We measured 15 agency websites: how do they hold up?

Speed, page weight, llms.txt, structured data and blog dates: what we measured on 15 digital agency websites, and how to run the same checks yourself.

· 14 min readHow we work

Short answer

On 10 November 2026 we measured the home pages of 15 digital agency websites. The median mobile performance score was 69; 3 sites scored above 90 and 4 below 50. 4 sites met the 2.5-second LCP threshold, 7 had an llms.txt file and 5 had invalid structured data.

You have just left a first meeting with an agency. The presentation mentioned site speed, Core Web Vitals and visibility in AI search, and all of it sounded right. There is a short way to see what stands behind those words: the agency’s own website is open to everyone and can be measured with free tools in a few minutes. Measuring an agency website will not make the decision for you, but it sharpens the question you ask in the next meeting.

We did this for 15 digital agency websites. On 10 November 2026 we measured the home pages of 15 agencies serving Türkiye with Lighthouse, on mobile settings. On each site we also looked at the llms.txt and robots.txt files, the structured data, the security headers and the date of the latest blog post. We do not name the sites; our aim is to show the overall picture rather than to rank anyone. Our own site is not in the sample. The method and its limits are set out at the end of the article.

The table below summarises the findings; the sections that follow take each row in turn.

Check Result across 15 sites
Mobile performance score Median 69, lowest 33, highest 99
Score of 90 or above 3 sites
Score below 50 4 sites
LCP within 2.5 seconds 4 sites (median 4.2 seconds)
CLS within 0.1 15 sites
TBT within 200 milliseconds 6 sites (median 370 milliseconds)
Passing all three lab thresholds 2 sites
Home page weight Median about 1.5 MB; from 0.25 MB to 32.6 MB
Has an llms.txt file 7 sites (2 of them stale or auto-generated)
Names AI crawlers in robots.txt 3 sites
Invalid or placeholder structured data 5 sites
Star rating declared without visible reviews 3 sites
No blog post in the last 6 months 5 of 14 sites
No blog post in the last 12 months 4 of 14 sites
Sends an HSTS header 5 sites
Sends a CSP header 2 sites

Mobile performance: 3 of 15 sites in the green

Lighthouse sorts the performance score into three bands: 0 to 49 poor, 50 to 89 needs improvement, 90 to 100 good. Of the 15 home pages we measured, 3 landed in the good band, 8 in the middle and 4 in the poor band. The lowest score was 33, the highest 99 and the median 69.

Lighthouse mobile performance scores for the home pages of 15 agency websites, sorted from lowest to highest 50 90 0 33 35 37 41 61 67 68 69 72 75 76 89 94 96 99 Poor (0 to 49) 4 sites Needs improvement (50 to 89) 8 sites Good (90 to 100) 3 sites Lighthouse mobile performance scores for the home pages of 15 agency websites, sorted from lowest to highest 50 90 0 33 35 37 41 61 67 68 69 72 75 76 89 94 96 99 Poor (0 to 49) 4 sites Needs improvement (50 to 89) 8 sites Good (90 to 100) 3 sites
Figure: Each bar is one site’s home page; 4 of the 15 sites fell in the poor band, 8 in needs improvement and 3 in good, with a median score of 69.

The 66-point gap between best and worst undoes the assumption that all agency websites are more or less alike. The other headings in the same report, however, hardly separate the sites. Under Lighthouse’s SEO heading 10 sites scored 100 and the remaining 5 scored 92. For accessibility the median was 92 and 3 sites got full marks. That is why an agency saying “our SEO score is 100” tells you little by itself: two thirds of the sample get the same score.

Core Web Vitals: 2 sites pass all three thresholds

The limits Google counts as “good” for Core Web Vitals are clear: LCP of 2.5 seconds or less, CLS of 0.1 or less, INP of 200 milliseconds or less. INP needs a real user interaction, so it cannot be measured in the lab; Lighthouse’s TBT metric is used in its place, and on mobile a value of up to 200 milliseconds counts as green.

