What actually happens when a homeowner asks ChatGPT for a contractor
Picture the search that used to feed you leads. A homeowner types "best HVAC company near me," scrolls the map pack, opens three tabs, reads reviews, and calls two. That behavior is moving. More owners are asking ChatGPT, Google's AI Overview at the top of the results page, or the assistant baked into their phone. They ask one question and expect one shortlist back.
When that question is "who should I hire," the answer engine has to decide which few businesses to name. It does not have a phone to call around. It reads text. Specifically, it reads the public reputation signals that already exist about you: your star rating, the count of reviews, how fresh they are, the words inside them, and whether the owner shows up in the replies. A business with 40 reviews that all say "good job" is thin material. A business with 300 reviews that name neighborhoods, describe the exact repair, and get a real reply from the owner is quotable.
This matters because the AI answer is often the only answer the homeowner sees. There is no page two to lose on. If ChatGPT names three plumbers and you are not one of them, the homeowner may never learn you exist, even if you rank fine in the old blue-link results. The click you used to fight for is now a mention you have to earn, and reviews are the raw material the engine reads to decide.
None of this replaces getting found in the map pack or ranking your site. It sits on top of it. But the lever that moves it, the thing that decides whether you make the shortlist, is the same reputation asset you already own: your reviews.
Which review signals AI answer engines actually read
Not all review signals carry the same weight. When we look at why one contractor gets quoted and a competitor two miles away does not, the difference clusters around a handful of things. Here is how they break down.
| Signal | Why the engine cares | What good looks like |
|---|---|---|
| Review volume | Enough text to summarize confidently | Deep, steady, not a burst then silence |
| Recency | Answers a "who is good now" question | New reviews every week or two |
| Star average | Simple confidence threshold | 4.7 and up, with the odd low one answered |
| Specific wording | Gives the model something to quote | Names the job, the trade, the town |
| Owner responses | Adds facts and shows the business is real | A reply on most reviews, good and bad |
Read the table and one thing jumps out: three of the five are about the words, not the number. An answer engine can only quote you if there is something worth quoting. "Fast and friendly" is filler. "They replaced our failed condensate pump the same afternoon in July" is a sentence a model can lift straight into an answer. That is the difference between being counted and being named.
Recency is the quiet one. A five-star average built over ten years, with nothing new this year, reads as stale. The engine is answering a present-tense question. A slow, steady drip of new reviews tells it you are open, busy, and current. That is why review generation is a system you run every month, not a campaign you do once.
The star average still matters, but as a floor, not a trophy. Below roughly 4.5 you look risky and get skipped. Above it, the number stops being the story and the words take over.
Why "specific" reviews beat "more" reviews for getting quoted
Every contractor is told to get more reviews. Good advice, but incomplete. For AI answers, the content of the review does as much work as the count. The engine is looking for language it can borrow to justify naming you.
Think about what a homeowner actually asks. Not "who has the most reviews." They ask "who does tankless water heater installs," "who is good with old knob-and-tube wiring," "who does metal roofs in a coastal county." Those are specific jobs. If your reviews never mention those jobs, the engine has no reason to connect you to that question, no matter how many five-star ratings you carry.
The fix is not to write fake reviews. It is to shape the ask so real customers describe the real work. A few ways to do that honestly:
- Send the review request right after the job is done, while the detail is fresh, and reference the specific work in the message.
- Ask an open question in the request: what did we fix, and how did it go, rather than "leave us a review."
- Let happy customers name the neighborhood or town themselves. That local wording is gold for "near me" style questions.
- Never coach the words. Prompt the memory, not the sentence.
Volume and specificity are not a trade-off. You want both. But if you only chase the count, you can end up with 200 reviews that all say the same three words, which gives an answer engine nothing to grab. Twenty reviews that each describe a real, named job can out-quote them. The goal is a body of reviews that reads like an accurate description of what your crew actually does, in the words of the people who paid for it.
How owner responses turn a review into something an engine can cite
Most contractors treat review responses as a chore, or skip them. That is a missed signal. Your reply is public text attached to the review, and answer engines read it right along with the customer's words. A good response does two jobs at once: it reassures the next homeowner reading, and it feeds the model facts it can use.
A reply that just says "Thanks for the kind words!" adds nothing. A reply that says "Glad the new heat pump is keeping the house comfortable, and thanks for trusting us with a same-week install" restates the trade, the job, and the timeliness. Now there are two voices confirming the same specific detail, which reads as more credible to a machine deciding what to quote.
The high-value responses are the ones on the bad reviews. A one-star with no reply looks like a business that does not care or is not there anymore. A one-star with a calm, factual, non-defensive owner response looks like a real business that stands behind its work. That contrast can matter more than the star itself. Both a homeowner and an answer engine can tell the difference between a company that hides from problems and one that handles them.
