“Not a definitive signal” is Google’s own phrasing about server location. It appears one sentence after the line hosting salespeople quote, and it is the more useful half.

Hosting decisions get made on two arguments, geography and speed, and the documentation qualifies both. That does not make hosting unimportant. It makes it important for different reasons than the ones on the pricing page.

This page sets out what the documentation says about location and response time, then what hosting genuinely determines.

Server location, in full

The two sentences belong together, and they are usually separated.

The part that gets quoted, verbatim. The server location is often physically near your users and can be a signal about your site’s intended audience.

The part that follows, verbatim. Some websites use distributed content delivery networks or are hosted in a country with better webserver infrastructure, so it is not a definitive signal.

What Google says it uses instead. Country-code top-level domain names. hreflang statements, whether in tags, headers or sitemaps. And other signals including local addresses and phone numbers on the pages, the use of local language and currency, and links from other local sites.

One more line worth knowing. Google states that it does not attempt to vary the crawler source used for a single site in order to find possible variations in a page.

What that rules out. Serving different content based on where the crawler appears to be. It will not see the variations, and building a strategy around that is building on something the documentation says does not happen.

The practical consequence for a B2B firm selling internationally. Your hreflang implementation, your ccTLD choice and the local signals on your pages do more for geographic targeting than which data centre you rent. Those are cheaper to fix, too.

Time to first byte, and why it is not a Core Web Vital

The other hosting argument is speed, and here the documentation is precise in a way that cuts both directions.

The published guide, verbatim. As a rough guide, most sites should strive to have a TTFB of 0.8 seconds or less. Good values are 0.8 seconds or less, and poor values are greater than 1.8 seconds.

The qualification, verbatim. Because TTFB is not a Core Web Vitals metric, it is not absolutely necessary that sites meet the good TTFB threshold, provided that it does not impede their ability to score well on the metrics that matter. Which is why the thresholds are described as a rough guide.

Why it still matters. Time to first byte is the floor under largest contentful paint. Every millisecond the server spends before sending anything is a millisecond the 2.5-second budget has already lost.

And why better hosting often does not fix the metric. On most B2B sites, largest contentful paint is broken by an unoptimised hero image, a webfont blocking the headline, or third-party scripts. A faster server shortens the first step of a chain whose later steps are the problem.

The diagnostic that settles it. Compare your time to first byte against your largest contentful paint. If TTFB is 0.4 seconds and LCP is 4 seconds, hosting is not your issue and buying more of it will not help.

When hosting genuinely is the issue. TTFB above 1.8 seconds, or a TTFB that climbs sharply under traffic, which is the failure mode shared hosting produces and synthetic tests never show.

The two documented hosting arguments and their published qualificationsAnalysis of the two arguments most commonly used to sell website hosting, each presented alongside the qualification the documentation attaches to it. On server location, the search platform states that the server location is often physically near users and can be a signal about a site’s intended audience, but immediately adds that some websites use distributed content delivery networks or are hosted in a country with better web server infrastructure, so that it is not a definitive signal. The platform names what it uses instead to determine the target locale of a site: country code top level domain names; hreflang statements whether expressed in tags, in headers or in sitemaps; and other signals including local addresses and telephone numbers on the pages, the use of local language and currency, and links from other local websites. It further states that it does not attempt to vary the crawler source used for a single site in order to find any possible variations in a page, which rules out serving different content according to apparent crawler location. On response time, the documentation states that as a rough guide most sites should strive for a time to first byte of zero point eight seconds or less, with good values at or below that figure and poor values greater than one point eight seconds, while adding that because time to first byte is not a Core Web Vitals metric it is not absolutely necessary for sites to meet the good threshold provided this does not impede their ability to score well on the metrics that do matter, which is the stated reason the thresholds are described as a rough guide. Time to first byte nonetheless constitutes the floor beneath largest contentful paint, since every millisecond the server spends before sending anything is deducted from that metric’s two and a half second budget.Both arguments, with their next sentenceArgument one: server locationQuoted: “The server location is often physically near your users and can be a signal about yoursite’s intended audience.”Next sentence: “…so it is not a definitive signal.”Used instead: ccTLDs, hreflang in tags/headers/sitemaps, local addresses, language, currency, local links.Argument two: response timeQuoted: “most sites should strive to have a TTFB of 0.8 seconds or less”Next: “Because TTFB isn’t a Core Web Vitals metric, it’s not absolutely necessarythat sites meet the ‘good’ TTFB threshold.”Poor above 1.8 s. And it is still the floor under your 2.5-second largest contentful paint budget.Google also states it does not vary the crawler source for one site. Geo-varying content by crawler is not a strategy.
Both quoted lines are real. Both are qualified in the next sentence, and the qualification is the useful part. Source : Google multi-regional sites guidance and web.dev TTFB documentation (2026)

Where the 2.5 seconds actually goes

Hosting owns the first segment of a chain. Knowing the segments tells you whether buying more of it will help.

