Best in Show · The agency study
August 2026 edition · Framework v1.1
What I found on 50 UK agency websites.
I looked at 50 UK agency websites to see how they compared with the public-page checks in my SEO + AI Standard. The results offer a view of how agencies present their own work online, and a chance to examine the framework itself.
The average score was 40.2 out of 100 under this method.
The study looks at selected public pages. It doesn't measure the quality of an agency's client work, its commercial results or how often it appears in AI answers.
Scored 19th August 2026 against the Modern SEO + AI Optimisation Standard: the 13 of 32 checks observable from public pages, re-based to 100. Full methodology.
What I did
On 19th August 2026, I assessed each agency's homepage, About page and one recent blog post. The study used 13 publicly observable checks from the 32-check framework, covering 47 of its 100 weight points. This was version 1.1 of the framework. Those results were then rescaled to 100. The other checks were excluded from the score. A full assessment could produce a different result.
The framework behind the study
The SEO + AI Standard is my working framework, informed by the sources linked in the guide. Its checks and weights reflect my judgement. Publishing them makes the assessment easier to inspect, discuss and challenge.
Read the current guide, or inspect the archived v1.1 criteria used in this study.
How the scores were distributed
The mean was 40.2 out of 100. Under the version of the framework used for this study, no agency reached the “Strong” band. These bands describe the assessment results, rather than independently measured visibility. I discuss what stood out to me in my account of the August findings.
These are the bands used in the August 2026 framework. They describe the assessment criteria, not independently measured visibility in AI answers.
Scores across the sample
Each column counts agencies in a 10-point score range, with the named top 10 shown individually in pink and the other 40 grouped as a grey block with a count.
View as a table
| Score range | Agencies | Named in the top 10 | Unnamed |
|---|---|---|---|
| 90 to 100 | 0 | 0 | |
| 80 to 89 | 0 | 0 | |
| 70 to 79 | 0 | 0 | |
| 60 to 69 | 2 | #1 Opace #2 Rapid SEO London | 0 |
| 50 to 59 | 6 | #3 SALT.agency #4 Vixen Digital #5 Get Found #6 Dark Horse #7 StudioHawk UK #8 Sierra Six Media | 0 |
| 40 to 49 | 20 | #9 Hallam #10 Artemis Marketing | 18 |
| 30 to 39 | 11 | 11 | |
| 20 to 29 | 9 | 9 | |
| 10 to 19 | 2 | 2 | |
| 0 to 9 | 0 | 0 |
Ranges: 40 to 49 covers 40.0 to 49.9, and so on.
Four patterns in the assessment
Expert contributions
The assessment marked 44 of the 50 sampled blog posts as having no credentialed expert quotation. That describes this criterion in these posts. It doesn't establish whether a quotation would improve every article or whether expertise was shown in other ways.
Attributed statistics
The assessment marked 43 of the 50 posts as having no attributed statistics. Where a numerical claim is relevant, a source helps readers check it. Not every subject needs a statistic.
A clear factual reference
Two agencies fully met the framework's canonical fact-base criterion, while 36 partly met it and 12 did not. The result suggests different levels of coverage under that definition. It does not mean the other agencies publish no reliable information about themselves.
Crawler files and access rules
The study recorded llms.txt files for 23 agencies and robots.txt rules blocking at least one checked AI crawler for two. These observations do not show whether a file is used by an answer engine, nor whether every system can access the site.
An established reputation and a score on this framework describe different things. The results should be read alongside the limited sample and method, rather than as a ranking of the agencies' overall quality.
Results for the 13 scored checks
The 13 checks scored, ordered by the number marked “No”. Hover a segment for the counts.
The top 10
These agencies received the highest scores in this sample. Opace ranked first at 68.1 and Rapid SEO London second at 64.9. The table includes the variation recorded across repeated assessments.
