Free guide:winning new clients predictably in 2026 · freeFree guide: win clients predictablyGet it now →

How to tell whether an agency takes its own website seriously

Before you speak to an agency, there is one work sample in plain view that nobody has picked out for you: its own website. Five checks that take a quarter of an hour and need no judgement of taste: speed for real visitors, what loads before you consent, who stands behind the legal notice, whether the site works without a mouse and whether the references can be verified.

Cover: How to tell whether an agency takes its own website seriously

A company is looking for an agency to relaunch its website. Three names are on the list, all three show attractive references, all three promise strategy, design and technology from a single source. Before anyone books a first meeting, there is one work sample in plain view that nobody has picked out for you: the agency's own website. When you are choosing an agency, that is where you can see how it works when no client sets the brief and no deadline is pressing. You do not need a verdict on colours or typefaces for this. You need five checks that take a quarter of an hour and whose result does not depend on taste. Below: why this work sample weighs more than the reference list, how fast the site is for real visitors, what it loads before you have consented, what the legal notice tells you, whether it works without a mouse and how to verify references.

Why does the agency's own website weigh more than its reference list?

Because the decision is made earlier than you might think. For its Buyer Experience Report 2025, the sales software provider 6sense surveyed more than 4,000 business buyers in North America, Europe, the Middle East, Africa and the Asia-Pacific region, published on 12 November 2025. 94 per cent of buying groups had already ranked their preferred vendors before they spoke to any of them for the first time. In 77 per cent of cases they ultimately bought from that preliminary favourite. The study covers B2B purchases in general, not agencies specifically, and 6sense sells software designed to make exactly this early phase visible. The finding still matches what anyone who has looked for an agency will recognise: the shortlist is drawn up on screen.

Two key figures with bars from the 6sense Buyer Experience Report 2025: on the left 94 per cent, the share of buying groups that had ranked their preferred vendors before speaking to any of them. On the right 77 per cent, how often they ultimately bought from that preliminary favourite.
Preferred vendors ranked before first contact: 94%. Ultimately bought from the preliminary favourite: 77%. Source: 6sense, Buyer Experience Report 2025, press release of 12 November 2025, more than 4,000 B2B buyers surveyed in North America, EMEA and APAC. Vendor study on B2B purchases in general, not on agency selection specifically.

References are weak evidence at this early stage. They are selected, they show what a client wanted and paid for, and you cannot see how much of it the agency decided. On its own website, the agency decides everything itself. One caveat is only fair: at many agencies the own site comes last, because client projects have deadlines and the own site does not. An older team photo or a post from last year therefore says little. What does say something are the things that cost care rather than budget: speed, consent, legal notice, usability and evidence. How a reference has to be built to convince is covered in our post on case study pages that sell. This one is about the site those references sit on.

How fast is the site for real visitors?

Open PageSpeed Insights and enter the agency's address. At the top you see how the site loaded for real Chrome users over the last 28 days; the lab test follows below. The difference matters. The upper values are field data from the Chrome User Experience Report, showing the three Core Web Vitals for loading, responsiveness and visual stability. The lower part is a Lighthouse run: a single session on a simulated mid-range smartphone on a mobile network. Google itself states in the PageSpeed Insights documentation that the two results may differ.

How demanding the field test is can be seen in the HTTP Archive's Web Almanac 2025, published in January 2026 and based on worldwide Chrome field data from July 2025. On smartphones, 48 per cent of websites pass all three values, up from 36 per cent in 2023 and 44 per cent in 2024. On desktop the figure is 56 per cent. Taken one by one, loading is where most sites fall short: only 62 per cent of mobile websites display their largest content quickly enough, 77 per cent manage responsiveness to input and 81 per cent visual stability.

Bar chart from the Web Almanac 2025, smartphone, share of websites with a good value: all three Core Web Vitals good 48 per cent, loading of the largest content (LCP) 62 per cent, responsiveness to input (INP) 77 per cent, visual stability (CLS) 81 per cent. The overall value and loading are highlighted.
Smartphone, share of websites with a good value: all three Core Web Vitals 48%, loading (LCP) 62%, responsiveness (INP) 77%, visual stability (CLS) 81%. Source: HTTP Archive, Web Almanac 2025, Performance chapter, published on 15 January 2026, Chrome User Experience Report field data from July 2025, worldwide.

For your check, that means two things. If an agency's field data passes, it belongs to the slim half that manages it. If there is no field data at all, that is no black mark: Chrome only includes addresses with a minimum number of visitors in the dataset, and according to the Chrome User Experience Report methodology Google does not disclose that number. Many smaller agency sites therefore only have the lab test. Look more closely when an agency advertises a Lighthouse score of 100. That figure proves a test run, not the experience of your future visitors. The better question for the meeting is: what field data do the client sites you have built show?

What does the site load before you have consented?

The second check takes two minutes. Open the site in a private browser window and look at the consent banner before you click anything. Is there a button to reject on the first layer that is as easy to find as the one to accept? If you want to look more closely, press F12 beforehand to open the browser's developer tools, switch to the Network tab and reload the page. If addresses of analytics or advertising services already appear while the banner is still open, measurement starts before you have decided. Legally, any access to your device that is not strictly necessary for the service requires consent: in Austria under Section 165(3) of the Telecommunications Act 2021, in Germany under Section 25 TDDDG. What consent management involves is explained in the glossary.