Segment one: the server. Time to first byte. Everything before a single byte reaches the browser. This is the part hosting owns, and the guide is 0.8 seconds.

Segment two: the blocking resources. Stylesheets, fonts and any script that blocks rendering. On a B2B site this is frequently longer than the server segment, and it is entirely under your control.

Segment three: the largest element itself. Usually a hero image. Whether it is correctly sized, in a modern format, and given priority.

Segment four: layout stability. Not part of the 2.5 seconds, but the same causes. Images without declared dimensions, banners injected late.

The arithmetic that decides your spend. If your budget is 2.5 seconds and the server takes 0.4, moving to a faster host might recover 0.2. If your hero image takes 2 seconds, optimising it recovers ten times more for no monthly cost.

Which is the common case. On most B2B sites the server is not the bottleneck. The bottleneck is a 3 MB header image and a font that blocks the headline.

When the diagnosis reverses. Database-driven pages, heavy plugin stacks, and any site where TTFB varies wildly between visits. There the server genuinely is the problem, and it usually shows as inconsistency rather than as a high average.

Decomposition of the largest contentful paint budget into segments and which segment hosting controlsDecomposition of the two and a half second largest contentful paint budget into its constituent segments, showing which segment website hosting actually controls. The first segment is server time, measured as time to first byte, covering everything that occurs before a single byte reaches the browser; this is the segment hosting owns and its published guide is zero point eight seconds or less, with values above one point eight seconds considered poor. The second segment covers blocking resources including stylesheets, fonts and any script that blocks rendering, which on a business-to-business website is frequently longer than the server segment and is entirely under the site owner’s control. The third segment is the largest element itself, usually a hero image, and depends on whether that element is correctly sized, delivered in a modern format and given loading priority. A fourth consideration, layout stability, falls outside the two and a half second budget but shares the same causes, namely images without declared dimensions and banners injected after initial render. The arithmetic determining whether hosting expenditure is worthwhile is that if the budget is two and a half seconds and the server consumes zero point four, moving to a faster host might recover approximately zero point two seconds, whereas if a hero image consumes two seconds then optimising that image recovers roughly ten times as much at no recurring cost. On most business-to-business websites the server is not the bottleneck; the bottleneck is an oversized header image together with a font that blocks the headline from rendering. The diagnosis reverses for database-driven pages, heavy plugin stacks, and any site where time to first byte varies substantially between visits, where the server genuinely is the constraint and typically manifests as inconsistency rather than as a high average.Where your 2.5 seconds goesServerBlocking resourcesThe largest element itselfHosting owns thisYour code owns thisUsually your hero imageServer takes 0.4 sA faster host might recover 0.2 s,every month, forever.Hero image takes 2 sOptimising it recovers ten times more,once, for nothing.The common B2B caseA 3 MB header image and a font blocking the headline. Neither is fixed by paying more for hosting.When the diagnosis reversesDatabase-driven pages, heavy plugin stacks, or a TTFB that varies wildly. That shows as inconsistency, not a high average.
Hosting owns the first segment. On most B2B sites the bottleneck is in the third. Source : web.dev LCP and TTFB documentation (2026)

What hosting actually decides

Six things, none of which is a ranking factor, all of which cost you money when they fail.

Whether the site is up. The unglamorous one. Availability is the only hosting property whose failure is total, and the only one your customers notice within minutes.

Response time under load. Not response time on a quiet Tuesday. What happens when a campaign, a newsletter and a conference all land in the same hour. This is where shared hosting fails and where synthetic tests never look.

Whether backups restore. Everyone has backups. Far fewer have restored one. A backup you have never tested is a belief, not a safeguard.

Whether a staging environment exists. The difference between testing a change and discovering it in production. This is a hosting decision more often than a budget one.

How quickly a human responds. At 2am on the day your certificate expires, the response time that matters is the support desk’s, not the server’s.

