What an audit reveals.
A worked example of how the rubric scores a real UK employer brand. The dimensions reveal what lies beyond the numbers.
The methodology described at /methodology produces a composite score from five dimensions. This page walks through one real audit run end-to-end.
The walk-through covers what the dimensions returned, what they reveal about the subject’s AI visibility, and where the lever sits. The numbers are real; the subject is anonymous. Audit run identifier and rubric version sit alongside so the result is reproducible from source.
Mid-market UK B2B software vendor.
Run ea0329fb-b7a7-439d-ba2d-49622f8f0fa1 · rubric 2026.05.7 · methodology hash sha256:974405f7d8f74a66…
Audited under rubric 2026.05.7, Q2 2026. Mid-market employer; UK-based; B2B software.
Audit ran the full five-dimension rubric across the four active LLM providers: Anthropic Claude, OpenAI GPT, Perplexity Sonar, Google Gemini. It also ran the structural signals: JSON-LD parsing, robots.txt against the canonical AI-crawler list, Wikidata resolution, web-search-mediated Glassdoor lookup. Wall-clock for the run: under four minutes.
Run identifier and methodology hash are pinned to the result; both are recoverable from the audit trail. The score here is pinned to rubric 2026.05.7 forever. Re-running under a future rubric produces a different number, by design.
The composite sits in the lower-middle band. Forty out of one hundred is not a failure score. The rubric returns floors near zero for genuinely invisible companies.
A mid-market vendor scoring forty has a discernible profile worth interpreting. The score is honest: this is what the methodology returns for a company with the strengths and gaps described next.
The five dimension scores produced this composite at the rubric’s fixed weights. Weights are citation 35%, schema 20%, knowledge graph 15%, crawlability 15%, social 15%.
The rubric excluded no dimension and rounded no signal up. The composite is a weighted average of all five at their actual returned values.
What each dimension returned.
Five dimensions scored 0–100 each, citation first per its 35% weight. Numbers are the empirical scores the audit returned. Paragraphs are the diagnostic interpretation those numbers support.
Citation likelihood
Of twenty candidate-research queries probed across the four providers, the company is named in five. Three of those five citations come from the trivial-form query — "what’s it like to work at {company}?" Here the company name is in the prompt and the LLM has been handed the answer.
The remaining two citations are organic. One provider names the company on questions about “top employers in its sector”. The same provider names it on “most innovative companies to work for in 2026”.
Of the four providers, Perplexity Sonar timed out on four of its five probes. Each timeout hit the per-call 20-second limit before returning. The fifth Perplexity probe responded in five seconds without citing the company.
The within-provider rule is honest about this. Timed-out probes contribute zero to the provider’s numerator; the responded-but-not-cited probe also contributes zero; Perplexity’s per-provider score lands at zero. The provider stays in the cross-provider mean.
The "exclude entirely" rule applies only when a provider returns no successful probes at all. Perplexity got over the line by one.
Per-provider scores: anthropic 20, openai 60, gemini 20, perplexity 0. The cross-provider mean of those four values, scaled to the 0–100 dimension scale, lands at 25.
The pattern is legible. One provider (openai) returns organic citations; three providers return trivial-form citations only; one provider’s measurement is substantively a non-result.
This is the rubric’s most direct measure of AI visibility. On the empirical evidence, the company barely surfaces when candidates ask organic questions.
Schema completeness
Five JSON-LD types parse cleanly off the careers site: Organization, WebSite, BreadcrumbList, ImageObject, and WebPage. The foundational entity layer is in place — schema.org-validated Organization markup, sitewide WebSite, BreadcrumbList for the careers section. This is what allows AI crawlers and search-side infrastructure to identify the company as a structured entity at all.
Three high-value types are absent. JobPosting is the most consequential omission for a careers-site audit. The page lists roles in its rendered output but does not expose them as schema.org JobPosting markup.
JobPosting is the canonical structured form Google for Jobs and AI crawlers expect. Person and ProfilePage are absent for the leadership team. FAQPage is absent.
