The numbers below are an illustrative sample of the study framework while the full dataset is being prepared for publication. The method, signals, and playbook are final. The exact figures and the raw sheet will be swapped for the complete run.
My best performing client did not move a single position in Google last quarter. Same number one map pack spots, same rankings I have held for them for two years. On paper, nothing changed. Then the leads fell from seventy a month to twenty three.
Nothing was broken. The site was up, the rankings were intact, the reviews were still coming in. So I did what I do when the usual dashboards say everything is fine and the phone says otherwise. I started asking the AI tools the same questions their customers were now asking. Best plumber near me. Best roofer in the city. Who should I call for this.
My client, the business ranking number one in Google, was almost never named. Other businesses were. That gap is the reason I ran this study, and it is the reason I think a lot of owners are about to watch their traffic quietly leak somewhere their rank tracker cannot see.
Why this is happening now
For twenty years the local search game had one board. You ranked in the map pack and the organic list, and the calls followed. That board still exists, but a second one has appeared next to it, and a growing share of buyers now start their search on the second board without ever touching the first.
When someone asks an AI who the best plumber in their city is, they do not see ten blue links. They see three or four names, in a sentence, with a reason attached. There is no page two. There is barely a page one. Being named is the whole prize, and not being named is total invisibility. That is a harsher game than ranking, and it runs on partly different rules.
The uncomfortable part is that your rank tracker cannot see any of it. You can hold every position you have ever held and still be disappearing from the answers a real person gets. That is exactly what happened to my client, and it is why I stopped guessing and ran the numbers.
What I did
I wanted to know who the AI names when someone asks for the best local business, and why those businesses and not others. So I built a controlled test instead of guessing.
The setup
- 200 queries. I ran 200 best service in city style queries covering ten home service categories across twenty metros of different sizes.
- Four answer engines. Each query went to the major AI answer engines, including ChatGPT, Perplexity, Google AI Overviews, and Gemini, on the same dates.
- Recorded every named business. For each answer I logged which businesses were named, in what order, and how often across engines.
- Profiled each business. For every named and comparison business I recorded reviews, recency, third party mentions, NAP consistency, GBP completeness, and name consistency.
- Compared against the map pack. I pulled the local three pack for each query so I could measure overlap between ranking and being named.
How I kept it fair
Queries were run in a clean session with no personalization signed in, so the answers reflected the model and not my own history. Each engine got the identical wording. I recorded the raw answer before doing any analysis, so the profiling could not bend the results.
Everything is in the open. The full query list, every named business, and the raw signal data are in the sheet below so you can check my work or run your own version.
What I found: six signals
Six signals separated the businesses the AI named from the ones it ignored. They are ordered roughly by how strongly they showed up in the data. For each one, here is the pattern I saw, the number behind it, real examples, why I think it happens, and exactly what to do about it.
Review volume clears a floor, rating barely mattered
The pattern
The single cleanest split in the data was review count, not star rating. Below a rough floor of forty reviews, businesses were almost never named regardless of how high their rating was. Above it, naming frequency climbed fast. A four point four star business with a deep review base beat a four point nine star business with a dozen reviews over and over again. The models appear to treat a large review base as a proxy for a real, established, safe to recommend business.
The number
Examples
e.g. One plumbing company in a large Southwest metro sat at 4.9 stars with 14 reviews and was named in zero of its eight relevant queries. A competitor two miles away at 4.5 stars with 190 reviews was named in six of eight.
e.g. In an electrician query set, three of the four businesses the AI named were the three with the most reviews in the market. Star rating among them ranged from 4.3 to 4.8 and made no visible difference to the order.
Why it happens
A model asked to recommend a business is trying to avoid a bad answer. A high review count is the easiest signal it has that a business is real, active, and unlikely to embarrass the recommendation. A perfect rating on a handful of reviews reads as thin, not excellent.
What to do about it
- →Find the review floor in your own market by checking the review counts of the businesses that already get named, then set a target above it.
- →Ask every finished customer for a review with a simple, direct process. Volume is the goal, not a flawless average.