What it costs to leave. Proprietary caching layers, custom build steps and undocumented configurations are exit costs. They do not appear on the pricing page and they arrive on the day you want to move.

Changing hosting is a documented, mild case

Worth stating because it is often treated as risky and the guidance is reassuring.

Its own category. The documentation covers a change in hosting infrastructure, meaning switching hosting providers or moving to a content delivery network, and applies only to migrations that do not affect the user-visible URL.

The steps. Prepare the infrastructure, change the DNS, monitor traffic, decommission the old setup.

Which step is the move. The documentation identifies the DNS change as the actual site move step.

What is absent. Redirects. There is nothing to redirect, because nothing about your addresses changed.

What that means for the risk conversation. A hosting change is not a migration in the sense that a URL change is. The main real risks are configuration drift, a missed redirect rule that existed on the old server, and TLS certificate handling at cutover.

The one thing to check that people forget. That your old server stops answering. Two servers serving the same site during a DNS propagation window is how inconsistent behaviour appears and disappears for two days.

The six properties that website hosting genuinely determinesTable of the six properties that website hosting genuinely determines for a business, none of which constitutes a search ranking factor and all of which impose costs when they fail. The first is availability, the least glamorous property, being the only hosting characteristic whose failure is total and the only one customers notice within minutes. The second is response time under load, meaning behaviour when a campaign, a newsletter distribution and a conference coincide within the same hour, rather than response time measured on a quiet weekday, since this is the point at which shared hosting fails and the point synthetic testing never examines. The third is whether backups actually restore, since most organisations possess backups but considerably fewer have ever restored one, and an untested backup constitutes a belief rather than a safeguard. The fourth is whether a staging environment exists, which determines the difference between testing a change and discovering it in production, and which is more frequently a hosting decision than a budgetary one. The fifth is how quickly a human responds, since at two in the morning on the day a certificate expires the response time that matters is that of the support desk rather than of the server. The sixth is the cost of leaving, comprising proprietary caching layers, custom build steps and undocumented configurations, which constitute exit costs that do not appear on any pricing page and become apparent only on the day a move is attempted. Separately, changing hosting is documented as a distinct and comparatively mild category of site move, covering only migrations that do not affect user-visible addresses, in which the domain name system change is identified as the actual move step and no redirects are involved at all.What you are actually buyingNone of these is a ranking factor. All of them cost money when they fail.1. AvailabilityThe only failure that is total,and noticed in minutes.2. Response under loadNot on a quiet Tuesday.Synthetic tests never see this.3. Backups that restoreAn untested backup is abelief, not a safeguard.4. A staging environmentTesting a change, versusdiscovering it in production.5. A human who answersAt 2am, the response timethat matters is theirs.6. The cost of leavingProprietary caching, custombuilds, undocumented config.The diagnostic that settles the speed argumentCompare your TTFB to your LCP. TTFB 0.4 s and LCP 4 s means hosting is not your problem.And when hosting genuinely is the problemTTFB above 1.8 s, or a TTFB that climbs sharply under traffic. That second one is the shared hosting failure.
Availability, response under load, restorable backups, staging, human support, and the cost of leaving. Source : Method, against Google's hosting change guidance (2026)
The documented hosting change procedure and its principal residual risksSummary of the documented procedure for changing website hosting infrastructure together with the residual risks that remain. The search platform documents a change in hosting infrastructure as a distinct category of site move, defined as switching hosting providers or moving to a content delivery network, and states that its guidance applies only to migrations that do not affect the user-visible address. The documented steps are to prepare the new infrastructure, change the domain name system records, monitor traffic, and then decommission the old infrastructure, with the platform identifying the domain name system change as the actual site move step. Redirects are entirely absent from this procedure because no address changes, which makes a hosting change substantially milder than a migration involving address changes and means it should not be treated with the same level of concern. The residual risks are configuration drift between the old and new environments, a redirect or rewrite rule that existed on the old server and was not reproduced on the new one, and transport layer security certificate handling at the moment of cutover. The single item most frequently overlooked is confirming that the old server stops answering requests, since two servers responding for the same site during a domain name system propagation window produces behaviour that appears and disappears inconsistently over a period of one to two days and is extremely difficult to diagnose while it is occurring, because any individual test may reach either server depending on which resolver cache is consulted.Changing host: milder than people think1. PrepareThe new infrastructure.2. Change the DNSThis is the move.3. MonitorTraffic, on both.4. DecommissionThe old setup.No redirects anywhere in this. Nothing about your addresses changed.The real residual risksConfiguration drift. A rewrite rule that existedon the old server. Certificates at cutover.And the one people forgetConfirm the old server stops answering.Two servers, one site, for two days.Why that one is hard to diagnoseAny single test reaches either server depending on which resolver cache answered. Behaviour appears and vanishes.
No redirects, because nothing moved. But make sure the old server actually stops answering. Source : Google hosting change guidance (2026)

How to choose, and how to test the claim

Six checks, all of which you can run before signing anything.

