SEE THE DELIVERABLE

Not another score.
A decision system.

This worked example follows one coherent business problem from crawl evidence to a 30-day implementation plan. Every number is illustrative, but every field, calculation and recommendation represents the practical detail a Professional report is designed to provide.

FICTIONAL CLIENT · ILLUSTRATIVE DATA · REAL REPORT STRUCTURE

SAMPLE PROFESSIONAL REPORT · HELIOFORGE RESEARCH30-DAY GEO GROWTH BRIEF

CLIENTHelioForge Research

Fictional European publisher of commercial-solar benchmarks for finance, property and procurement teams.

COMMERCIAL OBJECTIVEOwn the answer for solar-payback decisions

Increase qualified visibility when buyers compare payback periods, ownership models, incentives and PPA pricing.

MEASUREMENT SCOPE24 buyer questions × 3 providers

72 timestamped observations, one 128-URL crawl and a comparison against the sources returned in those answers.

EXECUTIVE DECISIONRepair and repackage 12 high-intent pages before commissioning more general articles.

The site already contains differentiated data. Its immediate problem is access, source attribution and answer packaging—not a lack of subject matter.

CITATION READINESS62/100BASELINE · 8 AUG 2026

URLS DISCOVERED128URLS ANALYSED121INACCESSIBLE7BUYER QUESTIONS24CITED RESPONSES19/72OBSERVED VISIBILITY26%

ACCESS86121/128 URLs retrievable

EVIDENCE4866/112 claims lack a nearby source

ANSWERABILITY5831 pages lack concise answer blocks

STRUCTURE74103 pages have a clear hierarchy

FRESHNESS5538 decision pages pre-date 2025

TRUST6027 pages lack reviewer credentials

WHAT THIS MEANS FOR THE BUSINESSHelioForge is discoverable, but its most commercially valuable evidence is difficult to verify and quote.

Across the 24-question buyer set, the domain earns 19 of 72 possible provider citations. The largest recoverable gap is the “payback by country” cluster: only 2 of 18 responses cite HelioForge even though its benchmark contains the relevant figures.

PRIORITY 01 · 18 OBSERVATIONS

Country-level payback data is trapped in a PDF table

The benchmark page describes the methodology, but the country figures exist only inside a 42-page PDF. Providers repeatedly return IEA PVPS, EU Commission and SolarPower Europe pages with equivalent figures exposed in HTML.

Decision: publish one indexable country comparison hub

PRIORITY 02 · 17 PAGES

Primary evidence is separated from 66 material claims

Claims name a percentage or payback period, but the publisher, study year and source link sit in a distant bibliography. A reader—or retrieval system—cannot reliably connect each claim to its provenance.

Decision: repair existing evidence before adding volume

PRIORITY 03 · 7 URLS

Five already-linked resources are blocked from retrieval

Seven research assets return 403. Five have external backlinks and two support high-intent query clusters, making access restoration more valuable than producing replacement articles.

Decision: send an exact URL remediation list to Web Ops

Recommendation order: unblock evidence → expose the country dataset in HTML → attach sources to claims → refresh dated incentive pages → rerun the same 72 observations. Do not commission a broad “solar trends” content programme until these assets are repaired.

PAGE-LEVEL PRESCRIPTIONSThe report names the exact asset, problem and definition of done.

The complete export contains all 121 analysed URLs. These six rows show the highest-priority work rather than a generic list of SEO advice.

PAGE / BUSINESS JOBSCOREDIRECT FINDINGPRESCRIBED CHANGESUCCESS CHECK

/research/commercial-solar-payback-europeCountry-payback authority page72

24 numeric claims; 18 lack an adjacent primary source. Country table exists only in the PDF.

Create an HTML table; show currency, unit, sample size, period and source on each row.Table extractable; 24/24 claims linked; canonical self-references.

/guides/lease-vs-buy-commercial-solarCommercial comparison page79

Already cited 7 times, but calculator assumptions are not visible in the article.

Add the financing assumptions and a five-factor lease-versus-buy matrix.All calculator inputs visible; answer block under 80 words.

/industries/warehouse-solar-roiSector consideration page58

312 words of generic copy; no roof-area, tariff or sensitivity evidence.

Add a worked 10,000 m² warehouse example and three sensitivity bands.Example includes formula, units, range and source date.

/research/incentives-germany-ukProcurement evidence page46

Visible update date is 2023; three linked programmes have changed.

Revalidate each programme against its government source and add an eligibility table.All programme links return 200; visible and schema dates match.

/data/commercial-ppa-prices.csvOriginal pricing dataset—

HTTP 403 to the crawler; no HTML description or licence statement.

Remove the CDN bot block and add a landing page with field definitions and licence.HTTP 200; CSV downloadable; Dataset schema validates.

/authors/dr-mara-veenExpert-entity page42

Author name appears on 27 pages, but no biography, qualifications or reviewed-by linkage exists.

Publish credentials, subject remit, disclosures and links to reviewed research.Author entity resolves from all 27 pages.

Each prescription is tied to the signal that triggered it, the query cluster it supports and a machine-checkable completion condition. Teams can assign the exported rows directly to editorial, research or web operations.

