I have argued several times on this site that AI-built websites create dependency for the organisations that commission them, and that they do not survive contact with reality once staff turn over. I stand behind all of it. None of those posts answered the question I want to put here, which is what happens when your own organisation starts using AI, and who is accountable for what it produces.

I should disclose something first, because it changes how you should weigh the rest.

I rebuilt socialectric.com on Astro and Sanity, working with AI through most of the build, and migrated a library of over 140 posts in the process. It was faster than anything I have done in seven years of building websites. It also transferred exactly none of the responsibility for the result away from me. Both things are true at once, and the second one is the part the sector is not discussing.

That stack is not what I build for clients. Client work stays on Webflow, deliberately. A bespoke Astro and Sanity build maintained by one person recreates precisely the dependency I spend my working life helping organisations escape. I used it on the only site where I am the sole stakeholder and the continuity risk is mine alone to carry. If I put an NGO on it tomorrow, that site would fail the first test in my own audit.

That is the whole argument in miniature. The tool moved. The judgement did not.

What actually changed

Writing for Edgar Allan in July 2026, Mason Poe made the point that the builders most exposed to this shift are the ones who tied their professional identity to an instrument rather than an outcome, and that what agencies now get paid for is strategic judgement rather than platform execution. He is describing his own industry, and he is right about it.

For your organisation the same shift arrives from the other direction. You are not worried about whether your supplier can still charge for platform expertise. You are dealing with something more immediate: everyone on your team can now produce publishable material faster than your approval process was designed to handle. Nobody decided this. It happened tool by tool, over about eighteen months.

Build speed collapsed. Drafting speed collapsed. Review speed did not move at all.

The two numbers that matter

The Charity Digital Skills Report 2026, co-authored by Zoe Amar of Zoe Amar Digital and Nissa Ramsay of Think Social Tech from a record 807 responses, found that 79% of UK charities now routinely use AI. Digital progress is accelerating with it: 81% of organisations reported progress this year against 60% in 2025.

Set that against the governance findings in the same report. Only 7% of charities have started reviewing AI use and risks regularly at board meetings. Only 6% have conducted risk assessments for AI use cases such as bias or misinformation. The proportion updating their risk register fell to 15%, down from 23% the year before, at the precise moment adoption became near universal.

The capability picture underneath is consistent. The report rates a quarter of charity CEOs and a third of boards as poor on AI skills, and finds that almost half of organisations have no trustee with relevant digital expertise. Separate research published by the Charity Commission with Pro Bono Economics put AI skills prevalence across the trustee population at 8%.

So: 79% adoption, 7% board oversight. The adoption is not the finding. The distance between those two numbers is the finding.

There is a third number that should concern anyone responsible for institutional funding. The same report found that 58% of funders say AI is already changing the applications they receive, mainly through increased volume and repetition, while 55% of funders do not know how many of their applicants are using it. Your funders are watching this happen and cannot yet see who is doing it. That uncertainty resolves eventually, and it resolves by them looking harder at everything you publish.

Why the website is where this surfaces

Every one of these gaps eventually becomes a page.

The Charity Digital Skills Report found that 31% of charities are now using AI for governance and compliance work, including policy drafting and board reporting, and 23% for knowledge management. Read that carefully. AI is being used to draft the documents that constitute the organisation's accountability record, and those documents are published on the website, where a regulator, an institutional funder or a journalist can read them at any hour without asking permission.

This is where my longstanding position on governance documents becomes sharper rather than softer. A stale governance document is worse than an absent one, because an absent document is a gap and a stale one is a claim. AI has not changed that. It has changed the rate at which claims accumulate.

An organisation that could previously publish four considered documents a quarter can now publish twenty. The approval process still handles four. The remaining sixteen do not stop existing. They go live, they get indexed, they get cited by answer engines, and they sit on the accountability surface indefinitely with nobody clear on who read them last.

That is not a technology problem. It is a throughput mismatch between publishing and review, and it shows up on the one asset your stakeholders inspect without telling you.

You cannot hold a model accountable

This is the part that has not moved and will not move.

You cannot hold an AI accountable. There is no mechanism. There is no entity to sanction, no duty owed, no reputation at risk, nobody to appear in front of a board. Accountability attaches to the operator, and it always has.

The regulatory position reflects this. The Charity Commission has said it does not anticipate producing dedicated AI guidance, and instead expects trustees to apply existing guidance as new technology emerges. That is easy to misread as permission. It is the opposite. No new duty was created, which means the existing duties on trustee decision-making apply to AI-assisted decisions exactly as they apply to any other. The bar did not move. The volume of decisions passing under it did.

The sector's own instruments have started to catch up. The Charity Governance Code was revised in November 2025, its most significant revision in eight years, and now encourages charities to hold policies covering the use of technology and artificial intelligence. The Fundraising Regulator published its first guidance on AI in charitable fundraising in December 2025, and put trustee board involvement in strategic AI decisions at the centre of it, alongside building enough board understanding to evaluate the risks properly.

None of that asks whether you use AI. All of it assumes you do, and asks who signed off.

What this means for who you work with

I am not going to pretend AI has made professional help optional, and I am not going to pretend it has made it more essential either. Both would be self-serving and neither is true.

Here is what is true. The reason to engage someone is no longer that they can do something your team cannot do. On a growing number of tasks, that gap is closing. The reason is that responsibility has to attach to a named person who can be asked why a decision was made, who reviewed it, and what happens when it turns out to be wrong. An organisation under funder scrutiny needs an answer to those questions that does not begin with the name of a tool.

That is why the speed is worth having rather than worth fearing. A build that takes a fifth of the time is not a smaller engagement. It is the same accountability compressed into a shorter window, which means more of the budget goes to the part that was always the point: deciding what belongs on the site, who it serves, what it claims, and whether those claims hold when someone checks.

I am not neutral here and there is no reason to pretend otherwise. Moving faster is most of what I do for organisations now. But the thing being accelerated has to be a decision someone owns.

The organisations actually at risk

It is tempting to assume the exposure sits with charities using AI carelessly. Obvious errors, invented statistics, a hallucinated figure in an impact report. Those are real and they are also self-correcting, because they get caught.

The exposure sits somewhere quieter. It sits with organisations using AI well enough that the output looks entirely credible, published at a rate the review process was never built to absorb, onto a website that funders and regulators read as a statement of institutional fact.

The risk is not that the work is bad. It is that it is good enough that nobody thought to ask who was checking.

If your organisation wants a clear account of what is currently published, who approved it, and whether it holds under scrutiny, that is what a Blueprint Audit establishes.