A homeowner with a cracked chimney used to type a few words into Google and pick from a list of websites. Increasingly they get a written answer instead, assembled from a handful of sources, and they never see the list at all. If your business is not among the sources that answer was built from, you were never in the running, and nothing in your analytics will tell you that you lost.
Dominguez Marketing works with home services companies across ten trades, and this is the question owners ask us most. Rather than answer it with theory, we are going to show you what happened on our own accounts: the results we can point at, the numbers behind them, and the ones that did not work. Every outside statistic here is attributed to the organization that published it, and every client figure comes from our own reporting.
What Actually Changed
Three findings are worth knowing, and all three come from organizations that publish their methods.
- SparkToro and Similarweb reported in 2026 that roughly 68 percent of searches now end without a click to any website.
- Pew Research Center found in 2025 that when an AI summary appeared, people clicked a result 8 percent of the time, against 15 percent when no summary appeared. Only about 1 percent clicked a link inside the summary itself.
- BrightLocal reported in 2026 that 45 percent of consumers now use AI tools for local business recommendations, up from 6 percent.
Put plainly: fewer people are clicking through to websites at all, and a fast-growing share are asking an AI tool to recommend a local business directly. For a company that has spent years thinking about its position in a list of blue links, the list is no longer the only thing worth competing for.
What Did Not Change
Almost everything else. The businesses getting recommended by AI tools are, with few exceptions, the same businesses that were already earning trust the ordinary way: real reviews, consistent information, actual expertise, and a website that answers the questions customers ask. We have not found a shortcut, and we have not seen anyone else demonstrate one.
This matters because the pitch you are most likely to hear right now is that AI search is a brand new game requiring a brand new product. In our accounts it has behaved more like a new distribution channel for work that was already worth doing.
Three Things That Actually Happened on Our Accounts
A family-run chimney company in eastern Pennsylvania, named as the recommendation
We searched Google’s AI Mode for a chimney cap replacement in this client’s town, using a prompt taken from the set we already track for them rather than one invented to flatter the result. The company came back named as the most highly recommended option, and the sources listed underneath the answer were their own pages.
That last detail is the part worth understanding. The AI did not recommend them because of a directory listing or a review aggregator. It recommended them and then cited the company’s own content as the basis for the recommendation. The pages doing that work are unremarkable service and question pages written over the preceding year.
A chimney and stove company in Central Massachusetts, same pattern
We ran the equivalent check for a chimney and stove company in Central Massachusetts, searching AI Mode for the best chimney sweep in their city. They were returned as the top-rated local option.
Two accounts is not a study, and we are not presenting it as one. What it does establish is that a local service business with no national brand and no unusual budget can be the name an AI tool gives when someone asks who to call.
A dryer vent page that became a source more than 2,500 times
The result that surprised us most came from the least glamorous content we produce. One client was cited more than 2,500 times over three months as the source behind Microsoft’s AI answers, and we traced it back to a handful of pages, with dryer vent content doing a disproportionate share of the work.
Nobody would call a dryer vent cleaning page a flagship asset. It exists because customers ask about lint buildup and dryer fires, so we answered the question properly, with specifics, at length. That turned out to be exactly the shape of thing an AI system will quote, because it is a clear answer to a question a person actually asked.
We have kept these clients unnamed on purpose. They hired us to market their businesses, not to be case studies, and we would rather describe the result accurately than trade their privacy for a more impressive-looking page. Anyone evaluating us is welcome to ask for specifics directly.
What Our Own Numbers Look Like With the Misses Included
Here is where most agency case studies stop. We are going to keep going, because the shape of the full dataset is more useful to you than the highlight.
Across our roster, AI citations run into the thousands over recent months. That total is real and it is also lopsided in three specific ways, all of which we would want to know if we were the ones reading this.
- The measurement windows are mixed rather than one clean period, so the total is an assembled figure and not a single tidy quarter.
- Roughly 92 percent of the citations come from the Microsoft Copilot and Bing ecosystem rather than from Google. Anyone quoting a big AI citation number without saying which system produced it is hiding the most important variable.
- The distribution is severely skewed. Two clients account for about 77 percent of the total. The median client sits closer to 50 citations than to 2,500, and two clients recorded zero.
Those two zeros are the most instructive entries in the whole dataset. Both are newer accounts where the content library is still thin. Nothing was done wrong on them. There simply was not yet enough published material for an AI system to have something to quote, which is the clearest evidence we have that this is a function of accumulated content rather than a switch anyone can flip.
So the honest position is that it is clearly possible for an ordinary local business to become a source AI systems quote at real volume, and it is not yet something anyone can promise on a timeline. If an agency guarantees you AI citations, they are ahead of the available evidence.
We Can Now Measure This Directly, and It Changes the Conversation
Google Search Console added a report showing impressions inside AI Overviews and AI Mode at the URL level. We piloted it on two accounts in our August 2026 reporting, and it is the first time we have been able to connect specific pages we wrote to AI surfaces with evidence rather than inference.
| Account | AI surface impressions, 28 days | What produced them |
|---|---|---|
| Chimney and stove company, Central Massachusetts | 8,110 | Roughly 85 percent traced to one pillar content package published in February 2026 |
| Chimney company, central Indiana | 2,370 | 47 percent from a single comparison guide on vented versus ventless gas logs |
Read the second column against the third. In both cases the large majority of AI visibility came from a small number of deliberately built pages, not from the site as a whole. That is consistent with what the citation data shows, and it is the single most actionable pattern we have found: depth on a few genuinely useful pages beats breadth across many thin ones.
We are extending this report to every client in the September cycle.