Test response time under load, not at rest. Ask the provider for a load test, or run one against a trial instance. Rested response time is the number everyone advertises and nobody experiences.

Ask what a restore takes, in hours. Not whether backups exist. How long a full restore takes, who performs it, and when they last did one.

Confirm the staging environment is included. If it costs extra, budget it. Testing in production is more expensive.

Check TLS certificate handling. Automatic renewal, and what happens if it fails. Expired certificates take sites down completely and predictably.

Ask what leaving involves. If nobody can answer clearly, that is the answer.

And do not buy geography you do not need. If you sell internationally, spend on hreflang and a content delivery network rather than on a data centre in each market. Google says location is not a definitive signal, and names what it uses instead.

Six pre-purchase checks for selecting website hostingSix checks to perform before committing to a website hosting provider, none of which concerns search rankings. The first is to test response time under load rather than at rest, either by requesting a load test from the provider or by running one against a trial instance, since rested response time is the figure providers advertise and the figure no visitor ever experiences. The second is to ask how long a full restore takes measured in hours, rather than merely whether backups exist, together with who performs the restore and when they last carried one out. The third is to confirm that a staging environment is included in the package and, where it costs extra, to budget for it, on the basis that testing in production is more expensive than paying for staging. The fourth is to check transport layer security certificate handling, specifically whether renewal is automatic and what occurs if renewal fails, since expired certificates take websites down completely and predictably. The fifth is to ask what leaving the provider involves, on the principle that an inability to answer that question clearly constitutes the answer. The sixth is not to purchase geography that is unnecessary: an organisation selling internationally should spend on hreflang implementation and a content delivery network rather than on a data centre in each target market, since the search platform states that server location is not a definitive signal and names country code top level domains, hreflang statements, local addresses, language, currency and local links as what it uses instead.Six checks before you sign1. Load, not restThe advertised number is the one nobody experiences.2. How long does a restore take?In hours. Who does it. When they last did.3. Staging included?If it costs extra, budget it anyway.4. Certificate handlingAuto-renewal, and what happens if it fails.5. What does leaving involve?If nobody can answer clearly, that is the answer.6. Skip the geography upsellhreflang and a CDN, not a datacentre per market.None of these six is a ranking factorWhich is the point. Hosting matters for reasons that have nothing to do with search.And the platform names what it uses for geography instead: ccTLDs, hreflang, local signals on the page.
Load, restore time, staging, certificates, exit cost. And do not buy geography you do not need. Source : Method, against the documented location and TTFB positions (2026)

Where to go next

You are working on speed. Website speed: what matters.

You are moving the site properly. Website migration without losing rankings.

You want the search requirements. Technical SEO: three requirements.

You are choosing a platform, not just a host. The best website platform for B2B.

You are wondering what a site should cost. How much does a B2B website cost.

You are considering a rebuild. Signs it is time for a B2B website redesign.

In short

  • Server location “is not a definitive signal”, in Google’s own wording, one sentence after the line hosting is sold on.
  • What it uses instead: ccTLDs, hreflang in tags, headers or sitemaps, and local addresses, language, currency and links.
  • Google does not vary its crawler source for a single site, so serving content by apparent crawler location achieves nothing.
  • Time to first byte: 0.8 seconds or less is good, above 1.8 is poor. But it is not a Core Web Vital and the thresholds are called a rough guide.
  • It is still the floor under largest contentful paint, so a slow server makes the metric that counts harder to reach.
  • Compare your TTFB to your LCP before buying hosting. If TTFB is fine and LCP is not, the problem is your images, fonts or scripts.
  • A hosting change is a documented, mild category of move: no address changes, no redirects, and the DNS change is the move.
  • What hosting decides is uptime, load behaviour, restorable backups, staging, support and exit cost. None is a ranking factor.

Test response under load, and ask how long a restore takes. Book a diagnostic, or see how we approach B2B websites.