| # | Agency | City | Score | Band | 3-run range |
|---|---|---|---|---|---|
| 1 | Opace | Birmingham | 68.1 | Emerging | 62.8 to 76.6 |
| 2 | Rapid SEO London | London | 64.9 | Emerging | 52.1 to 69.1 |
| 3 | SALT.agency | Leeds | 59.6 | Emerging | 56.4 to 63.8 |
| 4 | Vixen Digital | Brighton | 57.4 | Emerging | 48.9 to 61.7 |
| 5 | Get Found | Birmingham | 54.3 | Emerging | 54.3 to 57.4 |
| 6 | Dark Horse | Manchester | 54.3 | Emerging | 44.7 to 54.3 |
| 7 | StudioHawk UK | London | 50 | Emerging | 50 to 54.3 |
| 8 | Sierra Six Media | London | 50 | Emerging | 46.8 to 60.6 |
| 9 | Hallam | Nottingham | 48.9 | Exposed | 48.9 to 53.2 |
| 10 | Artemis Marketing | Sussex | 48.9 | Exposed | 48.9 to 50 |
Vixen Digital placed fourth in the sample and Artemis Marketing tenth. Both are based in Sussex. I'm based in Brighton, so it was nice to see a couple of local names in this part of the table.
How the study was carried out
Wording and scope explanation revised on 23rd September 2026. The recorded August scores are unchanged.
The study assessed 13 public-page checks from version 1.1 of the framework. It excluded process and off-site checks rather than assuming how the agencies work. A full assessment could produce a different result.
What was scored, and what deliberately wasn't
The 13 checks cover 47 of the framework's 100 weight points. These results were rescaled to 100. They do not measure actual search rankings or visibility in AI answers.
Crawler-related files were recorded as observations, not scored as proof that every AI system could access the site.
How each agency was assessed
On 19th August 2026, I fetched each agency's homepage, About page and one recent blog post. Scripts checked features such as headings, structured data and crawler-related files. An AI model assessed the content against the framework's criteria. I checked the selected blog URLs by hand.
For sites with an older latest post, I assessed the post available at the time. The date and limited sample matter when interpreting the result.
Repeated assessments
AI assessments can vary between runs. The recorded dataset contains three assessments for 19 agencies, including the published top 10, and one assessment for the other 31. For repeated assessments, the most frequent status was used for each check.
The table shows the range for the named results. This is variation between recorded assessments, not a statistical confidence interval. The entries do not all have the same repeat-check coverage.
How the sample was chosen
I selected agencies from UK search results for agency terms and added established industry names. The searches covered London, Manchester, Birmingham, Leeds, Brighton and the UK overall. Dog on the Table clients and partners were excluded. This was a selected sample, not a representative survey of every UK agency.
The final sample includes agencies of different sizes. It should not be interpreted as applying a uniform minimum staff count.
Edge cases worth knowing about
The recorded checks distinguish genuine llms.txt files from error pages. Plugin-generated files count as present. For one site that blocked the initial fetch, the study used a rendering service. These observations describe access during the assessment, not every crawler's access today.
This page records the August 2026 study. Any future edition will state its assessment date and method, including changes that affect comparisons.
Open data and how to cite it
Download the data
The CSV and JSON files hold the study's results and method data, without evidence captures or data about any single unnamed agency. They're free to reuse, share and adapt with credit to Dog on the Table under CC BY 4.0.
Licence: CC BY 4.0 · Dataset version 1.0 · August 2026 edition
Cite this
Dog on the Table (2026). Best in Show: a study of 50 UK agency websites, August 2026 edition [dataset, version 1.0]. https://dogonthetable.com/tools/seo-ai-standard/agency-study/. Licence: CC BY 4.0.
Competing interests
I sell AI visibility services through Dog on the Table, including to agencies. Dog on the Table isn't scored or ranked in its own index. No agency paid to be included or to change its result.
Questions about the study
Why did you carry out the study?
I wanted to apply the framework to a defined group of public websites and see which patterns emerged. Agency websites are a useful sample for that question because search and digital marketing are part of their work. The results also help me examine where my own method needs to improve.
Does a lower score mean an agency does poor work?
No. The score covers selected evidence on three public pages. It does not assess the agency's people, client work or results.
Why are some agencies unnamed?
The page names the 10 highest-scoring agencies. The other 40 appear only as counts in each 10-point score range, with no name, city, rank or score. A score from three public pages says little about an agency's people or client work, and I don't want the lower half turned into public sport. Any agency in the study can ask me for its own result.
How do I ask about a finding?
Tell me which agency is yours and which finding you'd like to discuss. I can explain the recorded evidence and review a specific concern.
Discuss a result or the method.
If your agency is included, I can share its result and check the recorded evidence if a fact looks wrong. Questions about the method are welcome too.