Monitoring a marketing site — uptime, errors and SEO signals

A site you do not monitor costs weeks you call a 'quiet pipeline'. The form sends to an inbox someone has left. DNS points somewhere the certificate does not cover. A redirect loop after a 'small' URL change. You think the market is cold. The market is hitting a 404, a timeout or an empty thank-you step. That is not a marketing problem first. It is a monitoring problem you have called operations.

Teams buy the wrong medicine. They see a slide with traces, RUM and 'full-stack observability' and draw an APM project on top of a site that gets a few thousand views and one form. Or they do the opposite: they have no check, so the first alert is a customer on LinkedIn. Both cost. One in theatre. The other in weeks you cannot get back in Search Console.

It is also an SEO question, not only an uptime question. Google notices a soft 500, a sitemap that 404s, or a page that suddenly is `noindex`, long before you look in Analytics. If you only measure 'is the server up', you can have a site that answers 200 and has still disappeared from search. Monitoring a marketing site is the boring bridge between operations and what you thought was content.

We write this because 'website monitoring' has become a badge on the status page you buy before you have a sentence about what may break. An honest setup is boring: the home page, the thank-you page, the sitemap, the certificate, Search Console, and a name that gets the mail. Without that sentence you have a dashboard. You do not have monitoring you can defend when someone asks why leads stopped.

Monitoring here is the checks that tell you the site is still the site you launched. An HTTP check on the URLs that matter. A synthetic submit on the form, or at least an eye on the endpoint answering. Certificate expiry. That `sitemap.xml` and `robots.txt` are still yours. That Search Console does not have a new coverage hole. That LCP on the home page has not slipped after a hero-video experiment.

It is not observability. Observability is being able to ask the system why it is slow when you have volume and distributed traces. A marketing site in Next.js on Vercel or Cloudflare has logs and a deploy history. That is enough for you to find a bad release. It is not a reason to buy an APM you never open because there is no on-call and no request storm.

It is not Analytics either. Sessions and conversions tell you whether people use the site. They tell you too late that the site was broken — and they tell you nothing if the cookie banner has turned measurement off. First-party analytics after consent is another post. The duty here: you must know the machine works, even when the graph is empty because consent is no.

And it is not a PageSpeed screenshot. A lab score you run before launch is a check. A lab score you run every week on the home page, and react to when LCP slips, is monitoring. A lab score in a pitch deck is theatre. Core Web Vitals in Search Console are closer to the truth than a single run in an incognito window on your fibre connection.

The first failure is only pinging the home page. The WordPress site answers 200, and wp-admin or a plugin endpoint is down. The Next.js site renders the home page from cache, and the server action that sends the mail fails. A CDN can hold an old 200 long after origin is dead. Ping what the user hits — and what happens when they press send.

The second failure is five tools and no owner. Uptime at the host. Vercel mails in a shared inbox. Search Console at an agency you no longer have. PageSpeed in someone's bookmarks. When something breaks, nobody knows which alert counts. One channel. One name. The rest may exist as evidence. They must not be what you trust at three in the morning — because you do not have a rota anyway. You have a Monday morning.

The third failure is ignoring the weeks the site typically breaks without 'being down'. A cookie banner that blocks the form. A campaign pixel that pushes LCP. A plugin update. A DNS move. A 'small' slug change without a redirect. Monitoring is also a checklist after those changes — not only a cron against the home page. If you deploy without opening Search Console and the form afterwards, you have a hope, not a check.

The fourth failure is buying app monitoring for a brochure. Traces, service maps and an 'error budget' are right when you have a product with users all day. On a marketing site that is how you get a dashboard you are ashamed to open because it is empty or noisy. Take the boring: HTTP, certificate, sitemap, Search Console, one synthetic path. Take APM when you have an app the observability post is actually about.

A WordPress site typically dies in the plugin layer. An update, a contact-form plugin that stops sending, a cache plugin that serves an old 200, a security plugin that blocks POST. Monitor admin if you must, but especially monitor the public POST and the thank-you page you think exists. Uptime on the home page is the check WordPress misses most often.

A Next.js site on Vercel or Cloudflare typically dies in a bad deploy, a missing env var, or an edge cache that holds an error. Preview deploys catch some of it. They do not catch that production is missing the secret preview had, or that a redirect in `next.config` has hit an old URL Search Console still loves. Deploy mails are evidence. They are not monitoring until a human has a check from the outside.

Campaign weeks are the third way the site breaks. A tag-manager container, a chat widget, a full-bleed hero, an A/B tool. LCP slips. The form waits on a script that did not arrive. The cookie banner double-blocks. Here PageSpeed and a real submit — not an uptime badge — are what tell you that you have bought traffic to a site you made heavy yourself.

If you are considering 'getting monitoring under control' by buying the same stack as the app, be honest: you are buying a language you will not speak every day. A marketing site needs boring checks and a Monday rhythm. An app needs traces. Mix them and you get an expensive dashboard for the site and a site you still discover is broken via a customer.