Original research · 2026-07 edition

AI SEO Statistics: Construction (2026-07 edition)

15 questions · 45 AI responses · 3 models · measured 2026-07-04

See how AI talks about YOUR businessFull data table

The question bank

The questions we tested — sampled from real buyer journeys in construction.

Each model answered every question once, same wording, same day. These are the prompts behind every percentage on this page.

Are these hairline cracks in my drywall a sign of structural issues or just the house settling?
Is it cheaper to buy my own renovation materials and just pay for labor, or let the contractor handle everything?
What specific questions should I ask a general contractor during our first walkthrough to see if they're reliable?
What is the average cost per square foot for a 500-square-foot master suite addition right now?
Should I hire a specialized roofing company or a general handyman to fix a small leak near my chimney?
How can I verify a contractor's license and check if they have any active complaints or lawsuits against them?
What are the biggest red flags to look for when reviewing a construction contract before signing?
My basement flooded and I need a restoration crew immediately; what's a fair price for emergency water extraction?
Show all 15 questions
Is it normal for a contractor to ask for a 50% deposit upfront before any work has started?
What's the difference between a fixed-price contract and a cost-plus contract for a home remodel?
I want to open up my floor plan; how do I tell if a wall is load-bearing without tearing into the drywall?
How much extra should I realistically set aside for hidden costs when renovating an older home built in the 1950s?
What kind of timeline should I expect for a full kitchen gut renovation, and what are the common causes for delays?
Can I legally fire my contractor midway through a project if the work quality is poor, and what happens to the money I've paid?
Does adding a second bathroom actually increase my home value enough to justify a $25,000 investment?

Model by model

21-point average divergence: which AI you ask changes the answer.

The divergence index is the average gap between the most and least likely model per behavior. Higher = the models disagree more about construction buyers.

Behavior rates across 15 construction buyer questions, 2026-07 edition. Last column: average across models.
ChatGPTClaudeGeminiConsensus
Recommends hiring a professional80%60%53%60%
Suggests DIY first20%20%13%87%
Names specific providers0%0%7%93%
Gives price or cost info27%60%47%60%
Tells to check reviews13%13%7%73%
Tells to verify credentials27%13%7%60%
Mentions case studies / portfolio7%0%0%93%
Mentions local proximity33%27%20%80%
Gives selection criteria33%27%27%60%
Warns about red flags27%27%20%73%
Asks a clarifying question67%73%0%20%
Recommends multiple quotes27%13%0%67%

By model

How each assistant handled Construction questions.

Reading the 45 answers model by model shows how differently the three assistants treat the same construction questions. On the most consequential behavior — whether to send the buyer to a professional at all — the rate ranged from 80% (ChatGPT) down to 53.3% (Gemini), a 27-point gap on an identical question set.

Across the 15 construction answers it produced, ChatGPT recommended hiring a professional in 80% of them and suggested a DIY approach first 20% of the time. It named a specific provider in 0% of answers (about 0 distinct providers per answer) and included price or cost information 26.7% of the time. ChatGPT asked a clarifying question before answering in 66.7% of cases, warned about red flags or scams in 26.7%, and told the buyer to verify credentials in 26.7%, averaging 598 words per answer. On the remaining cues it told the buyer to check reviews in 13.3%, pointed to case studies or a portfolio in 6.7%, and framed the choice around local proximity in 33.3%; a selection-criteria checklist appeared in 33.3% of its answers and a recommendation to gather multiple quotes in 26.7%.

Across the 15 construction answers it produced, Claude recommended hiring a professional in 60% of them and suggested a DIY approach first 20% of the time. It named a specific provider in 0% of answers (about 0 distinct providers per answer) and included price or cost information 60% of the time. Claude asked a clarifying question before answering in 73.3% of cases, warned about red flags or scams in 26.7%, and told the buyer to verify credentials in 13.3%, averaging 305 words per answer. On the remaining cues it told the buyer to check reviews in 13.3%, pointed to case studies or a portfolio in 0%, and framed the choice around local proximity in 26.7%; a selection-criteria checklist appeared in 26.7% of its answers and a recommendation to gather multiple quotes in 13.3%.

Across the 15 construction answers it produced, Gemini recommended hiring a professional in 53.3% of them and suggested a DIY approach first 13.3% of the time. It named a specific provider in 6.7% of answers (about 0.2 distinct providers per answer) and included price or cost information 46.7% of the time. Gemini asked a clarifying question before answering in 0% of cases, warned about red flags or scams in 20%, and told the buyer to verify credentials in 6.7%, averaging 253 words per answer. On the remaining cues it told the buyer to check reviews in 6.7%, pointed to case studies or a portfolio in 0%, and framed the choice around local proximity in 20%; a selection-criteria checklist appeared in 26.7% of its answers and a recommendation to gather multiple quotes in 0%.

Taken together, ChatGPT is the assistant most likely to route a construction buyer to a professional (80%) and Gemini the least (53.3%). ChatGPT produced the longest answers, at 598 words on average. Specific providers were named most often by Gemini (6.7%) — even there, roughly one answer in 15 carried a name.

Where they disagree

The behaviors where the choice of model changes the answer.

The divergence index for this study is 20.7 points — the average distance between the most and least likely model across the coded behaviors. The gaps below are where which assistant a construction buyer happens to ask matters most:

  • Asks a clarifying question: from 0% (Gemini) to 73.3% (Claude) — a 73-point spread.
  • Gives price or cost information: from 26.7% (ChatGPT) to 60% (Claude) — a 33-point spread.
  • Recommends hiring a professional: from 53.3% (Gemini) to 80% (ChatGPT) — a 27-point spread.
  • Recommends multiple quotes: from 0% (Gemini) to 26.7% (ChatGPT) — a 27-point spread.
  • Tells the buyer to verify credentials: from 6.7% (Gemini) to 26.7% (ChatGPT) — a 20-point spread.

