A score you
can look inside.

Two different questions. Two clear scores.

AI business exposure. When GPT analysis is available and selected, five factors estimate how the published offer may be exposed to AI-enabled alternatives. Higher means more indicated exposure. This assessment includes a company-specific verdict, evidence, uncertainties and a proposed automation blueprint.

Website foundations. Seventeen reproducible checks produce a score out of 100. The score uses the homepage and selected crawl signals, with the same published weights for every domain.

A broader evidence review. We inspect up to five public pages to build a website profile, identify patterns across the checked pages, review selected crawler policies and suggest a 30-, 60- and 90-day action plan. These findings add context without changing the foundation score.

Clear attribution. Observed page evidence and rule-based recommendations are labelled accordingly. An optional AI business assessment is a separate mode and is identified only when it has actually run. The website review and action brief work without it.

The five-factor AI exposure rubric

The current rubric is bougal-exposure-1.0. A GPT model assesses the following five factors using only the page evidence collected in the scan. It returns an integer from 0 to 100 for each supported judgment, or “Unknown” when the public information is inadequate. Higher values consistently mean greater potential exposure.

The calculation. When all five factors are assessed, the server adds their scores, divides by five and rounds to the nearest integer. The model does not choose the overall score. BOUGAL’s equal weights and bands are editorial choices; they are not calibrated against business outcomes.

Missing evidence stays unknown. If any factor is unassessed, there is no point score. Instead, we show the lowest and highest possible composite values by assigning each unknown factor 0 and 100 respectively, while retaining the observed factor estimates. Both endpoints are rounded to whole points. If all factors are unknown, we show insufficient evidence. These bounds are not a statistical confidence interval.

For example, four assessed factors of 40, 60, 50 and 70 with one unknown produce a range of 44–64, rather than treating the missing factor as zero exposure. The report shows how many factors were assessed and what information would resolve the gaps.

How the model is anchored

Each factor uses explicit low, medium and high reference descriptions. Routine workflows range from predominantly physical or bespoke human outcomes to repeatable digital tasks. Model dependency ranges from explicit value independent of foundation models to an explicitly described thin model interface. Moat vulnerability considers documented barriers and reproducible elements. Agent replaceability distinguishes a bounded digital task from the full outcome a company delivers. Pricing pressure considers published prices, included scope and delivery value together.

Undisclosed AI architecture does not establish model dependency. Missing moat evidence does not prove there is no moat. Unpublished pricing does not justify a pricing score. A high price alone does not establish vulnerability. Every factor includes its evidence, interpretation, confidence and uncertainty, with references to checked pages.

Relationship to the reference checker

The five concepts correspond to the publicly visible business-replacement dimensions used by Death by Clawd. BOUGAL implements its own evidence-based rubric, professional report and planning workflow. We do not have access to that service’s private prompt, model settings, overall-score calculation or calibration, and do not claim identical numerical results or affiliation.

The index is not a probability of company failure, a forecast of financial performance, an investment rating or an actual replacement test. Model judgments may vary, and public marketing information cannot establish every internal capability or competitive advantage.

What we inspect

We request the domain’s HTTPS homepage, its robots.txt and one sitemap location. From links actually found on the homepage, we may select up to four further pages on the same website. Product, service, company, pricing, proof and contact pages are useful candidates. This is a bounded sample of the website, not a complete crawl.

We read the HTML returned by the server and selected response headers. The report records each attempted page, its outcome and the evidence retained from it. A website profile reflects published page titles, descriptions, headings and excerpts; those statements have not been independently verified.

The scanner respects its crawl permissions and does not sign in, execute page scripts, follow external links or submit forms. If the homepage is inaccessible, blocked or provides too little readable content, no score is generated. A failed supplementary page is recorded as incomplete coverage while the available homepage result remains usable.

Recent website evidence and successful assessments may be reused for up to 15 minutes. The original scan time remains visible, and a cached complete result is labelled. If an AI request fails, the evidence can be reused for a fresh assessment without repeating the website crawl; a temporary AI failure is not cached as a completed assessment.

