Please purchase this course, or sign in if you’re already enrolled, to access the course content.
Lesson overview
This practical lesson gives you the corroboration checks from Lesson 3.5 as a procedure, with a ready-to-use prompt for the cold model test and a log format for the contradictions you find. You finish with your Corroboration score and the backlog it produces.
The checks run in the order the lesson recommends. You search Wikidata and Wikipedia for the brand and record what exists, then work through the profiles a third party is most likely to use — LinkedIn, directories, speaker bios, podcast show notes — comparing each against the canonical name and description from your naming table. Every disagreement goes into a contradictions log with what your site says, what the graph says and what the model says.
The cold model test comes with a prompt you can copy, asking a model what the brand is, who is behind it, what it is known for and what it is unsure about, with explicit permission to say it does not recognise the brand. The lesson explains how to make sure web search is really off, how to read the three possible outcomes, and why a confident paragraph about an obscure brand means the prompt is broken rather than the brand. It finishes with the scoring rubric and a template for a quarterly re-test that uses the same prompt and settings each time, so change can be measured.
Nothing in the practical requires paid tools. Wikidata, Wikipedia and profile checks are manual searches, and the cold test needs only access to a chat model where web search can be switched off, so the whole exercise can be run on any brand in under an hour.
This practical completes Module 3, Knowledge Graph Auditing & Entity Mapping, in The AI-powered Entity Audit, and leads into Module 4 on competitive entity research. It suits learners auditing their own brand or a client’s, and anyone who needs to report what AI systems know about a brand without overclaiming.
In this lesson you’ll cover: