The fastest fair price on HVAC work comes from vetting three contractors before you have an emergency. That used to take an afternoon. With ChatGPT, Claude, or Perplexity, it takes about 20 minutes and costs nothing.

This post is the exact workflow. Six prompts that do the lookups for you, plus an 11-point scorecard you fill in once the AI has done the legwork. If your equipment is over 10 years old, run the workflow now and keep the scorecard somewhere you can find it. The day your AC fails in August is not the day to start vetting.

What AI is actually good at here, and what it is not

AI is good at:

  • Pulling and summarizing public records (state license, BuildZoom permit history, BBB complaint patterns)
  • Cross-referencing claims on a contractor’s website against verifiable sources
  • Surfacing patterns in 24 months of reviews that you would have to read for hours
  • Searching local Reddit threads for honest takes on specific companies

AI is not reliable for:

  • Real-time prices (AI cost guides drift; cross-check against the HVAC Replacement Cost Guide and the Real Cost of HVAC Replacement on this site)
  • Whether a specific permit is valid right now (verify on the city’s permit portal)
  • Whether the human you are talking to today is the human who will show up tomorrow

So: AI does research, you do scoring. That is the pattern.

The 6 prompts

Run each one for every contractor you are considering. Most homeowners run this against three companies. Use ChatGPT, Claude, or Perplexity. Perplexity tends to surface citations more reliably; ChatGPT and Claude are better at the synthesis step. If you have a research-mode toggle (ChatGPT Deep Research, Claude with web search, Perplexity Pro), use it.

Prompt 1 — State or local license verification

Verify the contractor license status for [COMPANY NAME] in [CITY, STATE].
Return:
1. License number(s)
2. Issuing authority (state board, county, city)
3. Status (active, expired, suspended, complaint history)
4. Any disciplinary actions in the last 5 years
5. The specific URL on the issuing board's website where I can verify this myself

If the state does not require a state-level HVAC license (e.g., Colorado), confirm
that and return the relevant municipal license verification path for [CITY].

Do not infer or guess. Cite the exact source URL for each claim.

Prompt 2 — BuildZoom permit history

Look up [COMPANY NAME] in [CITY, STATE] on buildzoom.com and report:
1. BuildZoom score
2. Total permits pulled in the last 36 months
3. Average permit value
4. Number of municipal license endorsements
5. Any "cancelled at one point" flags on municipal licenses
6. Whether the permit volume matches the company's advertised work volume
   (a company that claims "thousands of installs" but pulls 12 permits per
   year is doing unpermitted work)

Cite the BuildZoom profile URL.

Prompt 3 — BBB complaint pattern read

Pull the BBB profile for [COMPANY NAME] in [CITY, STATE] at bbb.org.
Read the complaints from the last 24 months and report patterns, not just the
letter grade. Specifically look for:
- Water heater installs quoted over $8,000 (typical fair $1,500-$4,500)
- Furnace replacements over $15,000 (typical fair $5,000-$9,000 standard residential)
- "Cracked heat exchanger" or "carbon monoxide" diagnoses paired with same-day
  replacement contracts
- Diagnostic fees that do not credit toward repair
- Damage during install followed by repair charges
- Pressure tactics on elderly customers

Also report: BBB accreditation date, years in business, ownership changes,
and whether the company has been acquired by a private equity firm in the
last 5 years.

Cite the BBB profile URL and the specific complaint URLs you reference.

Prompt 4 — Cross-platform review consistency

Find the average star rating for [COMPANY NAME] in [CITY, STATE] across:
- Google Business Profile
- Yelp
- Angi (formerly Angie's List)
- HomeAdvisor
- Nextdoor

Report the rating on each, the number of reviews, and the spread.
Honest companies sit in a 4.3 to 4.8 range across all of them.
A 4.8 on Google paired with a 2.3 on Yelp is review manipulation.

Then search Reddit specifically for:
- "reddit [CITY] HVAC recommendations"
- "reddit [CITY] HVAC [COMPANY NAME]"

Summarize what local subreddit threads say. Reddit is the most useful unfiltered
source. Report direct quotes where useful and link the thread URLs.

Prompt 5 — Certification verification

For [COMPANY NAME], verify (do not just accept the company's claims):
1. NATE certification (North American Technician Excellence) — check natex.org
2. EPA 608 certification (refrigerant handling)
3. Factory authorized dealer status — check the manufacturer locator pages:
   - Trane Comfort Specialist
   - Carrier Factory Authorized Dealer
   - Lennox Premier Dealer
   - Rheem Pro Partner
   - American Standard Customer Care Dealer
4. Master mechanical license held by an actual employee (not just the LLC)

For each claim on the company's website, return either "verified at [URL]"
or "claim not verifiable from public sources."