How the points work

The five dimensions add up to 100 points. Each check receives its full weight when its rule is met. Checks that are absent or cannot be verified receive zero points. “Unverified” is shown separately so missing evidence can be distinguished from a confirmed finding.

The thresholds and weights are BOUGAL’s editorial choices. They are not a search-engine standard or a validated predictor of rankings, AI citations or commercial performance. The additional page findings, crawler matrix and any strategic assessment do not add hidden points.

How to read the score

A high score means these selected signals were found. It does not verify whether the business is reputable, the writing is compelling, the site is accessible or the offer is commercially strong. The foundation priorities are the three highest-weight checks that need attention or verification.

Read the coverage alongside the score. Five checked pages provide more context than one, but neither is a representative sample of every business process. A page that was not reached is an information gap, not proof that a capability or policy does not exist.

Search access and training access

The crawler matrix evaluates the retrieved robots.txt rules for selected search and model-training crawlers across the checked paths. “Allowed” means those rules permit the paths we evaluated. “Restricted” means they disallow them; “Mixed” means the outcome differs between paths. “Unverified” means the rules could not be reliably assessed.

Permission does not prove that a crawler visited, indexed or recommended the business. Firewalls, provider policies and other technical conditions can affect actual access. Search discovery and model training also serve different purposes: a business may reasonably permit one and restrict the other. Training restrictions do not reduce the score.

OpenAI documents independent controls for its search and training crawlers. Google’s guidance also explains that its AI search features use established search foundations, without an additional AI-specific markup requirement. The references below describe those policies.

How the action brief is prepared

The standard brief uses the observed checks and page coverage to suggest priorities for the first 30, 60 and 90 days. Each recommendation explains its rationale, links to relevant source pages and proposes a success measure. It is a rule-based planning aid, with recommendations connected to incubation, IT & AI, or brand scaling.

When a genuine AI business assessment is available, its opportunities and roadmap are identified as AI interpretation. It may propose a workflow pilot, a new offer or a visibility initiative. These are hypotheses to test with the business team, not claims that the website proves the initiative will succeed.

The horizons and effort labels are planning suggestions. Actual priorities, investment, staffing and delivery dates depend on information a public website cannot establish.

The automation blueprint

The GPT assessment proposes one bounded workflow related to the actual offer. It separates tasks AI could assist from responsibilities that remain with people, lists required inputs and access, and defines a pilot, a success measure and assumptions to validate. The downloadable Markdown document contains the same proposal shown in the report.

A generated blueprint is a planning specification. It has not built software, connected systems, tested a replacement or demonstrated that an entire company can be reproduced. Human judgment, physical delivery, permissions, relationships and specialist obligations must be considered before implementation.

About the optional AI assessment

An enabled AI service and sufficient readable evidence are required for this separate mode. When it runs, a bounded selection of public website evidence and the observed findings is sent to OpenAI for processing. Inquiry messages, visitor account details and private company records are not part of the assessment.

The model is asked to distinguish the apparent offer, audience and business model from its strategic proposals. Business-specific findings must refer to pages that the scan actually checked. Its structured response is validated for the five required factors, permitted source references, score ranges and confidence rules before display. A completed assessment shows its provider, model, generation time, rubric revision, stated confidence and questions that require further investigation.

Confidence is qualitative. It is not a calibrated probability. Source references help you inspect an assertion, but they do not guarantee the model interpreted the page correctly. The model has no browsing or action tools in this assessment and cannot independently verify the business’s claims.

If this mode is inactive or cannot produce a valid response, the website review remains available. The report explicitly marks the strategic assessment as unavailable; template recommendations are never presented as an AI assessment.

What remains outside the scan

We do not measure actual AI recommendations, search rankings, visitor behaviour, live page performance, reputation, internal AI adoption or competitor performance. The tool does not predict business failure, future revenue or investment value.

JavaScript-heavy websites may expose less information in their initial HTML. Navigation, unusual labels and language differences can affect extraction. Selected English, German, French, Italian and Spanish link labels are recognised; other wording may be missed. Structured-data checks cover JSON-LD, rather than every markup format.

A useful next step is to confirm the findings with the people who run the business, test the real customer journey and gather the operating data needed to validate a pilot.