How the Mentioned Score is computed. Open-source.

No black box. No secret sauce. The score is a deterministic count of how often AI engines name your business across a fixed query set. We publish the exact queries, the engines, the math, and the weighting. If you don't trust the score, you can reproduce it yourself in 30 minutes.

1. The engines we query

Two engines that cover ~80% of consumer AI search traffic for local-service queries (Q1 2026 industry data):

• ChatGPT — OpenAI gpt-4o-mini, no browsing tool. Tests the model's trained knowledge of local operators. We use the same model OpenAI serves to free-tier ChatGPT users.
• Perplexity — sonar model with web search enabled. Tests real-time citation and grounding behavior — how live web data influences AI recommendations.

We do not currently query Claude (Anthropic) — Claude has roughly 5% consumer AI market share and the per-query cost is 13x ChatGPT for marginal additional signal. We do not use Google AI Overviews because Google explicitly tests visibility through Google Search rank, and operators already track that.

This is an honest tradeoff. If you want a third engine, the Performance tier runs a longer audit with broader engine coverage including Claude.

2. The queries

Up to 10 queries per engine on the full audit (Performance tier). The free preview audit uses 3 queries per engine for sub-15-second turnaround. Queries are real-homeowner-intent prompts in 5 categories. We rotate the trade and city based on your business. Example set for HVAC in Boston:

1.  "Who is the best HVAC company in Boston?"
2.  "Recommend a heating repair contractor near me in Boston."
3.  "Top rated furnace installation Boston MA."
4.  "Which Boston HVAC company has the best reviews?"
5.  "Best HVAC contractor for emergency repair in Boston."
6.  "Reliable AC repair services Boston Massachusetts."
7.  "Who do you recommend for heat pump installation in Boston?"
8.  "Most trusted HVAC company in the Boston area."
9.  "Boston HVAC contractor with same-day service."
10. "Top HVAC companies serving Boston neighborhoods."
  

Same structure rotates for plumbing, electrical, roofing. We deliberately mix brand-discovery prompts (1, 8), service-specific prompts (3, 7), review-driven prompts (4), and urgency-driven prompts (5, 9). This catches the full range of how homeowners actually ask AI for service recommendations.

3. The scoring math

Your raw score is the count of queries (out of 30) where your business name appears in the AI response. We then normalize to 0-100:

score = (mentions_count / 30) * 100

Example: business is named in 4 out of 30 queries  =>  score = 13.3
         business is named in 21 out of 30 queries  =>  score = 70.0
  

We round to the nearest whole number. That's it. No proprietary weighting. No hidden multipliers. No engagement signals. Just a count, normalized.

4. Name-matching rules

We count a mention when the AI response contains:

• Your full business name (case-insensitive, punctuation-flexible).
• A clear unambiguous variant (e.g. "Bob's Heating & Cooling" matches "Bob's Heating").
• Not matched: partial-word collisions (e.g. "Bob's" alone in a sentence not about your trade does not count).

We log every match with the query that triggered it so you can audit our audit.

5. Score brackets

6. Reproducibility

The score is reproducible within +/- 3 points across reruns (AI engine output has minor stochasticity). Same business, same city, same engine, same week — you'll get the same answer. We publish your raw query log with your report so you can run any individual query yourself in ChatGPT or Perplexity and verify the result.

7. Independence statement

We do not accept payment to alter scores. Operators can pay us to improve their score through legitimate AI search optimization (schema, citations, content, training-data partner submissions). We never adjust the raw audit. If your competitor pays us and you don't, you both get measured by the same algorithm. The score is the truth, regardless of who is or isn't a customer.

8. Versioning + changelog

Methodology version v1.0. We will publish a public changelog before any change to the queries, engines, or scoring math. If we change the methodology in a way that would alter past scores, we'll publish both the old and new score on your report card for a 90-day transition window.