The widest single gap — asks a clarifying question, 73 points — means a construction buyer can receive materially different guidance on the same question depending only on which assistant they happen to open, so any visibility strategy built on a single model's behavior describes only part of the construction market.

Where they agree

The points of near-consensus in Construction.

On other behaviors the three models move almost in lockstep — the points of near-consensus for construction, where all three landed within a few points of each other:

  • Tells the buyer to check reviews: 6.7%–13.3% across all three (a 7-point spread).
  • Gives selection criteria: 26.7%–33.3% across all three (a 7-point spread).
  • Suggests a DIY approach first: 13.3%–20% across all three (a 7-point spread).
  • Names a specific provider: 0%–6.7% across all three (a 7-point spread).

Measured question by question, the three assistants coded a response the same way most consistently on "names a specific provider" (identical coding in 93.3% of questions) and least consistently on "asks a clarifying question" (20%).

Every behavior, measured

All twelve coded behaviors for Construction, averaged across the three models.

The behaviors AI models reproduce most often for construction are recommends hiring a professional (64.4% on average), asks a clarifying question (46.7%) and gives price or cost information (44.5%); the rarest are mentions case studies or portfolio (2.2%), names a specific provider (2.2%) and tells the buyer to check reviews (11.1%). Each figure below is the share of a model's 15 answers in which the behavior appeared at least once, averaged across the 3 models with the full per-model range in parentheses:

  • Recommends hiring a professional: 64.4% on average (ChatGPT 80%, Claude 60%, Gemini 53.3%) — a 27-point spread.
  • Asks a clarifying question: 46.7% on average (ChatGPT 66.7%, Claude 73.3%, Gemini 0%) — a 73-point spread.
  • Gives price or cost information: 44.5% on average (ChatGPT 26.7%, Claude 60%, Gemini 46.7%) — a 33-point spread.
  • Gives selection criteria: 28.9% on average (ChatGPT 33.3%, Claude 26.7%, Gemini 26.7%) — a 7-point spread.
  • Mentions local proximity: 26.7% on average (ChatGPT 33.3%, Claude 26.7%, Gemini 20%) — a 13-point spread.
  • Warns about red flags or scams: 24.5% on average (ChatGPT 26.7%, Claude 26.7%, Gemini 20%) — a 7-point spread.
  • Suggests a DIY approach first: 17.8% on average (ChatGPT 20%, Claude 20%, Gemini 13.3%) — a 7-point spread.
  • Tells the buyer to verify credentials: 15.6% on average (ChatGPT 26.7%, Claude 13.3%, Gemini 6.7%) — a 20-point spread.
  • Recommends multiple quotes: 13.3% on average (ChatGPT 26.7%, Claude 13.3%, Gemini 0%) — a 27-point spread.
  • Tells the buyer to check reviews: 11.1% on average (ChatGPT 13.3%, Claude 13.3%, Gemini 6.7%) — a 7-point spread.
  • Names a specific provider: 2.2% on average (ChatGPT 0%, Claude 0%, Gemini 6.7%) — a 7-point spread.
  • Mentions case studies or portfolio: 2.2% on average (ChatGPT 6.7%, Claude 0%, Gemini 0%) — a 7-point spread.

Trust signals

How well the models protect the construction buyer.

Beyond whether to hire, the rubric codes how carefully each assistant protects the construction buyer once a decision is made. Telling the buyer to check reviews or ratings appeared in 11.1% of answers on average. Verifying credentials or certifications appeared in 15.6%. Warning about red flags or scams appeared in 24.5%.

On structuring the decision, a selection-criteria checklist showed up in 28.9% of answers on average and a recommendation to gather multiple quotes in 13.3%. The single least-reproduced protective signal for construction is "tells the buyer to check reviews" at 11.1% on average — the clearest opening for content that supplies it, since the models are not yet reliably surfacing that guidance on their own.

Referral behavior

Do AI models name Construction providers?

For service providers the decisive question is whether these systems name anyone at all. Across 45 construction answers, a specific provider was named in 2.2% of responses on average — roughly 0.1 distinct providers per answer. In practice the assistants behave far more as an explanatory layer than as a referral engine for construction: visibility comes from being the reasoning a model reproduces, not from being the named recommendation.

The question set

What these 15 Construction questions cover.

The 15 questions behind every percentage on this page were drawn from real construction (home services; buyer hiring decisions for this specific service) buyer journeys. Each was put to all 3 models once, with identical wording, so the rates above describe how the assistants handled this exact construction question set — not a general prior or a hand-picked subset. The full list is shown earlier on this page; the coded percentages are what those specific questions produced.

How to read this

A note on the numbers.

A percentage here is the share of a model's 15 answers in which the behavior appeared at least once — not a confidence score. Because each model answered every question exactly once on 2026-07-04, the figures describe this specific construction question set and snapshot rather than a general prior. The full protocol and coding rubric are documented in the study methodology.

Methodology

A controlled snapshot, documented end to end.

15 standardized buyer questions per industry, one response per model per question (ChatGPT (gpt-5-mini), Claude (claude-sonnet-5), Gemini (gemini-3-flash-preview)), collected 2026-07-04, coded against a fixed 12-behavior rubric with human QA. AI outputs vary with model version, location and time — figures describe this sample and window, and are refreshed each edition. Read the full methodology →