Prompt 6 — Business model audit

Evaluate the business model of [COMPANY NAME] for predatory-pricing risk.
Search the company's website, reviews, and any news articles. Report:

Red flags:
- "Fully stocked trucks" or "same-day service guaranteed" language
- Membership / club programs that auto-renew
- Tech compensation explicitly stated as commission-based
- Sales reps separate from technicians (one diagnoses, another closes)
- Heavy TV/radio advertising spend
- Recent private-equity acquisition (search "[COMPANY] acquired" and
  "[COMPANY] private equity")

Green flags:
- Owner-operated or family-owned, multi-generational
- Technicians explicitly paid hourly with no commission
- Free estimates given by licensed technicians, not sales reps
- Permits advertised on every install
- Repair-before-replace mentality cited in reviews
- Smaller fleet, owner answers the phone

Report the count of each and quote the source for the strongest red and green flags.

How to run the prompts efficiently

Three contractors, six prompts each, is 18 prompt runs. That sounds like a lot. It is not. Two practical ways to compress:

Compressed approach (one chat per contractor): Open one Claude or ChatGPT chat per contractor. Paste the company name and city at the top, then paste the six prompts in sequence. The model keeps context across prompts inside the same chat, so you do not have to re-specify the company name. End-to-end this is about 7 minutes per contractor.

Parallel approach (three chats at once): Open three browser tabs, one chat per contractor. Run prompt 1 in all three, then prompt 2 in all three, etc. Comparison is faster because the answers are side by side.

Either way, copy the AI’s responses into a single doc as you go. You will score from the doc, not from the chat history.

The 11-point scorecard

Once the AI has done the research, you score the answers. AI does not score because the model has no skin in the game; you do.

#CriterionPointsWhat earns full points
1State or local license active15Active, valid in the city or county where the work happens. Pass/fail. Zero points equals disqualified.
2Pulls permits consistently15Verified through BuildZoom or city records.
3BBB complaint pattern (last 24 months)15No predatory pricing complaints. Subtract aggressively for repeated patterns.
4Cross-platform review consistency10Within 0.5 stars across Google, Yelp, Angi, HomeAdvisor.
5BuildZoom score1095 or above.
6Volume in healthy range1030 to 300 permits over 3 years.
7NATE / manufacturer certifications5Verified at natex.org or the manufacturer locator.
8Years in business55+ years under continuous ownership.
9Pricing transparency policy5Itemized written quotes. No commission-based techs.
10Owner accessibility5Owner answers calls or does estimates.
11Permit value fits typical job range5Recent permit dollar amounts match a fair install.
Total100

Score interpretation:

  • 85 to 100: Hire with confidence.
  • 70 to 84: Solid. Get a second quote anyway.
  • 55 to 69: Small jobs only. Never authorize a replacement on the first visit.
  • Under 55: Avoid.

Three failure modes to watch for in AI answers

AI is helpful here, but it has predictable failure modes worth knowing about.

Hallucinated license numbers. If a model returns a license number, verify it on the state board’s database directly. Models occasionally produce plausible-looking license numbers for companies that exist but cannot actually be looked up. Always cross-check.

Outdated complaint counts. BBB complaint windows refresh on a rolling basis. A model that says “no complaints in last 24 months” may be reading cached data. Click through to the BBB profile yourself for any contractor you are seriously considering.

Soft language on red flags. Models trained for safety will sometimes hedge on negative findings. If your prompt produces wording like “may have,” “could potentially,” or “users report,” push back: “Be specific. Cite the source URL and quote the exact language. Do not soften.”

Why this workflow saves the most money on the highest-stakes call

The biggest predatory pricing risk is not routine maintenance. It is the moment your equipment is at end of life, you feel urgency, and a tech quotes a $25,000 replacement on the same visit. Vetting three contractors before that moment means you have a pre-scored shortlist when the call has to be made fast.

If you are about to spend $15,000 to replace a furnace because a contractor told you “now is the time,” step back and run the bigger market check. HomeStats publishes the median price, property tax rate, insurance burden, price-to-income, and Buy-vs-Rent math for every U.S. state, county, metro, and ZIP code. If your home sits in a market where the equity math no longer favors holding, you may be replacing equipment for a buyer rather than for yourself, which changes how much you should spend. The Listing Analyzer flags every system that is approaching end of useful life on any home, with cited industry life-cycles, so you can run the same check the buyer’s inspector will run.

This post is informational, not legal, financial, or contractor-licensing advice. Mentions of third parties, including licensing boards, certifying bodies, and review aggregators, are nominative fair use. No affiliation is implied. AI tools occasionally produce outdated or inaccurate information; verify every license number, permit, and certification on the issuing source’s own database before relying on it.