Sabi Ventures · Field Study · August 2026
The Plumbers AI Forgets
We asked four AI assistants to recommend a plumber in Bedford, Massachusetts, 80 times. Then we checked who they named, and who they never mentioned once.
80 sessions · 4 engines · 5 questions · 82 companies tracked · run 28 August 2026
Question asked
"I need to find a good plumber and I don't have anyone I trust. Who are my best options? I live in Bedford, MA."
Representative answer
- Iovino Plumbing, Heating, and Cooling — Bedford
- Chevron Plumbing — Burlington
- Doug Anderson Plumbing and Heating — North Billerica
- Sedona Plumbing and Heating — Winchester
- Kirkland and Shaw — Burlington
Composite of the most frequently named companies across 80 recorded sessions. Individual answers named between 3 and 10 companies.
21 of the 39 plumbers in Google Maps' top 40 were never named. Not once, in 80 answers.
They are licensed, reviewed, and ranked well enough for a homeowner to find them on Google. In this study, to an AI assistant, they did not exist.
Curious what an AI visibility audit would find for your business?
This is the same kind of AI discovery test we run for individual businesses, tuned to your services and service area.
Seven findings
What we found
Over half of well-ranked local plumbers were invisible
Of 39 companies drawn from the top 40 Google Maps results for plumbers serving Bedford, 21 were never mentioned in any of the 80 recorded answers. Eighteen were mentioned at least once.
AI amplifies the top of Google and erases the middle
Google rank still matters. Companies ranked 1–10 on Maps drew 127 mentions between them; companies ranked 31–40 drew two. But the fall-off is far steeper than on Google itself, where a page-two listing still gets seen. Here, the middle of the pack does not appear at all.
The most-recommended plumber for Bedford is not in Bedford
The company named most often across all four engines is based in Winchester, roughly twelve miles away, and does not appear in the Google Maps top 40 for Bedford at all. It was named in 46% of answers, more than any company actually located in town. Across the whole study, companies based in Bedford itself drew 8.4% of all 393 mentions, and exactly one of the ten most-mentioned companies is located in town.
The four engines barely agree with each other
Out of 82 companies, exactly one was named by all four engines. Overlap between any two engines ranged from 8% to 31%. A company can be the top result on ChatGPT and completely absent from Gemini, and nothing on its own website would tell it so.
Answers are far more concentrated than search results
Three companies took 26% of all 393 mentions. Ten companies took 57%. A page of Google results shows ten businesses at once; an AI answer typically names three to six, and the same names recur.
Asking the same question twice gives different answers
Each question was asked four times on each engine, with identical wording in a fresh session every time. Any two of those identical runs agreed with each other only about half the time. Of every instance where a company was named, 42% happened in just one of the four runs, and only 27% held across all four.
Only fifteen companies were ever named first
Out of 82 companies, 15 ever held the first slot in an answer, and three of them took it in 58 of the 80 answers. Five of the ten most-mentioned companies were never named first, not once.
82 tiles · one per company
Every company in the study
Each solid tile is a company named at least once across the 80 answers. Each dashed tile is a company that was never named. Tile shade reflects how often a company appeared.
Named at least once — 55 companiesNever named — 27 companies
Companies that were never named are not identified here. If you want to know whether your company is one of them, ask and we will tell you privately.
Share of voice · every company named in 5% or more of answers
Who the engines actually name
Share of voice is the percentage of the 80 recorded answers in which a company was named. Position within an answer is not weighted here. Thirty companies cleared 5%. The remaining 25 that were named at all fell below it, most appearing only once or twice.
First-place finishes · 80 answers, 80 first slots
Being named is not the same as being first
Share of voice counts how often a company was named anywhere in an answer. It does not distinguish between being the first name a homeowner reads and being fifth on a list of six. That distinction decides who gets the call.
15 of 82
Only fifteen companies were ever named first in any of the 80 answers. Three of them held the first slot in 58 of them.
