We Checked 23 Home Service Websites Against AI Search. Here's What ChatGPT and Google's AI Actually Find.

Between September 10 and October 15, 2026, Adimize audited 23 home service businesses in 13 states for AI search visibility: 16 junk removal and hauling companies, two dumpster rental companies, a landscaper, a towing company, an electrician, a fence-post contractor and a multi-trade startup. Where we had AI-search tracking data, 6 of 13 businesses had never been mentioned by ChatGPT, Google's AI Overviews, Google's AI Mode or Gemini. Nine of the 23 had the same spam-link network pointing at them. Sixteen couldn't agree with their own listings on their name, phone number, address or hours. The median homepage took about 10 seconds to show its main image on a phone.

None of the businesses are named here. Every one of them got a private report. What follows is the pattern across all of them, with the numbers, because the pattern is more useful than any single site.

The numbers at a glance

Who we looked at, and how

These weren't picked for a study. They're businesses that asked for an audit, or that we audited before a sales conversation, over five weeks. So the sample skews toward owner-operators doing under a few million a year, toward junk removal, and toward companies that already suspected something was wrong. Treat the percentages as a field report, not a census.

For each one we read the live site on a phone and a desktop and the code behind it, ran Google's Lighthouse speed test on phone settings, read the Google Business Profile and searched Google Maps from the center of the towns the site claims, searched the web for the business name and phone number, checked the directories AI tools draw from (Yelp, BBB, Nextdoor, Thumbtack, Angi, MapQuest), and, for 15 of the 23, pulled Semrush's AI Search report, which tracks how often ChatGPT, Google's AI Overviews, Google's AI Mode and Gemini mention a business and cite its pages. Full method and limits are at the bottom.

1. AI tools have almost nothing to say about most of them

Of the 13 businesses where Semrush reported a mention count, 6 had zero mentions across all four AI tools, and 12 had five or fewer. One business, a junk removal company founded in 2025, had 12.

Google's AI Overviews, the answer box most people see first, were the emptiest: 10 of the 15 businesses with tracking data had zero AI Overview mentions. The visibility that did exist lived mostly in ChatGPT and Gemini.

The more interesting number is who gets cited instead. For 8 of the 23, when an AI tool answered a question about that business's own market, it cited someone else's website: national trash-schedule directories that don't own a truck, a used-parts site, Yellow Pages, Trustpilot (cited more often than the business's own site, for a company with 678 Google reviews), a state government page, Nextdoor, and in two cases 1-800-GOT-JUNK or Junkluggers, cited nearly as often as the local company being asked about.

Why: AI tools don't invent recommendations. They assemble them from what's already written about a business: the Maps listing, review sites, directories, articles, and the business's own pages. A company that exists only on Google and its own website gives them one source. The franchise has fifty.

2. Nine of them have the same spam-link network pointing at their site

In 9 of the 23 backlink profiles, the most common anchor text was an advertisement for selling links: "high quality dofollow backlinks DA 50," "premium seo link building experts," "proven guest post service," "trusted manual outreach backlinks." The linking pages carry titles like "Boost your Google rankings with Premium Backlinks" and sit on domains that sell links.

The same domains and the same anchor phrases showed up on a junk removal company in Pennsylvania, a towing company in New Mexico, a hauler in Texas, a fence-post contractor in Michigan, an electrician in Oklahoma, and a junk removal company in California. These businesses have nothing to do with each other. One had 1,600 links from 1,300 websites, with a single anchor phrase on 1,063 of them, and not one link from a real website in its own state.

Every one of the nine had a Semrush authority score between 2 and 9 out of 100. Several had more than 150 referring domains. That combination, lots of links and no authority, is what a link-seller's output looks like from the outside.

We can't tell from outside whether an owner bought a package years ago, a past agency did, or the site was simply added to a seller's list. It doesn't matter. Google's spam policies name bought links as link spam, these do nothing visible for any of the nine, and the links that would help, from a contractor who uses them, a supplier, a local directory, a community page, are the ones none of them had.

What to do: Export your backlinks (Semrush, Ahrefs, or Google Search Console's Links report). Sort by anchor text. If the top anchors are sales copy for an SEO service, that's the network. Disavow it if you can, and ignore anyone who offers to sell you more.

