Quick answer
ChatGPT recommends contractors based primarily on trade directory citations, review signal strength, and editorial presence in sources AI models index during training. In a Homerankr study of 100 ChatGPT contractor queries across 6 trades and 20 US markets, 94% of recommended contractors shared at least 3 of 5 identifiable citation signals. Only 6% of contractors ranking in Google Maps' local 3-pack appeared in the corresponding ChatGPT recommendation for the same query.
100
ChatGPT queries analyzed
6 trades · 20 US markets
94%
accuracy of 5-signal model
predicts ChatGPT recommendation
6%
Google Maps 3-pack overlap
with ChatGPT recommendations
90
days to first ChatGPT citation
Aquacade Pools case study
THE ORIGIN STORY
Aquacade Pools had been building inground pools across Nassau County for 15 years, with a five star average across more than 40 genuine customer reviews and a business built almost entirely on word of mouth and repeat referral work. The owner had never needed to think seriously about digital marketing, because the phone rang enough on its own.
That started to change in the spring of 2026. A homeowner in Old Westbury, mid-way through planning a full backyard renovation, mentioned during her first call that Aquacade was not the first name she had come across. She had already asked ChatGPT which pool builder to use in Nassau County before picking up the phone at all, and a competitor she had never worked with before was the name ChatGPT gave her. He called Homerankr the same week.
She had used ChatGPT to build her contractor shortlist before making a single call. A competitor she had never heard of appeared first. Aquacade was not mentioned at all.
When we audited Aquacade's AI search presence, we found a business that looked strong by every traditional local SEO measure and nearly invisible by every signal an AI model actually seems to weight. No listing in the APSP trade directory network. No Houzz contractor profile despite 15 years of project photos sitting in old phone backups. No editorial coverage in any Long Island publication, despite work that was easily newsworthy. No structured FAQ content anywhere on the website.
We had a hunch about why this was happening, based on a handful of other client audits earlier that year, but we did not have proof. We had never seen a systematic answer to the question of what actually determines whether ChatGPT recommends a contractor. So we designed a study.
THE METHODOLOGY
We ran 100 ChatGPT queries across a standardized grid:
6
trades studied
HVAC · Plumbing · Roofing · Remodeling · Pool Building · Electrical
20
US markets
Austin · Phoenix · Denver · Atlanta · Charlotte · Nashville · Tampa · Orlando · Dallas · Houston · Chicago · Minneapolis · Seattle · Portland · Sacramento · Las Vegas · Salt Lake City · Kansas City · Columbus · Raleigh
3
query formats per market
Format 1: “Who is the best [trade] near me in [city]?”
Format 2: “Find me a licensed [trade] in [city] with good reviews”
Format 3: “What [trade] companies are highly recommended in [city]?”
That produced more than 360 total queries. We used the first named contractor recommendation from each query response, deduplicated by business name to reach 100 unique recommended contractors.
What we documented for each recommended contractor:
Google Maps local 3-pack position
Total Google review count
Reviews in past 90 days (velocity proxy)
Trade directory listing (NRCA/PHCC/NECA/APSP/NARI)
Houzz verified contractor profile
Angi verified professional status
BBB accreditation
Local editorial mention in past 12 months
Website domain authority (Voodoo metric)
FAQPage schema on primary service page
Time period: Study conducted across 3 weeks in July 2026.
FINDING 01
This was the most important finding in the study and the one that surprised us most despite our suspicion going in.
6%
of ChatGPT-recommended contractors were also ranking in the Google Maps 3-pack for an equivalent query
94% were outside the local 3-pack. 71% were outside the top 10 on Google Maps entirely.
Of the 100 contractors recommended by ChatGPT, only 6 were also ranking in the Google Maps local 3-pack for the same trade and city combination at the time we ran the query. We checked map pack position for every recommended contractor within 48 hours of the corresponding ChatGPT query to minimize any drift between the two data points.
Going into the study, we expected some overlap, since it seemed reasonable that a contractor doing everything right for Google Maps would also show up favorably to an AI model. That was not what we found. 71 of the 100 recommended contractors did not appear in the top 10 organic or map pack positions on Google Maps at all for the equivalent search, meaning the majority of ChatGPT's contractor recommendations were coming from businesses that a homeowner using Google directly would likely never see.
Google Maps ranking and ChatGPT recommendation are separate systems governed by almost entirely different signals. A contractor optimized for one is not automatically visible in the other.
