How to design a technical SEO benchmark

What to take away

Decide first whether your benchmark describes your own site over time or a defined set of comparable sites, because the two need different data and different caveats.
Write the population, the metric definitions and the collection window down before you crawl anything, so the numbers survive scrutiny later.
Use published UK data and legislation only where they genuinely apply, and name the source and its date in the sentence.
Keep benchmark evidence separate from tool output so readers can see which figure came from where.

Should your benchmark compare sites or track one site?

A benchmark is only meaningful once you fix what is being compared. Cross-site benchmarks suit questions about relative performance, such as how many comparable pages return a soft 404. Longitudinal benchmarks suit questions about your own trajectory after a migration or template change.

Cross-site vs longitudinal benchmark

Cross-site

Question: Relative performance
Example: Soft 404 rate
Risk: Swapping sites

Longitudinal

Question: Own trajectory
Example: Post-migration change
Risk: No baseline

Name the comparison set before you run a crawl. A vague population lets you swap sites in and out until the result flatters you.

How do you define the population before crawling?

Write down the inclusion rules: domain type, sector, page templates, language and a crawl date. Record how many URLS you expected and how many you actually fetched.

Population inclusion rules

Domain type
Sector
Page templates
Language
Crawl date
Expected vs fetched URLs
Fetch success rate

Publish the fetch success rate. For example, a crawl that reached 400 of 1,000 seed URLS cannot support firm conclusions about the population.

Where a public body is in scope, check the accessibility requirements under the Public Sector Bodies (Websites and Mobile Applications) (No. 2) Accessibility Regulations 2018 before scoring accessibility. Covered websites and apps must meet WCAG 2.1 AA; WCAG 2.2 can be recorded as an additional benchmark, not as the statutory requirement.

Sample size should be stated, not implied. A benchmark built on eleven sites is a pilot, and should be labelled as one.

Which metrics belong in a defensible benchmark?

Choose metrics that a tool can reproduce. Screaming Frog SEO Spider and Sitebulb can crawl sites to check indexability, canonical consistency and response codes; Google Search Console provides data about your own site's search performance, while Lighthouse audits page performance. Structured data validity and Core Web Vitals field data are also measurable.

Keep the underlying counts alongside any composite score a tool produces. State any weighting behind that score, and keep the unweighted table so a reader can re-score it.

For each metric, note the tool, version and date. For wider context, the HTTP Archive and its Web Almanac publish crawl-based data and analysis; the Chrome UX Report provides field data for Core Web Vitals. Record the dataset or release, metric definition and extraction date so readers can identify the evidence. Your technical SEO measurement and reporting guide 2027 sets out how those definitions should flow into client reporting.

What legal and public data sources can you cite?

If your benchmark touches personal data, identify the applicable UK GDPR and Data Protection Act 2018 duties, and consult Information Commissioner's Office (ICO) guidance rather than paraphrasing from memory. This matters when crawls capture personal data or when analytics exports feed the benchmark. Record the specific UK GDPR article or Data Protection Act 2018 provision and the ICO guidance relied on, and date the citation.

If you use a National Audit Office report for context on public sector digital delivery, record its exact title, publication date and relevant passage. Do not treat a collection of reports as an undifferentiated dataset.

How do you run the benchmark step by step?

Run the benchmark step by step

Write the research question in one sentence and name the decision it informs.
Define the population, inclusion rules and crawl date.
Fix metric definitions and the tools that will produce them.
Run the crawl or export, and log failures rather than deleting them.
Clean the data against the definitions, documenting every exclusion.
Calculate results with ranges, not single figures, where the sample is small.
Publish the method alongside the findings, including limitations.
Archive the raw export and the cleaning script so a later reviewer can rerun the calculation.

How should the findings be reported?

Report the method first, then the results, then the caveats. Put the raw counts next to any percentage so a reader can check the arithmetic.

As an illustrative calculation, not a reported study, a team that checks forty URLS and finds six with duplicate canonicals should report both counts alongside the fifteen per cent calculation.

Say what evidence would change your mind, so readers can see which figures are decisive.

If the benchmark feeds a recurring client report, the technical SEO reporting dashboard in England explains how to keep those figures consistent month to month.

Where a benchmark justifies spend, say which figures are illustrative and which are measured. A labelled example, such as a team paying £400 a month for crawl credits, is illustrative, not a measured benchmark result.

Common questions

How many sites does a benchmark need?

There is no fixed threshold. State the number, describe how the sites were selected and treat small samples as indicative rather than representative.

Can I benchmark using only my own crawl data?

Yes, provided you say so. A single-site benchmark measures your own site over time and cannot support claims about the wider market.

How often should a benchmark be repeated?

Repeat it when the underlying templates, tools or definitions change, and record the date of each run so comparisons stay valid.

Do I need to cite legislation in an SEO benchmark?

Only where the benchmark touches personal data or regulated content. When it does, cite the relevant Act directly rather than summarising it.