This practical runs the analysis from Lesson 2.4 in the Claude Entity Analyst, a free browser tool built for the course. It replaces the earlier notebook: there is no code and nothing to install, and every answer Claude gives is checked against the page text before you use it.
What this lesson covers
- The key, the model and the four analyses
- One page, then a crawl export
- The price before anything is sent
- The page-text and NLP checks
- Confidence tiers
- Exports and limits
The tool asks Claude about one page at a time and requires its answer in a fixed structure: entities marked as stated or implied, with their role and a short quote as evidence, plus relationships, attributes and naming issues depending on the analysis chosen. Four analyses are available — an entity map, attribute gaps for the intended entity, a subject check against the intended entity, and a naming and ambiguity review — and each feeds a later part of the course, from Clarity verdicts to the gap matrix and the canonical naming table.
Before anything is shown as a finding, the tool looks up every quote in the page text. An entity whose quote is not on the page is marked Discard, the clearest sign of an invented answer. With the Entity Extractor’s export loaded, it also checks whether Google’s Natural Language API found the same entity. The result is four confidence tiers — Solid, Likely, Candidate and Discard — and CSV exports you can keep with the audit, along with the model and date. The lesson works through a real run on the demo site, in which all ten entities Claude named were verified on the page, at a cost of a few cents.
For more than one page, the tool reads a crawl export, shows the estimated cost before it runs, works through pages two at a time and keeps whatever has finished if you stop it. Its limits are stated in the tool itself: answers vary between runs and models, missing attributes are suggestions to check, and the content you analyse is sent to Anthropic.
This practical follows Lesson 2.4 in Module 2 of The AI-powered Entity Audit and leads into Lesson 2.5 on controlling cost and runtime. It suits anyone who wants an LLM’s judgement on their pages without writing code, and wants every line of it checkable.