FINDING 02
After documenting all 10 variables for each of the 100 recommended contractors, 5 signals stood out as far more common than the rest, and far more predictive of appearance than Google Maps ranking, star rating, or any single metric on its own.
Prediction rule
A contractor with 3 or more of the 5 signals above appeared in ChatGPT recommendations in 94 of 100 cases.
A contractor with 2 or fewer signals appeared in 6 of 100 cases.
FINDING 03
Trade category had a larger effect on the difficulty of appearing in a ChatGPT recommendation than any other variable we measured apart from the 5 core citation signals themselves. Some trades required a much higher review count to compete, while others rewarded a complete citation profile even at a comparatively modest review count.
Pool building and electrical work stood out as the two trades where recommended contractors had the lowest median review counts, 94 and 89 respectively, well below the 127 median across all 6 trades combined. We believe this reflects lower overall competition density in these two categories in most of the 20 markets we tested, meaning fewer contractors are competing for the same recommendation slot, and a complete citation profile carries more relative weight.
HVAC and roofing sat at the opposite end, with median review counts of 156 and 162 respectively, reflecting the volume of established, heavily reviewed competitors already active in most markets for those two trades. Remodeling and plumbing fell in between.
In every trade, however, the 5-signal model held: contractors with 3 or more signals appeared regardless of which trade they were in.
FINDING 04
Beyond raw review count, we noticed a pattern in the actual text of the reviews belonging to recommended contractors versus contractors in the same market and trade who did not appear. Certain phrases showed up disproportionately often among the recommended group, suggesting AI models may be parsing review content rather than only tallying star ratings and counts.
For HVAC and plumbing contractors, reviews mentioning some variation of emergency response, same-day service, or after-hours availability correlated strongly with ChatGPT recommendation. For remodeling and roofing contractors, reviews describing detailed estimates or a clear, itemized quoting process showed the same pattern. For pool builders specifically, reviews that described pool tile work, finish quality, or a specific material or technique correlated with recommendation more than generic praise like great job or highly recommend.
We cannot confirm the exact mechanism, since we do not have access to how ChatGPT's underlying training or retrieval process weights review text. But the pattern was consistent enough across 6 trades that we believe review content, not just review count, is doing real work in whatever process produces these recommendations.
ChatGPT is reading your reviews. Not just counting them.
FINDING 05
Before running the study, we expected the largest markets, Houston, Dallas, Chicago, to be harder to break into than smaller ones like Salt Lake City or Kansas City, simply because more contractors are competing for attention in a bigger metro. That was not what the data showed.
The percentage of contractors with 3 or more of the 5 citation signals who appeared in ChatGPT recommendations was within a few percentage points of each other across every market size we tested, large or small. A pool builder in a market of 200,000 people needed roughly the same citation profile to appear as a pool builder in a market of 2 million.
The citation requirements to appear in ChatGPT recommendations were almost identical across market sizes. A pool builder in Salt Lake City faces the same bar as one in Dallas.
The practical implication is that a contractor in a smaller market gets no discount on the work required, but also faces no real penalty for operating outside a major metro.
The Case Study
Aquacade Pools started this study. They ended it as a case study.
Week 1
APSP listing submitted. Review automation with photo request installed.
Weeks 2 to 3
Houzz profile with 20 project photos. Nassau and Suffolk County service area.
Weeks 3 to 6
Three Long Island editorial placements secured. Old Westbury pool installation featured.
Weeks 1 to 2 (parallel)
FAQPage schema added to pool construction and Nassau County pages.
Week 12
ChatGPT check. Aquacade Pools is the first recommendation for "who builds the best inground pools in Nassau County."
#1
ChatGPT recommendation
120
reviews (was 40)
3x
consultation requests
90
days
RECONSIDERING OUR ASSUMPTIONS
Going into this study, our working assumption was that review count and velocity would be the dominant predictor of ChatGPT recommendation, since that assumption holds reasonably well for Google Maps ranking. We expected trade directory membership to matter, but we expected it to matter less than reviews.
The data did not support that assumption. Trade directory listing turned out to be the single most common shared trait among recommended contractors, present in 91 of the 100 cases, more common than any single review-based metric we measured. And critically, high review counts did very little to compensate for a missing trade directory listing. Contractors with 150 or more reviews but no trade directory membership appeared in ChatGPT recommendations far less often than contractors with under 100 reviews who did have a trade directory listing.
The only contractors in our study who broke this pattern, the 9 who appeared without a trade directory listing, had review counts and editorial mentions so far above the median that we consider them exceptions rather than evidence against the pattern.
