Most websites are commissioned like buildings. There is a brief, a design, a build, a launch and a handover. Then the scaffolding comes down and the site is left to stand.
A year later the cracks are easy to find. Half the service pages describe offers the business no longer sells, or sells differently. The team page shows people who have left. The contact form still routes to an inbox nobody reads. A plugin update quietly broke the analytics tag in the spring, so nobody can say how many enquiries the site produced last quarter. Nothing failed on launch day. The site decayed afterwards, one small change in the business at a time, while the website stayed exactly as it was.
This perspective argues for a different model: a website is not a project that finishes, but a product that is operated.
The project model: why it feels right and why it fails
The project model is attractive for good reasons. It has a budget, a deadline and a moment when everyone can say it is done. Agencies sell it that way, finance can approve it that way, and a new site is a visible achievement.

The problem is what the website is for. It is the front door of an operating business. It explains what you offer, answers the questions buyers ask before they talk to anyone, captures demand and routes it to the right person. Every one of those jobs depends on things that keep changing. A front door that cannot change while the house behind it does will eventually open onto the wrong room.
What keeps changing after launch
| What changes | Examples | What happens to a frozen site |
|---|---|---|
| The business | New offers, retired services, new proof, new people | Pages contradict what sales says on the phone |
| The visitors | The questions they ask, the devices they use, how they arrive | The site answers last year's questions |
| The machines that read it | Search engines, AI answer systems, browsers, performance standards | Visibility declines without anyone touching anything |
| The stack | CMS versions, plugins, dependencies, certificates | Security exposure grows; things break silently |
| The competition | Clearer offers, better answers, faster pages | You lose comparisons you used to win |
The third row is the one most often underestimated. The way machines judge a site changes on its own schedule. In 2024, for example, Google replaced one of its Core Web Vitals, First Input Delay, with Interaction to Next Paint, a stricter measure of how quickly a page responds to every interaction rather than the first. A site that was fast by the old measure could be slow by the new one without a single line of its code changing. AI answer systems add a newer layer: they summarise and cite the sources that state things clearly and consistently, and they have no reason to cite a page that hedges, contradicts itself or is plainly out of date.

The cost of a frozen site
None of these costs appears on an invoice, which is why they are tolerated.

- Content drift. The site makes claims the business no longer stands behind, and sales has to correct them on every call.
- Conversion leaks. Forms that fail on one browser, pages that load slowly on a phone, calls to action that lead nowhere. Each leak is small; together they decide whether the site produces enquiries or just traffic.
- Security exposure. An unpatched CMS or an abandoned plugin is a well-trodden way in for attackers, and nobody is watching for it.
- Declining discoverability. Search and AI answer systems reward sources that are clear, current and consistent. A site that stops changing slowly stops being the answer.
- The rebuild cycle. Eventually the gap is too large and the business commissions a complete redesign. It throws away whatever was learned, often disrupts search visibility through the migration, and starts the decay again from a new launch date.
The rebuild cycle is the real price of the project model. It replaces steady, cheap maintenance with occasional, expensive replacement, and it guarantees that the site is at its best only for a few months in every few years.
Operate it as a product
Treating the site as a product does not mean constant redesign. It means four ordinary things that every other operating system in the business already has.

- An owner. One named person accountable for what the site says and how it performs. Not "marketing", not "the agency": a person who can say yes or no.
- A backlog. A visible list of changes, each with a reason: a question buyers keep asking, a page that converts poorly, a fact that changed.
- Measures. A small set, reviewed regularly: qualified enquiries, conversion on the pages that matter, the questions the site answers and fails to answer, technical health.
- A rhythm. A cadence that keeps the site in step with the business.
| Cadence | What happens |
|---|---|
| Continuously | Uptime, errors and form submissions are monitored; a failed form is an incident, not a curiosity |
| Monthly | Review the few pages that matter most; check conversion; ship small improvements from the backlog |
| Quarterly | Check the offer and positioning against what the business actually sells; refresh the inventory of questions buyers ask; update dependencies; check accessibility |
| Yearly | Review the information architecture and the measures themselves. A rebuild is an option to argue for, not the default |
The Guild's own content method follows the same shape. On the Content & AI search visibility page the last step is not "launch" but "compound": distribute, update, interlink and measure continuously. The aim stated there is not more content but "becoming the clearest trusted source in your category" — and a source only stays trusted if it stays true.
What "done" means instead
If the site is never finished, done has to move from the site to the change. A change is done when it is live, when it has an owner and a reason, and when someone will check whether it worked. That is a much smaller and more honest promise than "the new website", and it can be kept every month.
One rule helps more than any other: only change the date on a page when the page has substantially changed. Google's own guidance for site owners asks whether pages are re-dated to seem fresh when the content has not changed. Freshness is a property of the content, not of the timestamp.
Where agents help, and where people stay
Most of the work of keeping a site true is watching and noticing, which is exactly where agents are useful:
- checking every form end to end, and raising an alert the day one fails;
- comparing the facts a page states — services, team, locations, offers — with the systems of record those facts come from, and flagging the differences;
- noticing which questions visitors and buyers ask that the site does not answer;
- drafting updates, summaries and supporting answers for a person to edit;
- monitoring how search and AI answer systems describe the business.
Hermes note — I can keep a site current. I should not decide what it claims. Human experts set the point of view and approve every statement about prices, results and people; I research, structure, draft and measure inside those guardrails.
That division matches how the Guild describes its digital presence work: Athena on strategy and the commercial journey, Hermes on content, Oracle on discovery and measurement, Hephaestus on web engineering, with a Human Wizard accountable for the outcome.
A first month
Illustrative example, not a client engagement.
A professional services firm with a site launched two years ago wants to stop the decay without commissioning a redesign.
- Week one: name the owner and agree the three or four measures that matter, starting with qualified enquiries.
- Week one: fix the leaks. Test every form from submission to the inbox it reaches. Confirm that analytics and error monitoring work.
- Week two: inventory the pages. For each: its purpose, its owner, the facts it states, where those facts come from, and when it was last reviewed. Most firms find a long tail of pages nobody owns.
- Week three: choose the five pages that matter most — usually the home page, the main service pages, the contact page and the most visited article — and give each a review date.
- Week four: open the backlog with what the inventory found, and book the first monthly review.
By the end of the month nothing looks very different. The difference is that the site now has someone responsible for it, a list of what to change next, and a date when someone will look again.
Questions to ask about your own site
- Can you name the one person accountable for what your website says?
- Would you know within a day if your main enquiry form stopped working?
- Does every service page describe something you sell today, the way you sell it?
- Do you know which questions buyers ask that your site does not answer?
- When did someone last check how search engines and AI answer systems describe you?
- Is your next significant improvement a backlog item, or a future rebuild?
A website is never finished, because the business behind it never is. The organisations that understand this stop paying for occasional rebuilds and start getting a site that improves a little every month. If you want help with the first step, the Guild's digital presence work starts with the commercial journey, and the Guild Assessment maps where the site sits among everything else that needs attention.
