A useful finding should tell you what to check next.

A consultancy can explain its service clearly and still leave a prospective client asking how the work is delivered or what supports a claim. A technical access check will not answer those buying questions. We use this benchmark to make the different checks visible.

Three findings to put to work

Start with the evidence behind an important claim

The June records marked methodology present on 10 of 50 pages and limitations present on 6. These are recorded features of the reviewed pages, not a verdict on the businesses behind them.

Your next check: choose one claim a buyer relies on. Can they see the work, example, process or source behind it? A relevant case or a clear explanation may answer that question. A new dataset may add nothing.

Read the categories before the total

The median total was 52.5 out of 100. The median Citable score was 4 out of 30 under our DUCR rubric, which awards points for evidence and source presentation.

Your next check: ask what produced the low category score and whether those recorded features matter to the page’s job. Our rubric gives weight to research-style material; a service page can serve buyers well without containing every feature. See what the score means.

Check access and buyer usefulness separately

All 50 recorded primary service responses were HTTP 200. The stored robots flags marked Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot and PerplexityBot as allowed in every row. The study selected publicly accessible sites, so this result does not tell us how common access problems are elsewhere.

Your next check: investigate a specific access problem when there is evidence of one. Once a page can be read, review what a suitable buyer can understand, verify and do next. These records do not show actual platform visits or citation outcomes.

Choose one commercially useful question

What you are unsure about	A useful starting point
Buyers cannot tell whether the service fits	Review the audience, scope and next step using the service-page guide.
A strong claim has little visible support	Choose relevant proof for that buyer’s concern.
An important page seems absent from search	Separate indexing from weak visibility before prescribing a fix.
You want to know whether the work helped	Define the business question and use the measurement scorecard.

If you want our view on where to start, request the free Outside View. If you already have a decision to discuss, request a Clarity Call. The benchmark does not prescribe a rebuild or a content programme from a score alone.

The study behind the findings

Record	Detail
Study edition	2026.06
Sample	50 reviewed service-business websites; one record per sampled site
Collection	June 3, 2026 UTC / June 4, 2026 IST
Scoring	DUCR: Discoverable, Understandable, Citable, Routable
Public records	Site IDs with direct identifying fields removed from the corrected export
Review	22 approved records; 28 approved with caveats

The methodology explains selection, recorded checks and manual review. The decision boundaries explain what additional evidence a recommendation needs.

The recorded score breakdown

Layer	Median	Average
Discoverable	19/25	19.54/25
Understandable	15/25	14.38/25
Citable	4/30	5.90/30
Routable	13/20	13.54/20
Total	52.5/100	53.36/100

These are descriptive scores under our June rubric. They are not search rankings, citation probabilities or ratings of service quality.

How the scores were distributed

Total score band	Records
0–39	2
40–49	14
50–59	22
60–69	11
70–79	1
80–100	0

Within the Citable category, 32 records scored 0–5, nine scored 6–10, eight scored 11–15 and one scored 16–20. No record scored above 20 out of 30. Those bands describe this rubric, not a platform threshold.

What the robots checks recorded

User-agent token	Recorded as allowed
Googlebot	50/50
Bingbot	50/50
OAI-SearchBot	50/50
GPTBot	49/50
ChatGPT-User	49/50
ClaudeBot	50/50
Claude-SearchBot	50/50
Claude-User	50/50
PerplexityBot	50/50
Perplexity-User	50/50

These are stored robots flags, not verified visits by each platform. The robots fetch returned 200 in 48 rows, 406 in one and 404 in one; both non-200 rows still have all allowed flags set to true. The package does not explain how the collector handled those responses. Search, training and user-requested agents have different roles and policies. A business can decline training independently of search access. Read the access study for the other recorded signals.

Features recorded on the reviewed pages

Recorded feature	Pages
Methodology	10/50
Limitations	6/50
Dataset or download	5/50
At least one table	3/50
Original data	0/50
Author name	Not recorded; all 50 fields blank
Visible modified date	Not recorded; all 50 fields blank

The original-data field is explicitly false in all 50 rows. The author and date fields are blank, which means unrecorded, not absent. It does not establish that a business lacks expertise, evidence or useful information elsewhere. None of these features is a universal requirement for a service page.

Who was in the sample

Recorded category	Sites
Consulting	16
Accounting	10
Agency	9
Home services	5
Legal	3
Dental	2
Pest control	1
Junk removal	1
Roofing	1
Wellness	1
Med spa	1

Consulting, accounting and agency sites make up 35 of the 50 records. All original country fields were coded US; no UK comparison was recorded. The sample is not representative of a country or market, and these counts do not support comparisons of industry performance.

Four lenses for a page review

We organize recorded observations into access, clarity, evidence and useful routes. The DUCR score page explains the points and shows one retained record. Those lenses help structure a review; your audience and commercial goal determine the priority.

Add outcome evidence before drawing an outcome conclusion

The June study did not measure actual AI citations, private analytics, received enquiries or sales. To investigate those questions, use a recorded answer sample and your own business records. Keep the page observation and the business result separate until the evidence connects them.

Inspect the records

Download the reviewed CSV, reviewed JSON, codebook or original scoring rubric. The dataset page explains the export revision, summary files and checksums.

What supports each kind of statement

Statement	Evidence to inspect
A count or average from the sample	The 50 retained records and the named field
A point allocation	Our June DUCR rubric
A platform requirement	Current primary documentation linked from the methodology
A recommendation for your site	The observation plus its relevance to your buyers and goal

Take the next useful step

Read how to decide whether a finding merits work, inspect our AI SEO plan, or bring the decision to a Clarity Call.

Study and editorial history

Date	Record
June 4, 2026 IST	Edition 2026.06, the reviewed 50-site dataset and rubric
October 5, 2026	Interpretation and public-export correction; original observation dates and scores retained