LLM-BY-LLM COMPARISONObserved results stay separate from comparative estimates.

The first three providers were queried directly for this example. The remaining profiles show how the same measured evidence aligns with disclosed retrieval preferences; they are not presented as live citations.

MODEL FAMILYEVIDENCE CLASSRESULTWHAT HELD THE SITE BACKMOST RELEVANT CHANGE

OpenAIOBSERVED7 / 24 cited

Good retrieval of explanatory guides; weak selection of PDF-bound country data.

Expose country evidence in HTML with adjacent sources.

Google GeminiOBSERVED8 / 24 cited

Best current coverage; visible dates and government references influenced winning pages.

Refresh incentives content and preserve strong canonical signals.

Anthropic ClaudeOBSERVED4 / 24 cited

Returned research-led sources with clearer methodology and author credentials.

Connect every benchmark to method, author and primary evidence.

Microsoft CopilotMODELLED63 / 100

Indexable HTML and entity consistency are the largest comparative gaps.

Repair blocked assets and author entities first.

Meta AIMODELLED59 / 100

Short, self-contained answers are sparse on long commercial pages.

Add concise definitions and comparison summaries.

xAI GrokMODELLED58 / 100

Freshness and externally corroborated claims are the weakest matching signals.

Timestamp benchmark updates and cite current public datasets.

DeepSeekMODELLED56 / 100

Tables lack explicit units and explanatory field labels.

Publish semantic HTML tables with definitions.

Alibaba QwenMODELLED55 / 100

Regional comparisons lack consistent country and currency context.

Normalise geography, currency and period on every metric.

Moonshot KimiMODELLED54 / 100

Long reports are available, but executive claims do not link to supporting sections.

Add claim-to-section anchors and a structured contents map.

MistralMODELLED60 / 100

European relevance is strong; transparency and document access remain limiting.

Prioritise EU-source provenance and accessible report formats.

Observed total: OpenAI 7 + Gemini 8 + Claude 4 = 19 citations from 72 responses. Modelled scores are diagnostic comparisons only and are never added to the observed citation rate.

30-DAY MEASUREMENT TARGET19/72 → 29/72 cited responses

Raise observed visibility from 26% to 40% on the same fixed question set. This is a team target, not a guaranteed provider outcome.

READINESS TARGET62 → 75+

Zero blocked priority assets, at least 100 of 112 material claims source-linked, and all 12 priority pages passing the completion checks.

WHEN / OWNERACTIONWHY NOWDEFINITION OF DONE

Days 1–2Web operationsRestore seven blocked research assets

Five already earn backlinks; two support priority buyer questions.

All seven return 200 to declared crawlers; redirects ≤1; canonicals resolve.

Days 2–6Research + editorialTurn the payback PDF table into the primary HTML evidence hub

This single cluster contains 16 of the 53 uncaptured observed opportunities.

Country table includes units, period, sample, sources, method and downloadable data.

Days 5–10Editorial leadAttach primary provenance to 66 unsupported claims

Evidence proximity is the largest measured weakness across 17 pages.

At least 100/112 material claims have publisher, date and direct source link.

Days 8–14Policy analystRebuild the Germany–UK incentive comparison

The page is three years old and captures only 1 of 12 observed citations.

Every programme revalidated; eligibility table published; visible/schema dates align.

Days 12–18Content operationsPublish expert and methodology entity connections

Twenty-seven articles name a reviewer who has no resolvable credentials page.

Author/reviewer entities link bidirectionally to qualifications and reviewed work.

Days 25–30Growth leadRecrawl and rerun the identical 72 observations

A fixed test set distinguishes genuine movement from a changed benchmark.

Compare page checks, citation count, cited URLs and competitor movement to baseline.

DEPRIORITISE THIS MONTH

Generic trend articlesThey do not solve the observed source-selection gap.

Unverifiable “AI visibility” claimsOnly timestamped provider observations count as citations.

New PDFs without HTML evidence pagesThey repeat the format problem already suppressing retrieval.

THE DELIVERABLE, ON CAMERA

You do not leave with a diagnosis. You leave with the artefacts.

The implementation fix kit that ends every full report: a ready-to-adapt llms.txt, the robots.txt AI-crawler policy, JSON-LD drafted from what the page really contains, and answer capsules written to be quoted — copied to the clipboard, or exported as Markdown and JSON.

THE FIX KIT · 0:37REAL PRODUCT · SEEDED DATA · NO AUDIO

Every draft is generated from the measured findings of the report above it. Captions carry the narration; there is no audio track.

Read the transcript

Every full report ends in the same place: the fix kit. A ready-to-adapt llms.txt — the content map AI agents actually read. A robots.txt policy that says exactly which AI crawlers get in.

JSON-LD structured data, drafted from what the page really contains. Answer capsules: short, self-contained passages written to be quoted by an engine. Copy. Paste. Adapt. Ship it.

Or take the whole report as Markdown or JSON. You do not leave with a diagnosis. You leave with the artefacts.

This is the level of specificity customers are paying for.

The free report identifies the strongest gap and first action. Professional adds the complete URL inventory, query evidence, LLM-by-LLM interpretation, assigned backlog, success checks and repeat measurement.

Analyse my website →