If the AI Bubble Bursts, Will Your Practice Website Still Work?

If an AI company runs into trouble, your patients should not have to know about it. They should still be able to find your opening times, reach your online services and get the information they need. Your team should still be able to get the website changed or fixed.

Yet as AI becomes part of more software and services, that independence is worth checking. The question for a GP practice is not whether AI is clever. It is what happens when a supplier can no longer afford it, access it or work without it.

An AI bubble bursting would not be an off switch

Tamsin, our Information Governance Lead, suggested this subject after Ed Zitron’s appearance on The Diary of a CEO on 27 August 2026. The discussion asks whether the money pouring into AI can be justified by what the industry earns. His essay on OpenAI’s future develops that argument. It is a challenge to the industry’s financial assumptions, not proof that AI is about to disappear.

If confidence falls and funding becomes harder to secure, some businesses may cut services, change prices, be bought or close. Others may adapt and become stronger. Useful technology can survive a burst investment bubble. Nobody can reliably tell your practice which company will struggle or when.

You do not need that prediction to prepare. A service can become unavailable long before its supplier goes out of business. A discontinued product, a changed contract or an unaffordable subscription can be enough.

How could that reach your practice?

The risk is not the words “AI-powered” on a supplier’s website. It is a dependency with no workable alternative. There are three practical ways that could affect a practice.

1. The price changes, and you have little room to move

If an essential feature uses a paid AI service, its costs sit underneath your supplier’s price. Higher charges or tighter allowances may mean a bigger bill, less functionality or a search for an alternative. Moving becomes harder once your team has built its working day around the tool.

Pricing models already change: GitHub introduced usage-based Copilot billing in June 2026. That is a developer-tool example, not a prediction about GP website charges. Costs can fall too: Stanford’s 2025 AI Index recorded a more than 280-fold drop in the cost of querying models at a particular performance level between November 2022 and October 2024. A cheaper underlying technology does not, however, guarantee an unchanged subscription.

Ask what remains available if you stop paying for the AI part. Essential information should not become the bargaining chip.

2. A service stops, or nobody can make the next fix

An AI-dependent feature can stop when its connection fails. It can also need rebuilding when a model is retired: Anthropic’s documentation, for example, explains that requests to retired models fail. Someone must change and test the integration.

But there is a less obvious risk: the website stays online, while the supplier loses the ability to maintain it. If the only way they can change the code is by asking AI to do it, a broken form or an important update may have to wait until they restore access or find someone with the necessary skills.

That is why using AI to write code is different from needing AI to run a website. Code reviewed and maintained by experienced developers does not vanish when their coding assistant becomes unavailable. The concern is a supplier with neither the expertise nor a fallback to carry on.

3. You need a new supplier—and a way to recover your website

A business that cannot adapt to changing costs or access could become unable to keep trading. Your practice might then have to find a new provider at short notice, while staff are already dealing with disruption. That is a possible risk, not a prediction about any particular company.

If the only usable copy of your pages and files sits inside the supplier’s system, losing access could mean losing content or rebuilding it. Closure does not automatically erase a website, but recovery can be difficult when nobody is available to help.

Check your exit route before you need it: who controls the domain, how you export your content, and whether you have a current backup you can access independently. Those questions matter whatever technology the supplier uses.

For patients, the disruption is practical

A patient does not see a failed AI connection. They see a page that will not load, a request that will not send, or information they cannot trust. Frustration, repeat attempts and extra calls to reception can follow.

If the affected website or form is their usual route for repeat medication requests, that route may be unavailable too. It does not mean prescribing automatically stops. The NHS lists several ways to request repeat prescriptions, and an independently available NHS App or clinical service is not automatically taken down by a website outage. Practices need to know which alternatives still work and how to explain them to patients.

Security is a separate, longer-term concern. If maintenance stops and an attacker exploits an unfixed flaw, patient information could be at risk where the affected system holds or can access it. A public website is not automatically connected to clinical records, and an AI outage does not automatically cause a breach. The NCSC’s advice on prompt security updates is a reminder that getting software built is only the beginning: it needs capable people maintaining it.

A breach can bring patient distress, recovery work and reporting obligations under ICO guidance. The answer is ongoing maintenance, sensible access controls and tested recovery—not assuming that any supplier can promise zero risk.

What about a ban, or AI “going rogue”?

These concerns often appear alongside the bubble debate, but they are different risks. A financial downturn does not itself ban AI. A security incident does not prove that all AI systems are unsafe. In a 2 September 2026 parliamentary answer, the government described regulation in context, largely through existing regulators—not a blanket UK ban.

Particular tools or uses could still be restricted. Security concerns are real too: the NCSC warns that untrusted content can influence an AI system’s instructions. That is a reason to limit access and review what tools can do, especially around patient information. Whether the trigger is price, regulation or security, the continuity question stays the same: can you stop using the tool safely?

At TVD, AI helps our developers. It does not replace them.

We use AI because it can be useful. But Tree View Designs was built around understanding general practice, developing our own platform and supporting the people who use it. Those skills did not begin with generative AI.

Our full-time senior developers can write, understand, fix and maintain code without AI assistance. If a coding tool became unavailable or too expensive, we could change our working methods and continue maintaining the core website service. Access to an AI subscription does not decide whether our team can do its job.

Our next-generation CMS is being developed to use AI where it helps, while retaining a core that does not depend on a live AI service to keep your website online. AI-dependent extras are separate. As our security page explains, we can also commission external testing if AI-assisted security tools become unavailable.

That does not mean every task would take exactly the same time. Some enhancements might take longer, and an optional AI function might pause. It means the expertise to maintain your core website and the people supporting your practice remain in place.

Five questions worth asking before you commit

You do not need to be a developer to ask these. A good supplier should be able to answer them clearly—including us.

You should not have to bet your website on the AI market

AI can bring real benefits. NHS England’s work on AI notetaking is one example of pursuing useful tools alongside evidence and safeguards. Being careful about dependence does not mean rejecting innovation.

Whether the bubble bursts, deflates slowly or never bursts at all, the same foundations matter: people who understand the software, control over your information and a plan for when a service changes. Those are things a practice can check today.

Sources reviewed on 10 September 2026 and linked throughout. The consequences described are possible scenarios, not predictions of a crash, a ban or a particular supplier’s failure.