3. Sixteen of them can't agree with themselves on who they are

16 of the 23 businesses gave out a different name, phone number, address, hours, or service area on their website than on their Google listing or other directories.

The specifics were worse than the count:

• 8 spelled their own business name three or more different ways across their own site, Google listing and profiles. One dumpster company used five. One used eight names across two businesses and four web addresses. One misspelled its own name in the title tag of every page.
• 4 had conflicting phone numbers. One dumpster rental homepage dialed three different numbers depending on which button you tapped, and the number painted on the dumpster in the hero photo matched only one of them. One contractor's Google listing has a 517 number; his website, its code and its Call Now buttons have an 810 number.
• 5 told Google a different rating than Google has. Three sites say "5.0" on the page, two of them in the code Google reads too, while the Google listing says 4.9. Two show a review count on the homepage that's 23 to 46 percent lower than the listing.
• 6 share a name with another company somewhere else, and in most of those cases the other company has the better directory footprint. Search one Michigan contractor's exact name and the first result is a same-named company in Ontario.
• 2 had a street address on directories or public records that their own website never states. For one of them it's a residential street, published by MapQuest and Yahoo Local next to the phone number. If you're a service-area business and a directory has your street, it's probably your house, and it's syndicating.

Why it matters: Google and AI tools decide whether a listing and a website belong to the same business by matching name, address and phone across every source. A mismatch doesn't get you penalized. It gets you doubted, which is quieter and costs more.

4. Eight promise "upfront pricing" and publish no price

8 of the 23 websites use the words "transparent," "upfront" or "instant" pricing in their copy and never show a number. Of the 18 sites where we checked for pricing, 6 publish an actual range.

One pricing page is headed "Transparent pricing" and contains no figure, but does explain to the customer that the quote includes the company's profit margin. One site's own text says "Pricing Tier Matrix" and lists five tiers with no number next to any of them; the real prices sit inside an embedded CRM widget that search engines and AI crawlers don't render. One homepage promises a photo-based instant price, and the button goes to a contact form.

Why it matters: "How much does junk removal cost in [town]" is one of the highest-volume questions in the trade, and when an AI tool answers it, it quotes a page with a number on it. Pages without a number get skipped. The six businesses that published ranges ($99 to $199 for a small load, $375 for a 10-yard dumpster, $150 minimum pickup) are the ones whose pages get cited for cost questions. The fear is that a published price scares people off. What we see is the opposite: no price loses more jobs than an honest range does.

5. The median homepage takes 10 seconds to show its main image on a phone

Across the 12 sites where we measured it, the main image appeared after a median of 10.1 seconds on Google's phone test. Google's target is 2.5 seconds. Six were over 10 seconds; the slowest took 41.5 seconds on its best of three runs. All 8 sites where we measured page weight were over 3 MB, and one homepage weighed 40 MB.

The causes were almost never the hosting. Where we looked at time-to-first-byte, the server answered in under a quarter of a second. The weight came from things nobody looked at after launch:

• A logo file of 1.7 MB displayed at the size of a postage stamp, a third of the whole page.
• A 754 KB logo on every page of a Wix site, 18 percent of the page.
• Google reCAPTCHA loaded seven separate times on one homepage (2.4 MB of spam protection), three times on two others.
• A booking widget, payment processor and three captchas on a homepage, for a visitor who only wanted to read about estate cleanouts.
• 27 MB of photos, including one 8 MB and one 6 MB file, plus 6 MB of video, on a site that also showed 30 broken images because of how its image host was configured.

Google's own research found that more than half of mobile visitors abandon a page that takes longer than three seconds. A customer standing in a garage looking at a pile is not waiting 41 seconds.

6. Fourteen have a broken, missing or misdirected call or text button

14 of 23 sites had at least one place where the primary action, calling or texting the business, didn't work the way a customer expects.

We found a call button whose link was tel:000000000 (nine zeros, under the strongest sales copy on the page). A header call link of tel:%20615-... with an encoded space that some phones won't dial. A sticky call button on every page linked to tel: with no number after it. A "Text Us" button wired to place a phone call. A "Send us a text" button that opens a five-field form. A "Get a Quote" button that places a call. Two of the four pages in a site's main menu with no header, no menu and no phone number at all. A phone number that appears once, in the footer, as plain text, on a site whose own process says "text us a picture." Two "Book Now" and "Send Message" buttons that link to # and do nothing. A review pop-up that slides up over the sticky "Call" button on a phone, covering it at three of six scroll points we checked.