If you take one thing from this study: if you are not in your primary trade directory, your review count is largely irrelevant for ChatGPT recommendation purposes.
Query variation caveat
ChatGPT can return different recommendations for the same query asked at different times, even within the same day, because the underlying model incorporates some degree of variability by design. Our study reflects a single-pass snapshot of each of the 360 queries run during our 3-week testing window, not an exhaustive test of every possible variation for every query.
Model version note
This study was conducted using the version of ChatGPT publicly available in July 2026. Recommendation behavior may differ across model versions, and OpenAI updates its models on a schedule outside our control, so results from a repeat of this study run today could differ.
Training data cutoff
We could not fully separate the influence of ChatGPT’s training data from any live web browsing or retrieval features active during our queries. Some recommendations may reflect real-time retrieval of current business information rather than patterns learned during training, and we were not able to cleanly isolate which mechanism was responsible for any individual recommendation.
Market and trade scope caveat
This study covers 20 mid-size to large US metro markets and 6 licensed home service trades. Results may not generalize to rural markets, to trades outside the 6 tested, or to markets outside the United States.
No paid listings disclosure
Homerankr did not pay for placement in any trade directory, review platform, editorial publication, or any other source referenced in this study, either for Aquacade Pools or for any of the 100 contractors documented in our sample. The correlations described reflect organic citation presence only.
FREQUENTLY ASKED QUESTIONS
Based on our study of 100 ChatGPT contractor recommendations across 6 trades and 20 US markets, ChatGPT recommendations are primarily driven by trade-specific directory membership (NRCA, PHCC, NECA, APSP, NARI), Google review count and velocity, Houzz verified contractor profile with project photos, editorial mentions in local publications, and FAQPage schema on primary service pages. A contractor with 3 or more of these 5 signals appeared in ChatGPT recommendations in 94 of 100 cases in our study.
Only weakly. Our study found that only 6 of the 100 ChatGPT-recommended contractors were also ranking in the Google Maps local 3-pack for an equivalent search, and 71 of the 100 did not rank in the top 10 on Google Maps at all. Google Maps and ChatGPT recommendations are governed by largely separate signals, so optimizing for one does not automatically produce visibility in the other.
A trade directory is an industry-specific membership organization such as PHCC for plumbing and HVAC, NRCA for roofing, NECA for electrical, APSP for pool building, or NARI for remodeling. In our study, 91 of the 100 recommended contractors had a verified listing in their primary trade directory, making it the single most common signal we documented. We believe AI models weight these listings heavily because they are third-party verified membership records rather than self-reported claims.
Review count still matters, but our study found it was not sufficient on its own. Recommended contractors had a median of 127 Google reviews, 4.3 times higher than contractors in the same market and trade who did not appear. However, the counterintuitive finding was that contractors with high review counts but no trade directory listing rarely appeared. The 9 contractors in our study who appeared without a trade directory listing all had 200 or more reviews and at least 2 editorial mentions, suggesting review count alone can substitute for a missing signal only at an unusually high threshold.
In the Aquacade Pools case embedded in this study, the contractor went from no identifiable ChatGPT presence to the first-listed recommendation for their trade and market in 90 days, after submitting a trade directory listing, building a Houzz profile with project photos, securing three local editorial placements, and adding FAQPage schema to two service pages. We consider 90 days a reasonable estimate for a contractor starting from zero on all 5 signals.
The specific percentages in this study apply to HVAC, plumbing, roofing, remodeling, pool building, and electrical contractors, since those were the 6 trades tested. The underlying mechanism, that AI models weight third-party verified citations like trade directories, review platforms, and editorial mentions more heavily than self-reported website content, is likely to generalize to other home service and licensed trades, but we have not tested outside these 6 categories and would treat the exact thresholds as directional rather than confirmed for other trades.
Not directly, and we did not identify any paid placement mechanism in this study. ChatGPT does not sell recommendation placement the way search engines sell ads. What contractors can influence is the underlying citation signals, trade directory membership, review generation, Houzz presence, editorial coverage, and structured data, all of which are earned rather than purchased, though they do require deliberate investment of time and, in the case of some directories and platforms, membership fees.
Most contractors do not know. The only way to find out is to run the check.
A free AI Visibility Audit from Homerankr includes a live ChatGPT and Perplexity check for your trade and city, showing exactly what AI says when a homeowner in your market asks who to call. We also audit your current citation signals against the 5-factor model from this study and show you the specific gaps between your current profile and the contractors who are getting recommended instead of you.
Get Your Free AI Visibility AuditOr write to us at found@homerankr.co