Use AI to inspect evidence—not invent it.

Technical access evidence review

You are reviewing supplied HTTP evidence for one public URL.
Use only the response chain, headers, robots.txt, sitemap entries, and HTML I provide.
Report: observed status and redirects; canonical; robots directives; H1; primary content
availability without JavaScript; structured-data types; and inconsistencies.
For each finding, quote the exact supplied evidence location.
Separate observation from likely impact. Do not infer live behavior you cannot see.
End with missing tests required before a conclusion.

Case-study claim audit

Act as an adversarial research editor. Review the supplied case-study draft and evidence table.
Create four columns: exact claim, supporting evidence, strongest justified wording, and risk.
Flag overlapping date windows, missing denominators, selective cohorts, causal language,
ownership conflicts, private evidence, and conclusions that exceed the design.
Do not improve the marketing copy until the evidence boundary is correct.

Primary-source citation check

For each factual sentence in the supplied draft, identify its cited source.
Classify the source as primary documentation, peer-reviewed research, first-party observation,
secondary reporting, expert opinion, or unsupported. State whether the source directly supports
the exact wording. Recommend a narrower claim when support is partial.
Do not fabricate URLs or replace missing verification with general knowledge.

Entity consistency review

Compare the supplied public pages for one person or organization.
Extract names, current roles, start dates, locations, affiliations, profile URLs, credentials,
contact identity, and service claims. Produce a conflict table with source URL and exact text.
Recommend a canonical fact record. Do not treat structured data as independent proof.

Answer-first rewrite

Identify the primary reader question in this draft and the decision it should enable.
Write a 2–4 sentence direct answer that includes the most important qualification.
Then propose a heading outline that substantiates the answer with supplied evidence, examples,
exceptions, method, and next action. Preserve uncertainty. Do not add facts or citations.

How to use these safely

• Remove confidential or personal data unless the tool and account are approved for it.
• Provide the actual evidence; do not ask a model to reconstruct a live audit from memory.
• Verify status, page content, sources, calculations, and dates independently.
• Save the prompt, inputs, model/mode, date, output, reviewer, and corrections.
• Treat the result as an assistant’s analysis, not an authoritative source.