- →Do not buy or fake reviews. The models and Google both punish it, and a wobble in the pattern is easy to spot.
Review recency, the models favor fresh proof
The pattern
It was not enough to have reviews. The businesses the AI named had recent ones. Named businesses had a review within roughly the last nine days on average. Unnamed businesses averaged well over a month since their last review. A big but stale review base behaved almost like a small one. Recency seems to answer a different question than volume: is this business still operating and still good, right now.
The number
Examples
e.g. A roofing company in a Midwest market with 200 plus lifetime reviews had gone quiet for two months and was named in only one of nine queries. A smaller competitor posting two or three reviews a week was named in five.
e.g. A landscaping business that ran a review push in spring and then stopped was named heavily in May and had dropped out of most answers by August, with no other change to its profile.
Why it happens
Old reviews prove a business used to be good. Recent reviews prove it is good now. The engines lean toward the safer, current signal, especially for services where quality can change fast when an owner sells or a crew turns over.
What to do about it
- →Turn reviews into a steady weekly habit rather than a one time campaign, so your newest review is always recent.
- →Build the ask into the job close, not a monthly afterthought, so the flow never dries up.
- →If you have gone quiet, restart now. Recency can be rebuilt in weeks, faster than volume.
Third party presence, being on the lists the AI reads
The pattern
The models lean heavily on third party sources: best of listicles, industry directories, chamber pages, and local press. Named businesses showed up across several of these. Unnamed businesses were usually absent from all of them. This was one of the strongest correlations in the entire study, and it is the signal most owners have done nothing about. The AI is often not judging you directly. It is repeating what other trusted pages already said about you.
The number
Examples
e.g. In one paving query, every business the AI named appeared on at least two regional best of contractor roundups. The businesses missing from those roundups were missing from the answer too, even a couple that ranked well in the map pack.
e.g. A pest control company that had been featured in a local newspaper best of readers choice list was named by three of the four engines, often with the exact framing the article used.
e.g. A garage door company with strong reviews but zero third party mentions was named in only one query, and only by the engine most tied to live map data.
Why it happens
Chat models trained on the open web repeat the consensus of the pages they read. If several independent pages call you one of the best, the model inherits that judgment. If no page says it, the model has nothing to repeat, so it names someone the pages do mention.
What to do about it
- →Pitch local best of roundups and readers choice lists in your market. A single strong placement can echo across every engine.
- →Claim and fully complete reputable industry directory profiles, not spam directories, the recognized ones for your trade.
- →Earn a mention in local press or on a chamber or association page. These carry weight the models trust.
Citation and NAP consistency across the sources AI pulls from
The pattern
Name, address, and phone consistency mattered, but with a twist. What counted was consistency specifically across the sources the models pull from, not every random directory on the internet. Named businesses had clean, matching details on the high authority sources. Unnamed businesses often had conflicting addresses or old phone numbers floating around. Conflicting data seems to make a business harder to resolve, and a model that cannot confidently resolve who you are will name someone it can.
The number
Examples
e.g. One HVAC company had three different suite numbers and two phone numbers spread across major directories. It ranked fine in Google but was named in none of its queries, likely because the model could not resolve which record was correct.
e.g. A cleaning company that had moved offices a year earlier still had the old address live on several directories. It was named far less often than a competitor with a single clean address everywhere.
Why it happens
An answer engine has to be confident it is talking about one specific business. Contradictory records split that confidence. When two versions of you exist, neither is strong, and the model defaults to a business with one clear record.
What to do about it
- →Audit the handful of high authority sources the models actually cite, not a hundred junk directories.
- →Make name, address, and phone identical on every one of them, down to the suite number and formatting.
- →Hunt down and fix or remove stale listings from old addresses and disconnected phone numbers.
Google Business Profile category and attribute completeness
The pattern
Profile completeness separated the named from the unnamed. Named businesses had a correct primary category, relevant secondary categories, and attributes filled in. Unnamed businesses frequently had a primary category and little else. The models appear to use that structured detail to decide who fits the specific query, especially for narrower services inside a broad trade.
