TradeOps Consulting
Names 5 answer engines: ChatGPT, Gemini, Bing Copilot, Perplexity, Google AI Overviews.
Read 12 September 2026 · tradeopsconsulting.com/ai-optimization-for-hvac-companies/
The scores
| Criterion | Score | Weighted of its weight | Contribution |
|---|---|---|---|
| Named engine coverage | 5 / 5 | 25 / 25 | |
| Measurement method shown | 5 / 5 | 20 / 20 | |
| Citation change over time | 3 / 5 | 6 / 10 | |
| Entity and schema competence | 5 / 5 | 30 / 30 | |
| Falsifiability | 4 / 5 | 12 / 15 |
What the page actually says
- Engines
- Five named engines, and the only page in the whole set that correctly writes “AI Overviews (formerly SGE)” rather than still calling it SGE.
- Method
- A fixed HVAC prompt set sized by service and city, re-run monthly across the five named engines, with a scoreboard of citation share, competitor gap and engine-by-engine inclusion. Cadence and prompt families are stated on the page rather than left as “we track AI mentions”.
- Change over time
- Monthly re-runs of the same prompt set, with period-over-period citation share published for named client verticals. Not a one-off audit: the page dates the tracking window and shows movement rather than a single snapshot.
- Entity and schema
- Service, LocalBusiness, FAQ, Review and Offer schema named individually, entity linking to Google Business Profile, Yelp, Facebook and industry directories, and knowledge-graph framing.
- Falsifiability
- States that citations are not bookings and that model answers vary by phrasing, location and session. Publishes an engagement limit: if citation share does not move after three monthly cycles on the fixed prompt set, the work is treated as failed rather than re-sold as progress.
What we would ask them about
2 things on TradeOps Consulting's own page, worth raising before you sign. Not a judgement about the work.
- AskAsk for the raw engine answers behind the scoreboard, not only the summary share figures.
- AskConfirm the prompt set is locked in writing before month one so later reports stay comparable.
The check we would run on this one
TradeOps Consulting is one of the few here that shows a method, so the useful check is a level deeper than usual. Ask whether the prompt set is fixed in writing before the work starts, whether it stays identical between reports, and whether you can read the raw answers rather than a summary score. A company already publishing this much will have those answers to hand, and the one that matters is the second.
Disagree with any of this? The corrections log is where it gets settled, in public, with a date.