK Code of Fundraising Practice requires that fundraising does not place unreasonable pressure on a donor, and that donors can choose the type of gift they make.

We recommend one monthly suggestion rather than two, with the one-off continuation as a button of equal weight labelled as the donor’s existing choice.

Where someone has already chosen one-off, the second approach belongs on the thank you page or in the follow up email, where declining costs them nothing.

The part we are still working out

Asking for monthly donations raises money. NextAfter’s test with 7,734 participants found that defaulting a form to monthly lifted recurring gifts by 187.7 per cent and total revenue by 214.3 per cent, while one-time gifts fell 34.2 per cent.

Neon One’s 2026 figures put recurring donor retention at 78 to 80 per cent against 32 per cent for one-time donors, which is why fundraising teams want monthly donors.

Both describe a single session. One of our participants would count in that data as a success: they completed the donation, rated the experience zero out of five, and told us they would give elsewhere next time.

Fundraise Up’s Pulse of the Donor, drawn from more than a thousand organisations across four countries in 2025, reports the pre-conversion upsell converting 2.4 per cent of one-time donors in the US, up 20 per cent on the year, with a redesign of that screen associated with a 10 per cent lift in recurring conversions. They have far more data than we do, and their figure also means that roughly 97 in 100 people who see the screen decline it.

What that costs over the following years, nobody has measured, since the experiments that get funded stop at the checkout.

To truly evaluate impact, perhaps an interesting study would need to compare a calm donation flow against an insistent one over a two-year period. In the meantime, our strategy balances performance and user respect: offering a single monthly suggestion, presenting the donor’s original choice on an equally prominent button, and framing the option in clear, non-coercive language.

Uncovering overlooked friction points like these validates the importance of user testing. At Pixeled Eggs, testing ensures we meet our clients’ fundraising goals without relying on deceptive dark patterns.

When I was studying, one of my lecturers said something that stuck with me: “If you’re going to get into this game, you’ll need to get very comfortable with being uncomfortable, most of the time”. What they were referring to was that for a good portion of a design process, you may not know what ‘the final outcome’ looks like.

Add to this the commercial aspect of a deadline and it can become quite a daunting place to find yourself, but then I don’t think any designer got into this game for an easy ride.

The right answer is easier to find than ever

Any decent AI tool can plan a donation journey these days, in seconds. A sitemap, a user flow, the content each step needs, a page structure that follows best practice. Not long ago, that would have taken a few days of workshops, sketches and laying things out in Figma.

That’s genuinely useful, and I’m not criticising it. But it raises an awkward question for anyone who designs for a living: if the machine can get it right, what’s my role?

My thought? Stop focussing so hard on what’s right and turn your attention to what happens when things go wrong.

“Right” is an expectation

AI is good at ‘right’ because ‘right’ is well documented. More importantly, no one praises you or remembers you for getting things right under normal situations; it’s just expected. Millions of checkouts, sign-up forms and service pages have been built, tested, written about and copied. The patterns get averaged, and the average is pretty good these days.

Every agency, freelancer and in-house team has access to the same competent answer. When everyone arrives at the same well-organised donation flow, nobody stands out anymore. As a recent UX Collective piece put it, AI design isn’t ugly, it’s fluent, and that’s the problem. The average is built from what usually happens; in reality, people don’t behave like ‘usual’ all the time.

“Wrong” is a moment of hope

Ask the same tool what to do when things break, and things get interesting. A generic 404 with a search bar. A flat “No results found”. An error message in red. Technically correct, but emotionally absent, and not really understanding the context or situation of either the visitor or the brand. Nielsen Norman Group has noted that teams can get so focused on the ideal user path that anything outside it becomes an afterthought. They’ve also written about how to improve 404 pages and no-results search pages.

AI struggles most with the moments that don’t follow a script. Some pages need a single action to dominate, like a crisis line or an emergency appeal. Knowing a call to action should be prominent is easy; making it unmissable without dominating over everything else comes down to a designer’s instinct.

Then there’s the person who doesn’t know what to do. They don’t know what they need, or don’t have the words for it. They aren’t on a journey. They’re stuck.

None of these has a tidy pattern-library answer. They need judgement, empathy and an understanding of the specific people involved. And that understanding comes from talking to them.

It matters more when you’re helping people

Most of our work is for charities, not-for-profits and purpose-driven businesses. A large share of the people visiting those sites arrive at a hard moment. They’re worried, grieving, in debt, unwell, or trying to help someone who might be in one of these situations. The Government Digital Service has written about designing for people at critical stages of their lives.

For an online shop, a dead end costs a sale. For a charity, it can cost someone the help they came for, that could be their life. A person in crisis who hits an empty results page doesn’t ‘refine their search’, they leave, and often they blame themselves. Nielsen Norman Group warns against error messages that blame or shame users.

