Many times I've seen e-shops go offline in the busiest weeks of their year. News sites stall at 7am, when everyone opens them over breakfast. The traffic was expected. Their TTFB was often already two seconds on a normal day.
Nobody at those companies called it a technical SEO problem.
It was a line on an invoice. A hosting plan chosen years ago because it was cheap and never complained.
Then the peak arrived.
The server sets the floor for every other metric
TTFB, time to first byte, is how long the browser waits before your server sends anything back.
It isn't a Core Web Vital (yet). But everything else waits for it.
Google calls 800 milliseconds good. That bar is pretty generous. Personally, I only consider a site good under 250 ms. At that speed, each request is cheap, so the server has room to spare when traffic spikes.
But most sites don't get anywhere close. And most owners never check.
The slowness starts before anything appears on screen.
The plan is rarely the real problem
When a site is slow, the first instinct is a more expensive server.
Sometimes that's right. Cheap shared hosting or a server far away from your customers are worth fixing first.
Usually, though, the plan isn't the problem.
Every time someone opens a page, your website asks its database for information, builds the page and only then sends it. In my audits, that middle step is where the seconds go.
A bigger plan does the same slow work a bit faster. Until traffic grows into it again.
Hosting is the box. The real decision is what you make that box do on every visit.
Three patterns cause most of the damage.
One page, a hundred trips to the database
I see this problem often.
Take a category page full of products. Instead of loading everything in a few queries, the site asks separately for each product's price. Then its stock. Then its images. Then its reviews.
That's around a hundred trips to build one page.
Each trip is quick on its own. With a real catalog and real traffic, they can add up to seconds.
For a developer, the fix is usually small. You only need to know to ask.
Loading the whole warehouse to show eight products
The second pattern is loading far more data than the page shows.
A product card needs a name, a price and a photo. The site fetches everything stored about the product. Every variant. Every translation.
Picture a whey protein shop that loads its whole catalog to show eight bestsellers on the homepage. It works while the catalog is small. As it grows, the homepage falls over.
The visitor waits for data they will never see.
Work that should happen once happens on every visit
The third pattern is repetition.
The same menu, rebuilt for every visitor. The same exchange rate, fetched from an outside service on every view. A plugin recalculating related products on every visit instead of once a day.
None of it changes between two visits. All of it runs anyway.
Caching fixes most of this. The site saves the result and reuses it for a few minutes. Whole pages can be cached too, except personal ones like the cart, checkout or account.
A CDN alone doesn't solve this. Out of the box, most cache only images, styles and scripts.
The CDN got bought. Nobody told it to cache the pages.
Nobody owns this part
SEO owns the content. Developers own new features. The hosting plan belongs to whoever pays the invoice.
How fast the site answers falls between all of them. So a team spends a quarter compressing images, and the site barely gets faster.
That gap is where technical SEO comes in. Someone has to connect a slow server response to lost crawling and lost sales. Then sit with developers until it's fixed.
A site that needs two seconds to answer on a normal day has no reserve. Peak season doubles the traffic. Pages start timing out.
This type of issue can easily kill any website. Usually on the day it matters most.
AI crawlers raised the baseline
AI crawlers have spent the last two years adding traffic nobody planned for. In the first half of 2026, AI bots and crawlers like Googlebot made almost one in ten requests for pages on Cloudflare's network.
That traffic barely shows up in your analytics. It still lands on your server.
And it hits the expensive pages: deep categories, filter combinations, old products. Pages no cache has seen.
Your normal day got busier. Your reserve got smaller.
Slow answers cost you here too. Google crawls less when your server struggles. AI crawlers simply give up, and you'll see it in your server logs.
So, many site owners do "the obvious thing". They block AI training crawlers like GPTBot and ClaudeBot to take the load off.
I think that's the worst trade you can make. ChatGPT searches the web for only about a third of questions. The rest it answers from what it learned in training.
Block the training crawlers, and you basically write yourself out of what future models know. You fix a server problem by making your brand harder to find.
Check the first response before you compress another image
Open PageSpeed Insights and enter a page that earns money. Look at TTFB in the real-user data. Above 250 milliseconds? Start there.
Then check where the waiting happens: connecting to the server, or the server preparing the page. A slow connection points at hosting. Slow preparation usually means the website itself.
One check is a snapshot. In VitalSentinel, I track it with uptime checks every minute and real user monitoring, so I see the day and hour it starts creeping up.
If the website is the slow part, ask your developers how many database trips the page makes, which ones are the slowest and what could be cached.
If nobody can answer, that's your finding.
The question to ask first
How long does your best-selling page take to start loading? Who would notice if it doubled?
A bigger server only hides a slow website until the next peak. Fix what the server does, then pick the plan.

Martin Stepanek
Enterprise Technical SEO Consultant
I am a developer-led enterprise technical SEO consultant. 10+ years building the web before I started fixing it, and I still ship production code today. I read the responses your site actually sends, tell you which findings are worth a sprint, and take the fix to your developers myself.
The technical SEO newsletter engineers and SEOs actually finish
One specific problem, taken apart, with a position on what to do about it — argued against the primary documentation and against what I see in audits. Then three stories from the last two weeks, picked and explained.



Subscribe
A new episode every two weeks. Unsubscribe in one click.