The order changes when you weight by position. Sedona was named more often than anyone, in 37 answers, but led only 12 of them. Iovino was named 33 times and was first in 31 of those, so when it appears at all it is almost always the first name on the page.
Named firstNamed in a lower position
The ten most-mentioned companies, split by whether they led the answer. Bar length is total mentions out of 80.
Five of those ten companies were never named first. Doug Anderson was mentioned in 26 answers and led none of them. Woody's, Young Buck, Chase and North Point are in the same position: consistently present, consistently second or lower.
First-place rate
| Company | Times named | Times first | First-place rate |
|---|---|---|---|
| Iovino Plumbing | 33 | 31 | 94% |
| Dale Plumbing Services | 3 | 2 | 67% |
| Kirkland and Shaw | 32 | 15 | 47% |
| Roto-Rooter | 8 | 3 | 38% |
| Perricone Home Services | 11 | 4 | 36% |
| Pespisa Company | 6 | 2 | 33% |
| Sedona Plumbing | 37 | 12 | 32% |
| Jeneral Plumbing | 4 | 1 | 25% |
| Boston Tank Swap | 4 | 1 | 25% |
| Ciardi Plumbing & Heating | 17 | 3 | 18% |
| Chevron Plumbing | 16 | 2 | 12% |
| 978 Plumbing | 10 | 1 | 10% |
| Ostroski Plumbing & Heating | 12 | 1 | 8% |
Companies that led at least one answer, ranked by how often they led when named. Two further companies, Green Energy and Rezy Save, each led one of the two answers that named them; they are omitted here because a two-answer sample carries no signal. First-place counts sum to 80, one per answer.
Mentions by Maps rank band
Google Maps rank still predicts a lot
Across the 39 companies drawn from Google Maps, rank and AI mentions correlate at r = −0.60. Better Maps rank means more AI mentions. What changes is the shape of the drop.
| Maps rank | Companies | Named at least once | Total mentions |
|---|---|---|---|
| 1–10 | 9 | 8 | 127 |
| 11–20 | 10 | 6 | 68 |
| 21–30 | 10 | 3 | 11 |
| 31–40 | 10 | 1 | 2 |
A company at Maps position 25 still receives clicks from Google. In this study it received almost nothing from AI.
The same eight companies, split by engine
Which AI recommends which plumber
This is the finding with the most practical consequence. Visibility does not transfer between engines. The top eight companies split almost cleanly into two camps.
ChatGPTClaudeGeminiPerplexity
Iovino drew 33 mentions, all from ChatGPT and Claude, none from Gemini or Perplexity. Ciardi drew 17, all from Gemini and Perplexity. Chase Plumbing drew 14, every one from Claude.
How much the engines overlap
| ChatGPT | Gemini | Claude | Perplexity | |
|---|---|---|---|---|
| ChatGPT | — | 15% | 30% | 26% |
| Gemini | 15% | — | 8% | 31% |
| Claude | 30% | 8% | — | 15% |
| Perplexity | 26% | 31% | 15% | — |
Percentage of companies named by both engines, out of all companies named by either. One company out of 82 was named by all four.
How many seats each engine has
The engines differ in how many companies they will name at all. This sets a hard ceiling on how many businesses can be visible on each one.
| Engine | Names per answer | Range | Distinct companies ever named |
|---|---|---|---|
| Claude | 6.0 | 5–8 | 22 |
| ChatGPT | 5.8 | 4–9 | 26 |
| Perplexity | 4.0 | 3–5 | 23 |
| Gemini | 3.8 | 3–6 | 19 |
Averaged over 20 answers per engine. Across all 393 mentions, an answer named 4.9 companies on average.
Gemini gives roughly four seats and, across all twenty of its answers, only ever seated 19 different companies. For a plumber in this area, that is a concrete number: on Gemini there are about four chairs, and fewer than twenty businesses have ever sat in one.