  • LCP (the main content appearing): 4 sites stayed within 2.5 seconds, 2 sites were between 2.5 and 4 seconds, and 9 sites went over 4 seconds. The median was 4.2 seconds. On 3 sites the value was above 10 seconds; the highest measurement was 29.9 seconds.
  • CLS (the page shifting as it loads): all 15 sites are within the threshold. The highest value was 0.047.
  • TBT (the time scripts lock up the page): 6 sites within 200 milliseconds, 4 sites between 200 and 600, and 5 sites above 600. The median was 370 milliseconds.

Two sites passed all three thresholds at once. The picture is this: layout shift is a solved matter in this sample, and the real issues are main content arriving late and too much JavaScript.

Page weight: roughly 130 times between lightest and heaviest

The total data downloaded as the home page opened ranged from 0.25 MB to 32.6 MB; the median was about 1.5 MB. 4 sites stayed under 1 MB, 6 sites went over 2 MB and 4 sites over 4 MB.

On 3 of the 4 sites above 4 MB, the Lighthouse report pointed plainly at the cause: images that had not been compressed or converted to a modern format. On one of them a single animated image made up most of the page. Images are the easiest item to fix on a website; when they are left unfixed, it points less to a technical difficulty than to a missing check before publishing. We explained how speed relates to sales, and which metric tells you what, in why a fast site sells more.

AI search: llms.txt and robots.txt

7 of the 15 sites had a file at /llms.txt. 5 of those 7 files looked tidy and current. One was dated about 15 months before our measurement and contained none of the pages the site had published this year. Another had been generated automatically by a plugin, and the code of a contact form had been dumped into it.

On the robots.txt side, 3 sites listed AI crawlers by name and allowed them, and 1 site used a general permission line without names. The remaining 11 sites had no rule at all for these crawlers; where there is no rule, crawlers can enter the site by default. We did not come across a site that blocked the assistants’ crawlers.

llms.txt is not an official standard and its presence alone brings no result. Even so, if an agency offers you a visibility service for AI search, expecting its own file to be current and clean is a reasonable test.

Structured data: present, but not always valid

All 15 sites had some kind of structured data (JSON-LD). The problem was not presence, it was accuracy.

On 5 sites the markup either could not be parsed at all, or was structurally broken, or carried placeholder values left over from a template. Examples we saw: a question-and-answer block that could not be read because of a missing comma and a question list whose first question had been left empty. We also saw server code left unprocessed in an address field and a price field holding text where a number belongs.

On 3 sites a star rating and a review count were declared for which we could not see matching reviews on the page. The number of sites falling into at least one of these two groups is 7. We showed how to check structured data on your own site in step eight of the technical SEO checklist.

Blog: when was the latest post published

One site’s blog sat on a separate subdomain we could not reach; on the other 14 we established the date of the latest post. 8 sites had published within the last three months. 5 sites had no new post in the last six months, and 4 of those had none in the last twelve. The oldest “latest post” dated from 2020.

On at least 3 of the 8 sites that looked current, most of the posts were packed into a single day or a single week. How posts are spread across the calendar tells you as much as the date of the last one: a regular content programme and a one-off bulk upload do not look the same.

Security headers

Security headers are short lines in which the server tells the browser how to protect the connection and the page. 5 of the 15 sites sent an HSTS header and 2 sent a CSP header. 8 sites had none of the HSTS, CSP and X-Frame-Options headers. Missing headers do not make a site unsafe by themselves; they do give an idea of how a team that builds websites configures its own server.

How to run the same five checks yourself

For the website of an agency you are talking to, the five steps below take half an hour in total and need no account.

  1. Speed and weight. Paste the agency’s home page into PageSpeed Insights and look at the mobile tab. Note three numbers: the performance score, LCP and the total network payload line in the report. If real-user data appears at the top of the report, take it more seriously than the lab score.
  2. llms.txt and robots.txt. Open theagencysite.com/llms.txt and theagencysite.com/robots.txt in your browser. Is the file there, does it mention the site’s current pages, and is it readable text or a heap of code?
  3. Structured data. Give the home page and one service page to Google’s Rich Results Test. Look at the number of errors; if you see a star rating, check whether those reviews can really be read on the page.
  4. Blog dates. On the blog page, look at the dates of the top three posts. When was the last one, and are the posts spread over months?
  5. Security headers. Open the Network tab in your browser’s developer tools, reload the page, click the first row and look for the strict-transport-security and content-security-policy lines in the response headers.

