This practical turns Lesson 4.3 into settings you can use. It starts from a real test: with default settings, the topic “entity SEO” returned 33,160 AI mentions led by Instagram and Facebook; with the topic required in the user’s question, it returned 85 mentions led by Neil Patel, Schema App, Search Engine Land and InLinks.
What this lesson covers
- The working DataForSEO configuration
- A script that formats and checks topics
- Cost before calls
- Samples before aggregates
- Calibration with a known query
- A Claude skill for the MCP route
The lesson walks through every setting and why it matters: one topic per call, because several keywords in one target are combined as AND; the question scope, because answers often mention homonyms while questions rarely do; the default word match, which is already the tighter option; sources counted from cited links; and the filters that exist, which narrow by volume, platform or date but cannot disambiguate a topic. The same test on “semantic SEO” replaced semantic-HTML sites with Semrush, SE Ranking, Search Engine Journal and WordLift.
A script formats the topics, refuses lines that contain several topics, warns about single generic words, builds the requests and prices the run before sending anything. It then pulls a sample of real questions and answers for each topic, which you mark as on-topic or not, alongside the aggregated list of cited domains and the domains cited together with your own site. For learners using Claude, a reusable skill encodes the same rules for the DataForSEO MCP server.
Every setting in the lesson was tested on live calls on 31 October 2026, and the results are shown so you can see what changed and why. The script never sends a request without printing the plan and its maximum cost first, and it stores every raw response so a number in a report can always be traced back to the call that produced it.
The lesson ends with what to do when a sample fails the check: tighten the topic wording, add a qualifier that excludes the homonym, or label the result indicative and leave it out of decisions.
This practical follows Lesson 4.3 in Module 4 of The AI-powered Entity Audit. It needs a DataForSEO account; the trial credit covers a small run, and each topic costs roughly twenty cents.