Mentions per question · every company above 5%
Visibility depends on the question
A company can dominate one kind of request and vanish from another. The emergency question matters most commercially, because a flooding basement is the highest-value call a plumber takes.
| Company | Good plumber | Burst pipe | Water heater | Tankless | Small job | Total |
|---|---|---|---|---|---|---|
| Sedona Plumbing | 3 | 8 | 8 | 11 | 7 | 37 |
| Iovino Plumbing | 8 | 1 | 8 | 8 | 8 | 33 |
| Kirkland and Shaw | 4 | 4 | 9 | 10 | 5 | 32 |
| Doug Anderson Plumbing | 8 | 5 | 5 | 4 | 4 | 26 |
| Woody's Plumbing & Heating | 4 | — | 7 | 3 | 4 | 18 |
| Ciardi Plumbing & Heating | 3 | 7 | 3 | — | 4 | 17 |
| Chevron Plumbing | 8 | 4 | 3 | — | 1 | 16 |
| Young Buck Plumbing & Heating | 2 | 8 | 2 | 1 | 3 | 16 |
| Chase Plumbing Co. | 4 | — | 3 | 4 | 3 | 14 |
| North Point Plumbing | 3 | 4 | 5 | 2 | — | 14 |
| Ostroski Plumbing & Heating | 3 | — | 3 | 6 | — | 12 |
| Perricone Home Services | 2 | 4 | 1 | 4 | — | 11 |
| 978 Plumbing | 1 | 1 | 3 | 3 | 2 | 10 |
| J. Griffin Heating & Plumbing | 2 | — | 3 | 4 | 1 | 10 |
| Roto-Rooter | — | 5 | — | — | 3 | 8 |
| Xstream Plumbing | — | 3 | — | — | 5 | 8 |
| John's Sewer & Drains | — | — | — | — | 7 | 7 |
| Morris Plumbing | — | — | 1 | 4 | 2 | 7 |
| Drain & Heating Solutions | — | — | — | — | 6 | 6 |
| MillTown Plumbing | 1 | 3 | — | 2 | — | 6 |
| Pespisa Company | 1 | — | 3 | — | 2 | 6 |
| 128 Plumbing | — | 1 | 1 | 3 | — | 5 |
| Floros Plumbing & Heating | 3 | — | 2 | — | — | 5 |
| Rodenhiser Home Services | 1 | — | 3 | — | 1 | 5 |
| Adams Plumbing | 2 | — | 1 | — | 1 | 4 |
| Boston Tank Swap | — | — | — | 4 | — | 4 |
| Emerson Mechanical Contractor | — | 4 | — | — | — | 4 |
| Jeneral Plumbing | 1 | 2 | 1 | — | — | 4 |
| Mystic Valley Home Services | — | 4 | — | — | — | 4 |
| Precision Home Services | 1 | — | 3 | — | — | 4 |
Every company named in 5% or more of answers. Each number is out of 16 answers per question; a dash means the company was never named for that question. Totals are out of 80.
Two patterns show up. Established companies with strong overall numbers still have holes: Woody's was named in seven answers about water heaters and none about burst pipes. Chevron was named eight times for the general question and never for tankless installation. Those are gaps in an otherwise strong position.
Then there are companies that exist for one question and nothing else. Boston Tank Swap was named four times, every one of them about tankless installation. Drain & Heating Solutions was named six times, all for small jobs. Whether by design or accident, these companies have become the answer to one narrow question, which is a real position to hold and a fragile one to depend on.
Some questions have far fewer contenders
Across 16 answers each, the tankless water heater question named 20 distinct companies. The water heater replacement question named 29. A narrower field is easier to enter, which makes specialty questions the practical place for most companies to start.
Run-to-run variation · identical wording, fresh session
The same question, asked twice
Every question was asked four times on each engine. Same words, no memory, no history, no personalization. If these systems returned a stable ranking, all four runs would return the same companies. They did not.
42%
Of every instance where a company was named, 42% happened in only one of the four identical runs. Just 27% of appearances held across all four.
Each company named within a four-run block, counted by how many of those four runs included it. Two identical runs overlapped by 50% on average.