Some ground rules for responses that hold up under both human and machine reading:
- Respond to the negative ones first, and fast. Silence is the worst answer.
- Stay factual and calm. Never argue the customer's memory in public.
- Restate the actual work in your reply, without sounding like a keyword stuffer.
- Never mention private details a customer did not share themselves.
Done consistently, responses turn a passive pile of ratings into a two-sided record that reads as active, honest, and current. That is exactly the profile an answer engine wants to quote.
What review-gating and fake reviews cost you in AI answers
Two shortcuts tempt every contractor with a bruised rating: gate the reviews so only happy customers reach the public page, and buy or plant a few five-stars to pad the number. Both backfire, and the AI-answer era makes them worse.
Review-gating, filtering people to a private survey if they seem unhappy and only routing the happy ones to Google, violates the Federal Trade Commission's rules on reviews and can violate Google's policies too. It also produces a profile that reads as fake to anyone paying attention: a perfect five-star with zero low reviews and no owner responses to any friction looks staged, and increasingly it is treated that way. An answer engine leaning on that profile is a liability the platforms are actively working to reduce.
Bought or planted reviews are worse. They tend to arrive in bursts, use generic wording, come from accounts with no local history, and describe jobs that do not match your trade. Those are the exact patterns detection systems flag. When a batch gets removed, your count drops overnight and your average can swing, which is the opposite of the steady, current profile that earns a mention.
The honest path is slower and it wins. A real 4.8 with a handful of answered low reviews is more quotable than a suspicious 5.0, because it reads as true. Here is the contrast plainly:
| Shortcut | What it looks like | What it costs |
|---|---|---|
| Review-gating | Perfect stars, no friction ever shown | FTC exposure, reads as staged, low trust |
| Bought reviews | Bursts, generic text, no local history | Removal, count swings, flagged patterns |
| Honest system | Steady drip, specific wording, real replies | Slower to build, but quotable and durable |
The reputation you can defend in front of a regulator is the same one an answer engine wants to quote. There is no separate trick.
Review schema: helping the engine read what is already true
You have the reviews. The next question is whether the machines can read them cleanly. That is what review schema does. Schema is a small block of structured markup on your website that states, in a format search and answer engines parse directly, your aggregate rating and the review text you are allowed to display.
Think of it as labeling the box. Your reviews already live on Google and in your customers' words. Schema on your own site restates the rating and select reviews in a machine-clean format, so an engine reading your website does not have to guess what your reputation is. It sees a declared 4.8 average across a stated number of reviews, tied to your business, in a form it trusts.
A few honest guardrails, because schema is easy to abuse:
- Only mark up reviews that are real and that you have the right to display. Never invent aggregate numbers.
- The rating in your schema must match your actual public rating. Mismatches get ignored or penalized.
- Self-serving markup, rating yourself with no real reviews behind it, is against the rules and reads as fake.
Schema will not manufacture a reputation you do not have. It cannot turn 12 reviews into 300. What it does is remove friction: when your reputation is genuinely good, schema makes sure the engine reads it correctly instead of missing it. On a hand-coded site with the markup done right, it is one more place your true numbers are stated cleanly for the machine.
This is where reputation work touches your website directly, and it is worth doing once, properly, rather than bolting on a plugin that spits out markup that does not match reality.
A monthly review system that earns AI mentions over time
None of this is a one-time fix. The contractors who get quoted are running a loop, month after month, so their profile stays deep, current, and specific. You do not need software you will never open. You need a rhythm your crew can actually keep.
Here is the loop, in order:
- Ask every finished job, right away. Same day or next day, while the work is fresh. A text with a direct link beats a business card left on the counter.
- Prompt the story, not the star. Ask what you fixed and how it went, so the review names the real job in the customer's own words.
- Respond to every review inside a few days. Negative ones first. Restate the work, stay factual, never argue.
- Watch for new reviews across platforms, not just Google. A one-star on a smaller site still gets read by both homeowners and engines.
- Keep the schema on your site matched to your true numbers as the count grows.
Run that for a few months and the shift is visible: a steady drip of fresh, specific reviews, a full column of owner responses, and a rating that reads as real. That is the exact profile an answer engine reaches for when a homeowner asks who to hire. There is no cheat code that gets you there faster, only the loop, kept.
The honest truth about timelines: this is a build, not a switch. A profile deep and current enough to get consistently quoted usually takes several months of steady work, the same range competitive search terms take. If your rating is currently bleeding jobs, the first weeks are about stopping the bleed with responses and recent reviews, and the mentions follow the reputation, not the other way around.