Where AI Answers Get Their Sources
This has moved fast and it cuts both ways. Ahrefs reported in 2026 that 37.9 percent of the citations inside Google AI Overviews came from pages ranking in the top 10 for the query, down from 76 percent.
So ranking in the top 10 is no longer close to a guarantee of being cited. It also means a page that does not rank in the top 10 now has a real chance of being the source an answer is built from. For a small local business competing against national directories and franchise operations, the second half of that sentence is the more interesting one, and it matches what we see on our own accounts, where the pages earning citations are frequently not the pages with the best traditional rankings.
One Thing That Does Not Work
Structured data, also called schema markup, is code that describes your page to machines. It has been widely recommended as an AI visibility tactic, including by people acting in good faith.
The best available test does not support it. Ahrefs tracked 1,885 pages that added structured data against 4,000 matched control pages in May 2026 and found no lift in AI citations, along with a small but statistically significant decline for Google AI Overviews specifically.
Structured data remains worth implementing for rich results in traditional search, which is a genuine benefit and the reason we still do it. It is not defensible as a lever for AI visibility, and we stopped presenting it that way the week we read the study. If someone is selling you schema markup as an AI strategy, the fair question is which study it rests on.
What Appears to Help
With the caveat that this field is roughly two years old, our accounts and the published research point the same direction.
- Answer the actual question, plainly, near the top of the page. AI systems are assembling answers, so pages built as answers get used.
- Go after the specific and unglamorous questions, not only the high-volume ones. What it costs, how long it takes, what happens if you wait. The dryer vent result came from exactly this.
- Publish real numbers where you can stand behind them. Price ranges, timelines, and measurements give a system something concrete to quote. Vague copy gives it nothing.
- Be concrete about place. Naming the counties and towns you serve, and the conditions specific to them, is what makes a page usable for a local recommendation.
- Keep facts consistent everywhere. BrightLocal found in 2023 that 62 percent of consumers will avoid a business after finding incorrect information about it.
- Build depth on a few pages rather than thin coverage across many. Both of our measured accounts concentrated their AI visibility in a small number of substantial pages.
You will notice none of that is exotic. It is the same content discipline that has worked in search for years, pointed at the questions customers actually ask instead of at keyword lists.
How to Tell Whether It Is Working
Traditional reporting will understate this, because the entire premise is that people are getting answers without clicking. Watch a wider set of signals.
| What to watch | What it tells you |
|---|---|
| AI surface impressions in Search Console | Direct evidence of which pages are appearing in AI answers |
| Citations in AI answers across Copilot and ChatGPT | Whether you are being used as a source at all |
| Impressions holding or rising while clicks fall | Normal in a zero-click environment, not necessarily a problem |
| Branded search volume | Whether AI exposure turns into people looking you up by name |
| Calls referencing something specific from your site | The clearest real-world signal that a page did its job |
| Map pack position on core service terms | Still the primary driver of local calls for most trades |
That last row deserves emphasis. For most home services businesses, traditional local search still produces the majority of the phone calls today. AI search is where the trend points, not where all the revenue currently sits, and any plan that abandons what works now for what might work later is a bad trade.
A Realistic Timeline
On the accounts where this has worked, the sequence has looked similar. Content gets published. Traditional rankings move first, over roughly two to four months. AI citations follow, and in the clearest case the volume built over about a quarter once there was a body of material worth quoting. The accounts sitting at zero are the newest ones, which is the pattern you would expect if accumulated depth is the driver.
Anyone promising AI citations in thirty days is describing something we have not observed.
How We Work On This
Dominguez Marketing has worked with home services companies since 2018 out of Tonawanda, New York. Our approach to AI search is the same as our approach to everything else: publish work that genuinely answers customer questions, measure what we can measure, and tell you plainly which parts are evidence and which parts are still a guess.
We will show you the misses alongside the wins, because an agency that only shows you the wins is not giving you enough information to make a decision.
Want to know whether AI tools are already citing your business? Get in touch and we will run the check and tell you plainly what we find, including if the answer is not yet.
Frequently Asked Questions
What does an AI SEO agency actually do?
In practice, most of the work is making your website usable as a source when an AI tool assembles an answer: publishing clear answers to the specific questions customers ask, being concrete about pricing and location, and keeping facts consistent. It overlaps heavily with good traditional SEO. Any agency describing it as an entirely separate discipline should be asked to show the difference.
Are people really using AI to find local businesses?
Increasingly, yes. BrightLocal reported in 2026 that 45 percent of consumers use AI tools for local business recommendations, up from 6 percent, and SparkToro and Similarweb reported that roughly 68 percent of searches now end without a click to any website.
Do I need to rank in the top 10 to be cited by AI?
Not any more, and this is the most useful shift for a small business. Ahrefs found in 2026 that 37.9 percent of Google AI Overview citations came from pages ranking in the top 10, down from 76 percent. Pages outside the top 10 now have a real chance of being used as a source.
Does schema markup help with AI visibility?
The evidence says no. Ahrefs tested 1,885 pages that added structured data against 4,000 matched controls in May 2026 and found no lift in AI citations, with a small statistically significant decline for Google AI Overviews. It is still worth doing for rich results in traditional search.
Can an agency guarantee AI citations?
No, and a guarantee is a warning sign. We have seen one client cited more than 2,500 times in three months, and we have also seen two clients record zero. The median across our roster sits closer to 50. It is clearly achievable and it is not yet predictable.
How long does it take to show up in AI search?
On our accounts, traditional rankings moved first over roughly two to four months, with AI citations building afterward over about a quarter once there was enough published material to quote. The accounts with no citations are the newest ones with the thinnest content libraries.