As designers, we embrace the challenge, understanding the uncomfortable truths, and finding unique solutions that answer those challenges; but that’s not something we should expect our audiences to also enjoy. Not knowing what the solution is, and learning to trust you’ll find a solution through your intuition, experiences and process isn’t something we should be expecting people to do at a time of need or failure. Charities like Chayn show what the alternative looks like: their trauma-informed design principles, including safety, agency and hope, are a great starting point for designing these moments.

So for us, things going wrong isn’t an anomaly, it’s a core use case. We need to know we’re getting it right when things are going wrong.

Recognition builds trust

People want to be seen and understood. Nobody wakes up thinking of themselves as a “user”, the brands and organisations who are succeeding, recognise this. Their designs make the most important thing to people, obvious.

That kind of care can’t easily be generated. It comes from qualitative, human-to-human research: sitting with people, hearing the words they use, noticing where they hesitate. Nielsen Norman Group argues that research with real people builds empathy in a way AI-generated ‘synthetic users’ can’t. User research strategist Nikki Anderson argues that AI should sharpen a researcher’s judgement, not stand in for it. Of course AI can help you make sense of that research, in the same way it’s helped me write this article, but it can’t do the listening and observation for you, nor can it make the judgement calls you’ll make in the moment, in context.

In summary

AI is already part of everyday work in our sector; the Charity Digital Skills Report 2026 found that 79% of charities now use it routinely. AI-assisted design workflows are going to produce a lot of competent, well-structured websites that all behave with familiarity, and having been in this industry for a while, that’s no bad thing. We shouldn’t be putting cognitive load on our customers and visitors. The work that truly connects, builds trust and becomes memorable, though, will be the work that knows what to do when things aren’t going right, the work that actually feels like it recognises the visitor’s situation. That’s where human experience is still invaluable and priceless. Maze’s 2026 research report reached the same conclusion: automation is now the baseline, and human judgement is the differentiator.

As designers, we’ve learned to be comfortable being uncomfortable, however the people we design for shouldn’t be expected to learn the same.

So worry less about getting it right, and get really good at understanding what to do when things go wrong. That’s where people feel seen, and where your work stands out from the mediocre noise.

The term Artificial Intelligence was first coined at Dartmouth in 1956, but the ideas behind it, machine learning and task automation based on programmable rules, had already been around for years. Generative AI, too, isn’t exactly new. In 1961, Joseph Weizenbaum created ELIZA, a computer program that could hold basic conversations with humans using natural-sounding, empathic language.

Fast-forward to 2022, when OpenAI launched ChatGPT, combining Generative AI with powerful Large Language Models, and suddenly AI moved from niche tech to something that touched daily life. It’s had an impact on how we live and work not seen since smartphones actually became smart in the mid-2000s (anyone remember Blackberrys and Windows Mobile?).

Whether you love it or hate it, AI isn’t going away and it’s going to shape our lives for the foreseeable future, until the next big revolution comes along (still holding out hope for cloning myself to get more done… wait, erm, maybe that’s what AI already is?!).

There are now so many AI tools that help us design and develop websites, from tools that help with wireframing and templating to coding assistants, there are hundreds of excellent and not-so-excellent applications that can help get us to a solution faster than we used to be able to do before. At Pixeled Eggs, we’ve been experimenting with many of them, and using a few (responsibly) in our process. These system are getting more powerful and much better at handling complex solutions. They are also evolving fast, and as they grow, our jobs as designers and developers will evolve with them.

In this article, I want to focus on practical, actionable ways AI can help you maintain and improve your website and make managing your content just a little easier.

A few final thoughts on AI

There are lots of other areas where AI is making an impact on website performance, like security and personalisation. But today, we’ve focused mainly on content and usability.

A few things to keep in mind:

• Human oversight is essential.
  AI is a great assistant, but it can’t replace your brand’s voice, values, or nuance. Always add the human touch.
• Be ethical and mindful.
  AI tools raise real concerns around data privacy, bias, and environmental impact (especially energy and water usage for generating images and video). Use it responsibly!
• Stay curious.
  The world of AI is evolving fast. I find some of the AI News sites a bit too nerdy (even for me!) but if you want to keep up, sites like the AI Tools Directory offer a good, user-friendly overview of new tools.

And remember: You don’t always need a brand-new website.

Sometimes a good tune-up, smarter content, better performance, and more thoughtful design is all it takes to make your existing site even more powerful.

If you’d like a hand making the most of your website and your content, I’d love to chat.

This post has been revised on 29.07.2026

Photo by Nik on Unsplash