The score reflects the gap. Foundational schema present, expected careers-page schema not.
Knowledge-graph presence
The Wikidata resolution step found five candidate entities for the company name. None matched the entity-keyword filter. The filter confirms a candidate is the actual company rather than a homonymous game franchise, plant, or unrelated firm.
The audit returns unresolved.
This is the audit’s clearest single finding. The company has no Wikipedia entry, no resolvable Wikidata Q-number, and no Knowledge-Panel-grade entity in the structured-knowledge graph. Knowledge-graph presence is a baseline AI awareness signal that operates independently of the careers site.
The score is honest at zero. The lever is not on the careers site — it is at the encyclopaedic-presence layer.
LLM crawlability
Robots.txt allows thirteen of the fourteen AI crawlers in the canonical list. GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and the other major retrieval agents pass through. One blocked.
The site’s content is reachable to the crawlers that matter for AI visibility.
llms.txt is absent. The standard is emergent and not yet a critical signal. The rubric records its absence rather than scoring around it.
The site renders in partial server-side mode. Roughly 58% of the final rendered content sits in the raw HTML response, with the remainder hydrating client-side. AI crawlers without JavaScript execution see the SSR portion.
Render-readiness is a real signal that the site does not optimise for AI crawlers specifically. Enough renders server-side that the major crawlers receive substantive content.
Crawl is the dimension where the rubric’s quietness is the right signal: the plumbing works.
Social-signal density
The social dimension probes Glassdoor presence via web-search-mediated retrieval. The audit matched the company on Glassdoor with thirty-two reviews and a 4.3 average rating.
The base score of 86 (rating 4.3 × 20) capped at 70 because the review count sits below fifty. Above fifty reviews, the rubric treats the rating as uncapped.
The credibility cap is honest pessimism on small-N samples. A 4.3 rating from twelve reviews carries less signal than a 4.3 rating from twelve hundred. The rubric reflects that without needing to throw out the data.
At fifty-plus reviews the cap lifts and the rating reads through directly.
Social v1 ships Glassdoor only. Indeed, LinkedIn, Reddit, and the broader employer-review surfaces are queued for v2. The dimension currently reflects the Glassdoor sub-signal alone, weighted against the rubric’s 15% accordingly.
What this audit reveals.
Across the five dimensions, the audit reveals a clean distinction the rubric exposes by design. The company exists at the site layer.
AI crawlers can reach the careers page, the foundational schema parses, social proof is real and directionally positive. The plumbing works.
The company does not exist at the structured-knowledge layer. Wikidata disambiguation fails; there is no Wikipedia entry; the company name resolves to five other entities before this one. This is the awareness gap, and it cascades.
The cascade is visible in the citation dimension. Three of the five cited probes come from the trivial-form query, where the prompt names the company directly. The other three providers do not surface the company organically.
Two organic queries: “best employers hiring software engineers in the UK” and “most innovative companies to work for in 2026”. Only one provider names the company once across these queries. This is what citation looks like for a company AI does not recognise as a structured entity in its corpus.
The audit does not prescribe a fix. What it reveals is the shape of the gap: site-level signal acceptable, awareness layer empty, citation tracking the awareness layer.
Investing in careers-page schema or content alone would move the schema score but probably not the composite. The lever sits at the encyclopaedic-presence layer, outside the careers site.
This is the consulting interpretation the methodology produces. The numbers are diagnostic; the read of where the lever sits is the work the rubric enables.
The methodology in full.
This audit was scored under rubric 2026.05.7. The rubric defines the five dimensions, their weights, the query templates, the canonical AI-crawler list, and the brand-platform scoring rules.
Content-hashing against the source files makes the score reproducible. Anyone with the rubric definition, the query JSON, and the reference data can re-derive the score.
Read the full methodology →Audit your employer brand against this methodology.
Discovery calls run thirty minutes. We’ll walk through how the methodology applies to your sector.
We’ll cover what your current scores would look like and where the leverage is. No pitch deck, no slides — the methodology and the audit are the pitch.