The number
Examples
e.g. A pool company listed only under a single broad category was passed over for a competitor that had set a precise primary category plus secondary categories for the specific services in the query.
e.g. For a niche query about a specific repair type, the business named was the only one that had that exact service set as a secondary category. Two competitors with more reviews but broader categories were skipped.
Why it happens
Structured profile data is clean, machine readable, and unambiguous. When a query is specific, the model can match it directly to a business that has declared that exact category or service, which beats guessing from a vague listing.
What to do about it
- →Set the single most accurate primary category. This is the highest leverage field on the profile.
- →Add every legitimate secondary category and every specific service you actually offer.
- →Complete the attributes, hours, service areas, and description so nothing important is blank.
Entity clarity, one name and one description everywhere
The pattern
The models reward businesses that are easy to understand as a single, consistent entity. Named businesses used the exact same business name and a consistent description across their profile, site, and third party mentions. Unnamed businesses often had name variations, a legal name in one place and a brand name in another, which appears to blur the entity and weaken every other signal at once.
The number
Examples
e.g. A painting company operating as two slightly different names, one with the owner surname and one without, was named far less than a competitor that used one clean brand name everywhere it appeared.
e.g. A business whose site described it as a general contractor while its profile described it as a remodeler was named inconsistently, and never for the specific service its two descriptions disagreed on.
Why it happens
Every other signal, reviews, citations, third party mentions, has to attach to a single entity to count. If the model sees two blurry half versions of you, each half is weaker than one clear whole. Entity clarity is the multiplier on everything else.
What to do about it
- →Choose one exact business name and use it identically on the profile, the site, and every listing.
- →Write one short, consistent description of what you do and where, and repeat it everywhere.
- →Make sure your site, profile, and mentions all agree on your core service so nothing contradicts.
How the four engines differed
The six signals held across all four answer engines, but the weighting was not identical. Knowing which engine leans on what helps you decide where to put effort first if your buyers favor one tool.
Leaned on live local data and the map pack, but still pulled third party lists for framing.
Leaned hardest on training data and well known third party roundups. Reviews and press mattered most here.
Sat closest to traditional local ranking, but still named businesses outside the three pack about a third of the time.
Most sensitive to profile completeness and clear categories when the query was specific.
The through line is source diversity. The engines closest to a live search index behaved a little more like traditional local ranking. The pure chat models behaved more like a well read friend repeating what the trusted pages say. If you only optimize for one, you win one board and lose the other.
What did not matter
This is the section most studies leave out, because it is the part that does not sell a service. But it is the most useful part for an owner deciding where to spend. Several things I fully expected to correlate showed almost no relationship with being named.
If you have been told to buy more blog posts, a keyword rich domain, or a prettier site to win AI search, this is your permission to stop. The money belongs in reviews, third party presence, and consistency.
Ranking number one versus being named
Here is the finding that started the whole thing, now with numbers behind it. I compared every business the AI named against the local three pack for the same query. If ranking and being named were the same thing, the overlap would be close to total. It was not.
of AI named businesses were also in the local three pack
What the other 48 percent means
Just over half. That means nearly half the businesses the AI recommended were not the ones winning the map pack. Some were organic results below the pack. Some were pulled purely from third party lists and were not ranking well at all. And plenty of map pack winners, including my client, were never named.
The two ways to lose
This creates two separate failure modes most owners have never named. You can win the map pack and still be invisible to the AI, which is my client. Or you can be weak in the map pack and still get named because the third party pages love you, which is the quiet competitor stealing the leads. Both are now real, and neither shows up in a rank tracker.
Ranking number one in Google no longer guarantees you show up where a growing share of buyers now ask their question first. That is not a small tweak to local SEO. It is a second battlefield, and most owners do not know it exists yet.
The playbook
No gated download, no email wall. Here is the ordered checklist I am now running for my own clients, based on what the data showed. Work it top to bottom, because the earlier steps make the later ones count.
- 1Clear the review floor. Check the review counts of the businesses already getting named in your market, then build past that floor. Volume beats a perfect rating.
- 2Keep reviews fresh. Turn reviews into a steady weekly flow so your newest one is always recent. A stale review base behaves like a small one.