This changes what the rest of this report means. A company at 20% share of voice does not hold a fixed position that can be checked once and confirmed. It enters a drawing each time somebody asks, and wins about one in five. The companies at the top of this study are not there because they are locked in. They are there because they are in the drawing more often, and 27 companies are not in it at all.
It also means a single spot check tells a business owner very little. Asking an assistant once and seeing your own name is roughly as informative as one coin flip.
This cuts both ways. The variability shown here is real and outside any one business's control, but it is not pure noise: a company with strong, clear, well-structured information about itself online, the practical target of AEO work, is entered in more of these drawings and wins more of them, which is why the same handful of names keep turning up across runs while 27 companies never appear at all.
Some engines and questions are steadier than others
| Engine | Agreement between two identical runs | Named in all four | Named in only one |
|---|---|---|---|
| Claude | 55% | 31% | 31% |
| ChatGPT | 53% | 32% | 38% |
| Gemini | 52% | 23% | 46% |
| Perplexity | 40% | 21% | 57% |
Perplexity was the least repeatable engine in this study: 57% of the companies it named appeared in only one of four identical runs.
| Question | Agreement between two identical runs |
|---|---|
| Burst pipe, emergency | 64% |
| Tankless water heater | 58% |
| Small job, affordable | 48% |
| Good plumber, general | 41% |
| Water heater quotes | 40% |
The emergency question produced the most repeatable answers and the general question among the least. Narrower questions appear to have a smaller and more settled pool of candidates.
Interpretation
What we think this means
A homeowner with water coming through the ceiling is not going to compare ten search results. Increasingly they ask an assistant and call the first name it gives them. In this study, that name was one of about six companies, and 21 well-ranked local plumbers were not among them.
None of this means Google no longer matters. Maps rank was the single strongest predictor of AI visibility we measured. It means the reward curve has changed shape. On Google, ranking 22nd gets you some business. In an AI answer, ranking 22nd gets you nothing at all, because there is no 22nd position.
What this practice is called
Getting a business named in AI-generated answers is usually called answer engine optimization, or AEO. Some people call it generative engine optimization, or simply AI visibility. They mean roughly the same thing.
It overlaps heavily with search engine optimization, and this study found Google Maps rank to be the strongest single predictor we measured. But it is not the same discipline, because the output is different. A search engine returns a page of ranked links and a person scrolls. An answer engine returns four or five names in a sentence and the person stops reading. Optimizing for a list you can scroll and optimizing for a list you cannot are different problems.
Correlation, not proof
This study records what four assistants said on one day in one town. It shows which companies were named and which were not. It does not establish why, and we are not going to claim otherwise. The recommendations below are the well-supported starting points, not a guaranteed route to being named.
Where a local company would start
Free, and fully in your control
- A page on your own site for each town you serve, saying plainly that you serve it.
- Content that answers the questions people actually ask, in their words. A page about emergency plumbing in Bedford, not a page titled "Services."
- A complete Google Business Profile with service areas listed, since Maps rank predicted AI visibility better than anything else we measured.
- The same business name, address, and phone number everywhere they appear.
Cheap, and outside your site
- Claim your Better Business Bureau listing.
- Claim free profiles on the directories that assistants cite when answering local questions.
- Get listed in your local chamber of commerce directory.
Paid, and worth thinking hard about
- Paid directory placements such as Angi carry a real monthly cost and hand the customer relationship to a third party. Some companies named in this study use them. We list this as an observed pattern, not a recommendation.
What we think generalizes, and what we don't know
Does this apply to businesses that aren't plumbers?
We measured one trade in one town on one day. Anyone telling you what that proves about dentists in Newton or law firms in Boston is guessing. Still, some of what we found looks structural rather than specific to plumbing, and it is worth separating the two.
Likely to hold anywhere
- There are only a handful of slots. Answers named 4.9 companies on average. That is a property of how assistants write answers, not of plumbing. Whatever your industry, the list is short.
- The engines disagree with each other. One company out of 82 was named by all four. There is no reason to expect four independently built systems to converge more in one industry than another.
- The same question produces different answers. Identical prompts agreed about half the time. This comes from how these models generate text, not from anything about pipes.