None of these generate an error. The tap fails silently and the customer calls the next company. There's no way to know how many calls any of these have eaten, because a click that reaches nothing is never logged.

7. Page-count SEO is producing page counts

Six of the 23 sites had been built up with large numbers of templated pages, and in every case the traffic those pages produced was close to zero.

• More than 160 pages on a dumpster rental site, including dozens of micro-neighborhood pages and several that had nothing to do with dumpsters at all. All but two of those pages got zero organic visits, and the company's own name ranked on more than two dozen pages at once with no clear winner. The domain had real authority that most competitors would trade for, and it produced fewer than 20 organic visits a month.
• 45 "what we take" pages on a junk removal site (gaming console disposal, bed-frame removal, leaf-pile disposal). 387 ranking keywords, 111 visits a month, 72 of them people searching the company's name.
• 148 blog posts on a contractor's site, published at exactly eight a month by the website platform, each 476 to 574 words, not one naming a town in the company's state. Organic visits: zero.
• 33 town pages launched at once from two templates, illustrated with files named ai-generated-IMAGE.jpg.
• Six sprinkler-repair town pages that were 77 to 91 percent identical once you swap the town name. 58 of 78 town pages with no town in the headline.
• Nine hand-written blog articles that rank, one of them sixth for a 70-a-month term, producing one visit a month combined, because none contains a price, an item list, or anything a machine can extract.

The pattern is the same everywhere: the number that gets reported (pages, keywords) grows, and the number that rings the phone doesn't. Google's spam policies now name "scaled content abuse" specifically. AI tools skip templated pages for the simpler reason that there's nothing on them to quote.

8. Nineteen are still wearing the builder's leftovers

19 of 23 sites carried visible leftovers from the template, the platform or the person who built them.

A page title of "Home | My Site 1" on every page. An "Under construction" item in the main menu. A booking page whose delivered code reads "No categories and services added!" with a button labeled "Test Button." A homepage share card that lists the web designer's personal Gmail address as the author. Another that carries a marketing agency's name as "Written by." A hidden copy of the menu listing Malibu, Beverly Hills and Koreatown on a Pennsylvania site, with each one linked to a Pennsylvania town page. A Google Maps link whose place ID is 0x123456789abcdef. A stock photo in the hero with another company's sign, street address and phone number legible in it. A logo whose file name, and therefore its alt text, is "ChatGPT Image Aug 26, 2026." Six social icons that link to facebook.com and instagram.com, not to any page of the business. An About page at /blank. Services set up as $0.00 products with "Add to cart" buttons on an industrial electrician's site. "Report abuse" twice in the footer of a site built on Google Sites. British spelling ("kerb," "programme," "labourer") on a site serving the DC suburbs.

Customers notice some of these. Machines read all of them. A file named ai-generated-IMAGE with the alt text "our truck and crew in Tampa" is a claim about work that didn't happen, read by a system that is deciding whether to trust the page.

9. Three locked search engines or AI out by accident

One site served <meta name="robots" content="noindex"> on every page, so no search engine or AI tool could index any of it. One had its Services and About pages set to members-only, so visitors and crawlers hit a sign-up wall. One had a robots.txt rule that refused an AI crawler while the owner was actively trying to get cited by AI tools.

Related, and more common: on 7 of the 23 sites, the most important content only existed after JavaScript ran. A quote form and the prices inside an embedded CRM widget. Reviews in a third-party widget, so the delivered page shows a heading and no reviews. A "years of experience" counter that reads "0+" until a script animates it. A service-area map that loads as about:blank. A FAQ accordion that may only deliver its first answer. Most AI crawlers don't run JavaScript. To them, those pages say the business has no prices, no reviews and zero years of experience.

10. Outside Google, eight of them barely exist

8 of the 23 had either no listing we could find outside Google and their own website, or a single thin one. Two had no Google Business Profile at all. Two had a Google listing that appeared to be unclaimed, one of them carrying 122 five-star reviews.