This cannot be taken for granted. For a study published at USENIX Security 2024, a team at ETH Zurich led by Ahmed Bouhoula examined 97,000 websites that are widely visited in the EU, selected from the Chrome User Experience Report of March 2023. Of the 48,843 sites with a cookie notice, 73.4 per cent probably set analytics or advertising cookies before anyone had interacted with the notice. Of the 16,231 sites with a reject button, 65.4 per cent still did so after it had been clicked. 56.7 per cent of the notices examined offered no way to reject at all. The researchers speak of likely violations because the cookies were classified automatically.

Dot grid of 100 dots, 73 of them filled: 73.4 per cent of 48,843 websites widely visited in the EU with a cookie notice probably set analytics or advertising cookies before anyone had interacted with the notice. Alongside, two further values: 65.4 per cent of the 16,231 sites with a reject button still did so after rejection, 56.7 per cent of notices offered no way to reject.
Analytics or advertising cookies before any interaction: 73.4% of 48,843 sites with a cookie notice. Even after rejection: 65.4% of 16,231 sites with a reject button. No reject option offered: 56.7% of notices. Source: Bouhoula, Kubicek, Zac, Cotrini and Basin (ETH Zurich), Automated Large-Scale Analysis of Cookie Notice Compliance, USENIX Security 2024. 97,000 websites widely visited in the EU from the Chrome User Experience Report of March 2023, automated measurement, described by the authors as likely violations.

For an agency this is more than a legal question. It will install the same tools on your site, and how it has handled them on its own is the most honest preview you will get. Which other questions matter with third-party services is covered in our post on data protection on your website.

The legal notice looks like a formality, but it is the one place that tells you with legal force who stands behind the offer. The imprint obligation in Austria, under Section 5 of the E-Commerce Act, requires among other things the name or company name, the geographic address, contact details including an email address and, where applicable, the company register number and register court, permanently, easily and directly accessible. In Germany, Section 5 of the Digital Services Act (DDG) has governed this since May 2024 with similar details, including legal form and authorised representatives for companies. Missing details can cost up to 3,000 euros in Austria under Section 26 ECG and up to 50,000 euros in Germany under Section 33 DDG.

For your selection, the content matters more than the penalty. Is it a company or a sole trader? Both are fine, but you should know who your contractual partner is and who is liable. Do the names in the legal notice match those on the team page? Is the address a real place of business, and is there an email address next to the form that reaches a person directly? Is the agency based in the country whose law is to govern your contract? A complete legal notice does not prove good work. An incomplete one does show how seriously someone takes obligations that nobody checks until there is trouble. Why address and contact details should read the same everywhere is covered in our post on address, opening hours and phone number.

Can you get through the site without a mouse?

Put the mouse aside and press the Tab key. Can you see where you are at every step? Can you reach the menu, the services and the contact form, and can the form be submitted? Then zoom the page to 200 per cent in your browser. Does everything remain readable and usable? These two checks take three minutes and reveal what no score shows. The framework for them is set by the WCAG, the testable criteria for accessibility.

The WebAIM Million 2026 shows how often something is missing here. In February 2026, the US organisation WebAIM automatically tested the home pages of the one million most visited websites. 95.9 per cent had detectable WCAG 2 failures, 56.1 per page on average. The most common: low text contrast on 83.9 per cent of pages, missing alternative text for images on 53.1 per cent, form fields without labels on 51 per cent and links without discernible text on 46.3 per cent. WebAIM itself points out that automated tools cannot detect every failure. The real picture is therefore more likely to be worse.

Why this matters when choosing an agency: contrast, alternative text and labelled form fields are not a question of budget but of craft. An agency that has mastered them on its own site does not need to learn them on yours. When such text is a legal requirement and how much work it takes is covered in our post on subtitles, transcripts and alt text.

Can the references be verified?

The last check takes you away from the agency's site. Take two references and open the clients' websites yourself. Do they still exist, and do they look like the agency's images? Does the reference name the client, a time period and which part the agency handled: concept, design, technology or all of it? Then repeat the first three checks on those sites: field data, banner, Tab key. An agency's own website can be the exception, for better or for worse. Two client sites show what it delivers day to day. Who the people behind the projects are is, at best, shown on the team page. Why that counts for more than any photo is covered in our post on the team page as a sales page.

Five-step sequence for checking an agency website in around 15 minutes: 1. Speed, look at field data in PageSpeed Insights, about 3 minutes. 2. Consent, check the banner in a private window and the Network tab, about 2 minutes. 3. Legal notice, identify the partner by legal form, address, email and names as on the team page, about 2 minutes. 4. Usability, Tab key and 200 per cent zoom, about 3 minutes. 5. References, open two client sites and repeat checks 1 to 3, about 5 minutes. Editorial assessment from our own work, not a measurement.
Five checks, around 15 minutes in total: speed (field data in PageSpeed Insights, about 3 minutes), consent (banner in a private window, Network tab, about 2 minutes), legal notice (legal form, address, email, names as on the team page, about 2 minutes), usability (Tab key, 200% zoom, about 3 minutes), references (two client sites, repeat checks 1 to 3, about 5 minutes). Editorial assessment from our own work, not a measurement.

Three levers for choosing your agency

First: check every candidate against the same list. Five checks, a quarter of an hour per agency, the results in one table. Only the comparison turns a single finding into a statement.

Second: take the findings into the first meeting as questions, not as grounds for exclusion. Something like: your banner loads an analytics service before consent, how do you handle that for clients? The answer shows how an agency deals with feedback, and you will need that more often during the project than any presentation.

Third: ask to see field data from client sites, not scores. A Lighthouse score is a test run. The values of real visitors on a site the agency has built are the evidence you are actually after.

If you are comparing agencies right now: ask us the questions on this list in the very first conversation. 🤝