- Being findable is not the same as being named. Twenty-one companies ranked well enough on Google for a customer to find them and were never mentioned once. Any industry with more businesses than answer slots will have a version of this gap.
Likely to differ
- Which sources matter. Home services lean on general directories. Lawyers, doctors, contractors and restaurants each have their own dominant directories and licensing bodies, and those are probably what the assistants lean on instead.
- How much geography counts. Plumbing is bounded by drive time. For a business whose customers can be anywhere, the location signals that mattered here may matter much less.
- Whether people ask at all. A flooding basement at 11pm produces exactly the kind of question that gets typed into an assistant. Purchases that run on referral, relationship or repeat business generate fewer of them. Some industries have far less exposure to this than plumbing does, and should spend accordingly.
- How concentrated the results are. Three companies took 26% of all mentions here. In a denser market, or a thinner one, that could look very different.
What we genuinely do not know
Whether the invisibility rate in other industries is anywhere near half. Whether the same fixes move the needle. Whether any of this holds three months from now. We intend to run this study on other trades and other towns, and we will publish those the same way, including the parts that do not support our business.
If you want to know what these systems say about your own business rather than about plumbers, the test is the same one described in the method above and you can run it yourself. We also do it for people, across their real service area and the questions their customers actually ask.
Questions and Method
FAQ
Which plumbers do AI assistants recommend in Bedford, MA?
Across 80 sessions in August 2026, the most frequently named companies were Sedona Plumbing and Heating (46% of answers), Iovino Plumbing, Heating, and Cooling (41%), Kirkland and Shaw (40%) and Doug Anderson Plumbing and Heating (33%). Answers typically named three to six companies.
How many local plumbers are invisible to AI assistants?
Of 39 plumbers drawn from the Google Maps top 40 for Bedford, Massachusetts, 21 were never named in any of 80 recorded AI answers.
Do ChatGPT, Gemini, Claude and Perplexity recommend the same businesses?
No. Of 82 companies tracked, only one was named by all four engines. Overlap between any two engines ranged from 8 to 31 percent, so a business can be prominent on one assistant and absent from another.
Does Google Maps ranking affect whether AI assistants recommend a business?
In this study the two correlated at r = -0.60. Companies ranked 1 to 10 on Google Maps received 127 mentions; companies ranked 31 to 40 received two. The relationship is strong, but the drop-off is much steeper than in Google search results.
Do AI assistants give the same answer every time?
No. In an August 2026 study of 80 sessions, each question was asked four times per engine with identical wording in a fresh session. Any two identical runs shared only about half their recommendations. Of every instance where a company was named, 42% occurred in only one of four runs and 27% held across all four. Perplexity was the least repeatable at 40% agreement between runs; Claude the steadiest at 55%.
Which plumber is named first most often by AI assistants in Bedford, MA?
Iovino Plumbing, Heating, and Cooling led more answers than any other company, appearing first in 31 of 80 answers. It was named in 33 answers overall. Only 15 of the 82 companies tracked were ever named first, and three companies took the first slot in 58 of the 80 answers.
How many businesses does an AI assistant recommend at once?
Across 80 answers in this study the average was 4.9 companies. Claude named the most at 6.0 per answer, ChatGPT 5.8, Perplexity 4.0 and Gemini 3.8. Gemini named only 19 different companies across all twenty of its answers.
How was this study run, and what are its limitations?
Companies. We pulled the top 40 Google Maps results for plumbers serving Bedford, Massachusetts, and kept those offering residential plumbing to the town. We added companies recommended in local community groups, and any company an assistant named that was not already on the list. Final universe: 82 companies. Where a comparison needs a clean sample, we use only the 39 drawn from Google Maps, because the remainder were discovered through the assistants themselves and cannot be used to measure what the assistants do.
Questions. Five, phrased the way a homeowner would type them, each ending with the location. They covered finding a plumber with no existing relationship, an emergency burst pipe, water heater replacement quotes, a tankless installation, and a small affordable job.