Meanwhile, in 6 of the 23 markets, the town page a customer or an AI tool lands on belongs to a franchise. 1-800-GOT-JUNK has a page for nearly every town these businesses serve. College Hunks has a page with prices and five FAQs for a Kentucky town where the local company's entire site is under 500 words. Junkluggers, LoadUp and Lugg all have a Bethesda page; the local startup serving 106 towns has none.

And the proof these businesses do have is often on the one platform AI tools can't read. One company has 7,799 Facebook followers, more than most of the companies ahead of it on Maps, and one Yelp review. Facebook keeps posts behind a login. Yelp doesn't.

What the ones doing well had in common

A few of the 23 were ahead, and they were ahead for boring reasons.

• A published price range in plain text. Small load $99 to $199, medium $250 to $450, large $500 to $850. Ten-yard dumpster $375 with the weight limit stated. Minimum pickup $150. Those pages get cited for cost questions.
• Town pages that say something true about the town. One site's page for a waterfront town talks about lanais, docks, boat equipment and snowbird turnover. Its page for the inland town next door talks about apartment move-outs near the town center. That's the opposite of swapping the town name into a template, and it's why those pages rank.
• Reviews written into the page, with names, service and town. Not a widget. One site has eight on the homepage in plain text. Another has ten.
• A named owner with a real story, on the page people land on. Seven of the 23 are owned by a veteran, a firefighter or a paramedic. That's exactly the kind of verifiable, repeatable fact AI tools include in a recommendation. In four of those seven, the homepage never names the owner.
• Hours and phone that match the Google listing exactly. Rarer than it should be.
• Fast, because it's simple. The fastest site in the set (3.4 seconds, zero blocking time) was a three-page Elementor site with a caching plugin and nothing exotic on it.

The 10-minute self-check

You can run most of this audit on your own business before lunch.

1. Ask the tools. In ChatGPT, Google's AI Mode and Perplexity, ask: "What are the best [your trade] companies in [your town]?", "How much does [your service] cost in [your town]?", and "Is [your business name] in [your town] any good?" Save the answers. Note who gets named and which websites get cited.
2. Search your name and your phone number in Google, in quotes. If the first results are a different company with your name, or nothing, that's your footprint.
3. Search your trade on Google Maps from your own town, then from two towns you claim on your website. Count where you are.
4. Open your site on your phone and tap every call and text button. All of them. On every page.
5. Compare your Google listing to your website: name, phone, hours, address, rating, review count. Character for character.
6. Run your homepage through PageSpeed Insights on mobile. Look at the main-image time and the page weight.
7. Look at your site with JavaScript off (or paste your homepage URL into a tool that shows the raw HTML). Are your prices and reviews there?
8. Pull your backlinks and sort by anchor text. If the top anchor is an ad for an SEO service, you've met the network.
9. Search your website for the words "transparent," "upfront" or "instant" next to "pricing." Then search it for a dollar sign.
10. Open your robots.txt (yourdomain.com/robots.txt) and your homepage source for the word "noindex."

Do that, and you'll know more about your AI search position than most agencies will tell you.

What to fix first

Across all 23 reports, the order was nearly always the same, because the cheapest fixes were also the ones blocking everything else:

1. Anything that tells search engines or AI to leave: noindex tags, login walls, crawler blocks. Minutes.
2. The call and text buttons. Minutes.
3. One name, one phone number, one address, one set of hours, identical on the website, in its code, on Google and on every directory. An afternoon.
4. The single heaviest file on the homepage. Usually a logo or a loaded-three-times script. An hour.
5. A real price range, on a page a machine can read. An hour.
6. Reviews and the owner's story as plain text on the pages people land on. A day.
7. Then, and only then, pages for the towns you actually drive to, each saying something true about that town.

Everything else, including the spam links, the builder leftovers and the structured data, comes after those seven.

Frequently asked questions

Why doesn't ChatGPT recommend my business? Because it has nothing to build a recommendation from. AI tools assemble answers from the Google Maps results, review sites, directories and articles that already mention a business. If your business exists only on Google and your own website, you have one source and the franchise has fifty. In our set, 6 of 13 businesses with tracking data had never been mentioned by any of the four major AI tools.

What is an AI mention, and how is it different from a citation? A mention is an AI tool naming your business in an answer. A citation is the tool linking to one of your pages as a source. You can have citations without mentions (your page gets used, your name doesn't get attached) and mentions without citations (the tool names you from what others wrote). Both matter. Semrush's AI Search report tracks both across ChatGPT, Google's AI Overviews, AI Mode and Gemini.

