A monthly prompt-bank run across ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews for a national US retail brand, scored against a fixed rubric and fact-checked before it reaches the agency's dashboard.
Agencies fielding client questions about ChatGPT and AI Overviews had no repeatable way to answer them. A one-off check in a chat window doesn't hold up as reporting, and it misses the real question a client is asking, not just whether the brand gets mentioned, but whether the answer sends anyone back to the brand's own site. Without a fixed method, the same question produces a different answer depending on the prompt, the day, and who's asking.
A prompt bank runs monthly across five engines: ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews. Every answer is scored against a fixed rubric, answer influence, recommendation strength, competitive position, sentiment, and entity recognition, so the same question produces a comparable result month over month. Claims inside AI answers are checked against a maintained fact bank, and anything touching price or promotions is flagged for a human to review before it's treated as fact. Verified results write directly into the agency's live scorecard.
A fixed set of prompts covers the brand's category, competitors, and buying-stage questions, built to be re-run identically every month rather than reworded each time.
The same prompt bank runs across ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews, capturing exactly what each engine answers that month.
Each answer is scored on five fixed criteria: answer influence, recommendation strength, competitive position, sentiment, and entity recognition. The rubric doesn't change month to month, so scores are comparable over time.
Claims inside each answer are checked against a maintained fact bank. Price and promotional claims are flagged automatically for a human to review before anything is reported as accurate or inaccurate.
Verified results write into the agency's live scorecard, so the agency sees the same dashboard its client would, branded as its own work.
20 minutes. No slides. One process, one prototype.
Book a 20-min intro