Sessions. Each question was asked four times on each of ChatGPT, Gemini, Claude and Perplexity. 5 questions × 4 engines × 4 runs = 80 sessions, all on 28 August 2026. Every session was run in a fresh temporary or incognito chat with memory and history disabled, from the same location, so that personalization did not affect results. Running each question four times was intended to capture variation rather than a single snapshot. It turned out to be the minimum useful number: two identical runs agreed with each other only about half the time, so a single run per question would have produced a materially different study.
Recording. For each session we recorded every company named and the order in which it appeared. No count was requested; the assistants decided how many to name, returning between 3 and 10.
Limitations. One town, one trade, one day: these systems change week to week, and results elsewhere may differ. Four runs per question per engine captures some variation, but not all of it. Given that identical runs agreed only about half the time, share-of-voice figures here should be read as approximate rather than precise; a company shown at 20% might land anywhere in a band around that number on a different day. We recorded which companies were named, not the reasons given, and did not systematically capture the sources the assistants cited, so this study makes no claim about which directories or pages drive the results. Several companies named by the assistants have pages on their own sites targeting Bedford specifically; we noticed this, but because we found those companies through the assistants in the first place, we cannot treat it as evidence — testing it properly means checking the never-named companies too, which is planned for the next round. Finally, we are an AI visibility consultancy: this study is research we ran to understand our own market, and it is also marketing. The method above is written so you can run it yourself and check us.
One last thing worth saying out loud: this page is built the way we would build a client's. A specific question answered directly, findings described in structured data, plain language a machine can quote back. Whether that works is checkable — ask an assistant about AI visibility for local businesses in a few months and see whether this study comes up.
What were the exact questions used in the study?
- "I need to find a good plumber and I don't have anyone I trust. Who are my best options? I live in Bedford, MA."
- "A pipe burst in my basement and it's flooding. Which plumbers should I call right now? I'm in Bedford, MA."
- "My water heater is failing and I want to get a few quotes before I commit. Which plumbers should I contact? I live in Bedford, MA."
- "I'm thinking about switching to a tankless water heater. Which plumbers do that well? I'm in Bedford, MA."
- "Which plumbers should I call for small jobs like a leaky faucet or clogged drain? Looking for reliable, affordable options. I live in Bedford, MA."
What is answer engine optimization (AEO)?
Answer engine optimization is the practice of getting a business named in the answers AI assistants give, rather than in a list of search results. It overlaps with SEO and depends on many of the same signals, but the target is different: a search engine returns a scrollable page of links, while an assistant names four or five businesses and stops. In an August 2026 study of 80 AI answers about local plumbers, answers named 4.9 companies on average, so there is no equivalent of page two.
How do I find out whether AI assistants recommend my business?
Open a fresh temporary or incognito chat with memory and history turned off, phrase the question the way a customer would including your town, and repeat it several times on each assistant. Repetition matters: in a study of 80 sessions, 42% of the times a company was named it appeared in only one of four identical runs, so asking once tells you very little.
Does ranking well on Google mean AI assistants will recommend my business?
It helps but is not sufficient. In an August 2026 study of plumbers serving Bedford, Massachusetts, Google Maps rank was the strongest measured predictor of AI visibility at r = -0.60. Even so, 21 of the 39 businesses drawn from the Google Maps top 40 were never named in any of 80 AI answers, including companies ranked in the top twenty.
Does AI visibility matter for industries other than home services?
Some findings appear structural rather than industry-specific: AI answers name only a handful of businesses, different engines recommend different companies, and identical questions produce different results. What differs by industry is which sources assistants draw on, how much geography matters, and how often customers ask an assistant rather than relying on referral or repeat business.
Was your company one of the 21?
We are not publishing the names of companies that were never mentioned. If you run a plumbing business in the Bedford area and want to know where you stood in these 80 answers, we will send you your own results and what the visible companies were doing differently. No charge, no obligation.
Not a plumber? The same test runs for any business customers look for by need and location. We will tell you which assistants name you, which name a competitor instead, and where the gap is.
Ask what AI says about your business