My website has hundreds of backlinks. Why is my authority score so low? Almost certainly because the links are spam. In 9 of 23 profiles we reviewed, the most common anchor text was sales copy for a link-selling service, and every one of those sites scored between 2 and 9 out of 100. Volume from a link network adds nothing; one link from a local contractor, supplier or news site is worth more than a thousand of them.

Does it hurt to have my phone number or business name slightly different on different sites? Yes, quietly. Google and AI tools confirm that a listing and a website belong to the same business by matching name, address and phone across sources. A mismatch doesn't trigger a penalty. It lowers confidence, and a business the tools aren't confident about is one they don't recommend. Sixteen of the 23 businesses we checked had at least one such mismatch.

Should I publish my prices? Yes, as a range. "How much does [service] cost in [town]" is one of the most common questions people ask search engines and AI tools, and the answer is always built from a page with a number on it. In our set, only 6 of 18 sites we checked published a range, and 8 promised "upfront pricing" without one.

Do AI tools read JavaScript? Mostly not. Most AI crawlers read the page as the server delivers it. Anything that appears only after a script runs (an embedded quote form, a review widget, an animated counter, a map) isn't there as far as they're concerned. Seven of the 23 sites kept their most important content, prices or reviews, in exactly that kind of widget.

Is a 5.0 rating on my website a problem if Google shows 4.9? It's a small trust problem and a possible markup problem. Google doesn't show star ratings for reviews a business publishes about itself, and a self-reported rating in your code that doesn't match your Google listing is one more thing for a tool to doubt. Five of the 23 sites had a rating or review count that disagreed with their Google listing.

Method and limits

Sample. 23 businesses, reviewed between September 10 and October 15, 2026: 16 junk removal and hauling companies, 2 dumpster rental companies, and one each of landscaping, towing, electrical, post-hole digging and a multi-trade startup, across CA, CO, DC, FL, KY, MI, NC, NM, OK, PA, SC, TN and TX. These were inbound audit requests and pre-sales reviews, not a random sample. Owner-operators doing under a few million a year are over-represented, and so are businesses that already suspected a problem. One review covered two businesses under one owner.

What we did. Read the live site on a phone and a desktop, including the code; ran Google's Lighthouse or PageSpeed Insights on mobile settings (three runs where we ran Lighthouse locally; one run from PageSpeed Insights otherwise); read the Google Business Profile from the public panel without logging in; searched Google Maps from the center of the towns the site claims (Maps results shift with searcher location, and we report the view from the town center); searched the web for the business name and phone number; checked Yelp, BBB, Nextdoor, Thumbtack, Angi, HomeAdvisor, MapQuest and Yahoo Local where applicable; looked up domain registration records. For 15 of the 23, we pulled Semrush's Domain Overview, AI Search, organic keywords and backlink anchor reports; for 13 of those, Semrush reported a mention count we could read. Page speed was measured on 13 sites and a main-image (LCP) figure recorded for 12; page weight was recorded for 8.

What we didn't do. We did not log into any Google Business Profile, submit any form, or run ChatGPT, AI Mode or Gemini ourselves as a measurement (the AI numbers are Semrush's tracking, not our own prompts). "No structured data could be confirmed" means none was visible in the delivered HTML; it doesn't rule out markup injected by script. Traffic figures are third-party estimates. "Zero" in Semrush means the tool recorded nothing, which for a domain under a year old can mean "not yet tracked" rather than "never mentioned"; we counted those as zero only where the domain was old enough to be tracked.

Counting. Each finding was counted once per business, from the delivered client report, re-read line by line for this piece. Where a check wasn't run on a business, that business is excluded from that denominator, which is why some figures are "of 13," "of 15," "of 18" or "of 12" rather than "of 23."

We'll update this page as the set grows. The next version will be built from the same findings sheet, with each new audit added as a row.

---

Justin Hubbard runs Adimize, a marketing agency for home service businesses, and has owned and operated a junk removal and dumpster rental company for over 12 years. The audits in this report were performed by Adimize; no business is named and no client data beyond the aggregate counts above is published. If you want to know where your own business stands, the self-check above is where we'd start.

Want this run on your business? Request an audit.