Choosing a website provider

1. Load their site on your mobile phone, on mobile data, not wifi. Count. If the page you came for is not usable in a few seconds, they have not solved the problem they are offering to solve for you.
2. Enter their address at pagespeed.web.dev and choose Mobile. If the top section shows real-visitor data, read whether it passes Core Web Vitals. If it says there is not enough data, that is a fact too: the site does not get many visitors. Below, the lab score is a rough guide; it varies from run to run, so treat anything within ten points as the same.
3. Do the same for two of their clients’ sites. Pick the ones they are proudest of. A provider’s own site can be a showpiece; the clients’ sites are the work.
4. Scan all three at developer.mozilla.org/en-US/observatory. It grades encryption and security headers from A+ to F in about ten seconds. A provider whose sites score below B has not thought about this.
5. Open one of the client sites in a private window. If a cookie banner appears before you have done anything, ask what it is for. Press Tab a dozen times and watch where the outline goes. Zoom to 200 per cent.

If a provider’s own site and their clients’ sites pass, you are talking to people who build what they sell. If they do not, their proposal is describing something they have not done.

We looked at ten firms competing for the same clients as us. Nine use testimonials. Four put a row of big numbers at the top of the homepage. One publishes measurements of its own site. None describes looking after the site after launch as part of the job; the usual offer is 30 or 60 days and then a retainer. Half publish prices, which is to their credit.

None of this is dishonest. It is what a proposal looks like when the measurements are not there. A testimonial is a claim by someone you cannot question. A big number without its method is decoration. A measurement with its tool, its date and its settings is something you can check yourself in the next five minutes, and that is the difference to look for.

Ask these of anyone, and listen for whether the answer is a number, a method or a mood.

1. What will the main content’s load time be on a mobile phone, for the slowest quarter of visitors, and how will you measure it? The answer should name a threshold, a tool and a method. “Blazing fast” is a mood.
2. What on the site will need patching, how often, and who will do it? Count the components in the answer. On a WordPress build with 30 plugins the honest answer is “30 things, within days of each fix, and that is a job”. On a static build it is a short list.
3. If we leave you, what do we take with us, and in what form? The code, the content and the domain, as files you can read, with a document saying how it fits together. Anything held in a system only they can operate is a lock.
4. What happens to our existing addresses? Every one, kept or redirected, with the list shown before launch. “We’ll set up redirects” is not the same as a list.
5. What do you do after launch, and what does it cost? A cadence and a price, or an honest “nothing”. Build-and-forget is the commonest way a good site becomes a bad one.
6. When would you tell us not to do something? A provider who has never advised against a feature, a redesign or an AI project has never put your interest above the invoice.

An engineer answers with numbers, methods and lists, and will say “that won’t pay off” when it won’t. A salesperson answers with adjectives and a discount. Both may be perfectly pleasant. Only one of them is accountable for the result, because only one of them has said what the result will be in terms you can measure.

Run the five-minute test on this site and on the sites in our Work section. Then ask us the six questions. An engineer will answer within one working day, with numbers.