AI Visibility Benchmark · DUCR Score
Read the score before you spend on the fix.
DUCR organizes public-page observations into four layers. The useful question is what the breakdown reveals about a page your buyers rely on.
Benchmark record
- Model
- 100-point DUCR rubric
- Layers
- D 25 · U 25 · C 30 · R 20
- Edition
- 2026.06
Benchmark corpus
A total can hide very different problems. Two pages with the same score may need different work, or no immediate change. Start with the category, inspect the recorded observation and ask what it means for the person choosing your service.
What each category organizes
| Category and maximum | What our historical rubric groups |
|---|---|
| Discoverable, 25 | Public access signals, robots rules, index/snippet directives, sitemaps, links, canonicals and visible text |
| Understandable, 25 | Consistent naming, topic and page-role clarity, headings, schema and audience context |
| Citable, 30 | Recorded evidence, methods, supporting references, self-contained explanations, source presentation and attribution |
| Routable, 20 | Relevant next steps, internal connections and contact or measurement signals |
The names describe our review framework. “Citable” does not mean a platform has cited the page or will choose it. The points are our editorial weights, not published Google or assistant ranking factors. The archived routing criterion even names a benchmark and the Search + AI Visibility service page. That is a site-specific assumption in the June rubric, not a requirement for every service business.
Follow one retained record
Record SB-0061, a consulting site, was retained as approved with caveat. The June export stores the following category scores:
| Category | Stored score |
|---|---|
| Discoverable | 19/25 |
| Understandable | 7/25 |
| Citable | 3/30 |
| Routable | 10/20 |
| Total | 39/100 |
The same record stores a primary service HTTP response of 200, an H2 count of zero and an answer-block count of zero. You can find this site ID in the reviewed CSV or JSON.
What this supports: a successful response to the study’s collector coexisted with low recorded clarity and source scores. The public file does not show the individual point awards, so those three fields cannot be used to reconstruct why each category received its exact score.
The next question: can a buyer find the service scope, supporting evidence and next step on the current page? If that information is hard to find, clarifying it may be useful. Recheck the live page before assigning work; these historical fields do not establish a current problem or a business outcome.
The June sample in context
| Category | Median across 50 records |
|---|---|
| Discoverable | 19/25 |
| Understandable | 15/25 |
| Citable | 4/30 |
| Routable | 13/20 |
| Total | 52.5/100 |
The average total was 53.36. The lowest recorded total was 37 and the highest was 73. None of those figures establishes a pass mark for your business. The overview shows the distributions and sample composition.
What a low Citable score means
It means few of the features rewarded by this category were recorded. The June rubric gives up to eight points for original data or method, five for claim support, five for self-contained answers, five for tables/charts/downloads, four for author/date/version/limitations and three for restrained claims.
That weighting favors source and research material. A service page’s job may be to explain a specific offer, establish fit and help a buyer contact the right team. It does not need to become a research publication to do that job well.
Thirty-two of the 50 records scored 0–5 in this category. This tells us about the recorded features and chosen weights. It does not show that those pages failed to attract clients or receive citations.
Choose a useful change, not a higher number
| Recorded gap | Check before commissioning work |
|---|---|
| An important service page appears restricted | Is the restriction intentional, and does a current access or indexing check confirm a problem? |
| Audience or service meaning is unclear | Can a suitable buyer identify the work and decide whether it fits? |
| A valuable claim has little support | Would a relevant case, example or process explanation resolve that buyer’s concern? |
| The next step is unclear | Is there an appropriate destination, and can the visitor complete the action? |
| The page has no table or download | Would either help a real decision, or simply add material? |
Use the service-page guide for structure and the proof guide for evidence selection. Define what you expect to improve before the change, then use the measurement scorecard.
How the historical rubric treats blockers
The June rubric includes caps for specified restrictions. A core benchmark page marked noindex has a total cap of 40; nosnippet or max-snippet:0 has a cap of 55. A Googlebot or Bingbot block caps Discoverable at eight and the total at 60. Its specified AI-search block rule caps the total at 70.
It also includes credibility and visible-content checks for unsupported statistics, false claims, dataset markup without a download, hidden findings and schema that does not match the page. These are rules in our archived rubric, not platform penalties. All 50 final records have an empty critical-blocker field; that is not certification that every technical or business issue was absent.
Training permissions are a separate choice. User-triggered fetches also need their own interpretation. Read the method’s bot distinctions before applying an old score to a current access question.
Inspect the rubric or discuss the decision
The June rubric JSON retains every original point allocation. The dataset retains category scores and observable fields. The public records support checking totals and distributions; they do not include a complete per-criterion point ledger for independently recreating every score.
Read how the study was assembled and when a finding is enough to act on. If you want to discuss a priority on your own site, request a Clarity Call. You do not need to calculate a score first.
The Outside View
See where you stand with Google, AI and your buyers.
Send us your website. In two business days you have it in writing: how you look from the outside, who really competes with you, and the ten fixes that matter. Free.