You can apply the same steps to your own site; for something fuller, the technical SEO checklist works through nine steps. We are an agency too, and we would like these five checks to be run on our site as well.

Common mistakes when reading the results

  • Judging from a single measurement. A Lighthouse score moves from run to run. Measure the same page three times and take the middle value.
  • Taking the lab score for real-user data. A lab measurement simulates one device and one connection; what Google’s assessment rests on is field data collected from real visitors.
  • Inferring the whole site from the home page. The home page is often the heaviest page on a site. Measure a service page too.
  • Treating a good score as proof of a good agency. A fast site does not show that the agency will manage your work well. A weak score should not be a reason to rule anyone out alone either; ask the agency why, and see how concrete the answer is.

Measurement is only one part of choosing. We gathered the questions on your point of contact, account ownership and the contract in how to choose a digital agency, and what to look for in an audit report in how to choose an SEO agency. In our SEO service, the first month goes to a technical audit that includes the checks described here.

Method and limits

  • Sample: 15 digital agency websites. We chose them from among the sites that came up in the searches someone looking for an agency would run. It is not a random sample and does not represent every agency in Türkiye.
  • Date: All measurements were taken on 10 November 2026.
  • Tool: Lighthouse 12.8.2, mobile profile, default throttling. Each page was measured once; the performance figures in this article belong to home pages.
  • Environment: The measurements were run from a machine in a data centre. Absolute values differ from the field data seen on real visitors’ phones. In some measurements third-party resources such as analytics, fonts and video could not be reached; on those sites the real weight and TBT may be higher than what we measured.
  • What we looked at: Lighthouse scores and metrics, home page weight, the llms.txt and robots.txt files, the JSON-LD on pages, response headers and the date of the latest blog post.
  • What we did not look at: Search rankings, traffic, backlinks, field data (CrUX), client work and service quality. This measurement does not say how well an agency works for its clients.
  • The margin of a single run: A site scoring close to the 50 or 90 boundary could move into the neighbouring band on a second run. The overall shape of the distribution is not affected by this; the position of individual sites may be.

Common questions

Should I avoid an agency whose own website is slow?

Do not treat it as a reason to rule them out alone, but do ask. There are good teams that have not found time for their own site. What decides it is the quality of the answer: can they say concretely what causes the problem and when it will be fixed, or do they play the measurement down? If you are buying a speed or technical SEO service, the question carries even more weight.

Is a Lighthouse score the same as the speed Google sees?

No. Lighthouse is a lab measurement: it measures a single load with a simulated phone and connection. The Core Web Vitals assessment looks at the 75th percentile of real visits. The lab score helps you find where the problem is; field data shows how the site is really experienced.

There is no such rule. llms.txt is an optional file, and 8 of the 15 sites we measured did not have one. Whether crawlers can enter a site is decided by robots.txt; if nothing is blocked there, the pages can be read. If the file exists it has to be kept current, because a stale list describes the site as it used to be.

If you do just one thing today, do this: open PageSpeed Insights in two tabs, paste the home page of the agency you are talking to into one and your own home page into the other. Write the mobile performance score and LCP side by side, then show both numbers to the agency at the next meeting and ask how they read the difference.

Sources

This article is general information, not legal, financial or technical advice. Platforms, their rules and the law change over time, so some details may have changed since 10 November 2026 or contain small errors. Before you make a decision for your business, check the current official sources or write to us.

Tell us about
your project.

Fill in the form or write to us directly, and we will agree on a day to talk.

Your first meeting is with Deniz Sezer, founder of Detmore.

First meeting free · Reply within one working day · Written proposal in 2 to 3 days · NDA ready

What do you need? pick as many as you like

Name, email and phone are required.

Contact

We reply within one working day.