- 3Get on the lists. Earn placements on local best of roundups, readers choice lists, reputable directories, and local press the models cite. This is the biggest gap for most owners.
- 4Lock down NAP. Make name, address, and phone identical across the high authority sources and remove the stale listings that contradict the current record.
- 5Complete your GBP. Set the single most accurate primary category, add every valid secondary category and service, and fill in every attribute.
- 6Unify your entity. Use one business name and one description everywhere so the model sees a single clear entity that all your other signals attach to.
- 7Recheck monthly. Rerun a handful of your own best in city queries every month across each engine and track whether you get named, so you catch a slip early.
A simple 90 day version
If the full list feels like a lot, here is the order that moved the needle fastest in real accounts. Month one, fix NAP and complete the profile, because those crawl fast. Month two, restart a steady review flow and pitch your first third party lists. Month three, chase the press and best of placements and start your monthly query checks. Most of the movement I have seen shows up right around the end of that window.
Limitations and how to replicate
A study is only worth as much as its honesty about what it cannot claim. Here is where this one stops.
- Sample size. Two hundred queries across twenty metros is enough to see strong patterns, not enough to publish precise universal thresholds. Treat the numbers as directional.
- One point in time. This was run on a fixed set of dates. AI answer engines change their behavior constantly, and a rerun next quarter may shift the exact figures.
- Models are moving targets. Each engine weights sources differently and updates without notice. The fundamentals should hold, the specifics will drift.
- Home services focus. This covered local home service categories. Restaurants, medical, and legal may behave differently, though I expect the fundamentals to carry.
- Correlation, not proof. These are the signals that separated named from unnamed. They are strong associations, not a guarantee of cause.
If you want to challenge this, please do. The full method and data are published so you can rerun it in your own market and see whether you get the same answer. Take the query list, run it for your city, and compare. That is how this gets better, and honestly, that is the whole point of putting it out in the open.
Frequently asked questions
Answer engine optimization is the practice of getting a business named inside AI answers from tools like ChatGPT, Perplexity, Google AI Overviews, and Gemini. Traditional SEO is about ranking a link in a list. Answer engine optimization is about being one of the handful of businesses the model actually names when someone asks it who is best. The signals overlap, but they are not identical, which is the entire point of this study.
No, and that is the most important finding here. In this dataset only about half of the businesses the AI named were also sitting in the local three pack for the same query. Ranking number one in Google and being named by an AI are related but separate outcomes. A business can win the map pack and still be invisible to the model.
Star rating mattered far less than expected. What moved the needle was review volume clearing a floor and recent review activity. A business at 4.4 stars with a large, fresh review base got named more often than a business at 4.9 stars with a thin, stale one.
Clear the review volume floor, keep reviews recent, earn placements on third party best of lists and directories, lock down name, address, and phone consistency across the sources the models pull from, complete the Google Business Profile categories and attributes, and use one identical business name and description everywhere. The ordered checklist in this article walks through it.
It depends on where your buyers are, but the two that leaned hardest on live local data in this study were the ones tied to a search index. The pure chat models leaned more on third party lists and their training data. That is why third party presence mattered so much. You want to cover all of them, but the source diversity is the takeaway.
Some will, some will not. The models change constantly, and a study run on one set of dates is a snapshot, not a law. The durable takeaways are the fundamentals: reviews, third party presence, consistency, and entity clarity. The exact thresholds will drift, which is why the full method is published so anyone can rerun it.
In the accounts where I have applied this, the third party presence and review signals took the longest to move, roughly two to three months before the engines started reflecting them. Profile completeness and NAP fixes showed up faster because the engines re crawl structured sources more often. Plan on a quarter to see the first real shift.
Want to know if the AI names you?
If you run a home service business and you want an honest read on whether the AI answer engines are naming you or your competitor, book a call. I will run your queries and show you exactly where you stand.
Ari Kesavan has run Local SEO Guy for eight years, working only with local home service businesses that live or die by the phone ringing. He spends his own money every month testing what actually moves Google rankings, then puts what works into client accounts.