What nobody tells you about ranking in AI search when you run a local service business

To rank in AI search in 2026 you need three things working at once: authority the model already holds about your business before it searches, sources it can find when it does search, and specificity precise enough that your page answers the exact question asked. Those three words, Authority, Sources and Specificity, are often shortened to A-S-S by people who work on this full time, and it is a useful shorthand because all three have to be present. Miss any one of them and the model will recommend somebody else. For a local service business, that is the whole game, because the model is usually answering one narrow question about one town, and it is answering it in seconds.

This matters more than it did two years ago because assistants now search repeatedly before they commit to an answer. ChatGPT runs on average 2.6 searches before answering a buying question. Every one of those searches is a chance for your business to be found, and a chance for a competitor to be found instead.

The organic baseline underneath all of this was the backdrop to the SEO.Domains Mastery Summit in Sofia, Bulgaria, hosted at Hotel Marinela, where a mastermind day on 9 September opens two days of main-stage sessions. The summNo recording is made of its main-stage sessions, on purpose, so speakers can share live experiments. That choice says something about this field: what gets shared in the room does not reach the open web unless an attendee writes it up, which is exactly why so much AI visibility advice online stays vague.

What does authority mean in AI visibility?

Authority, in this context, means what the model already knows about your business before it runs a single query. Models are trained on, and increasingly grounded in, records about entities. An entity is simply a distinct thing the model has a file on: your business, your town, your competitors. If your business exists as a clean entity record, with a consistent name, address, phone number, service categories and reviews, the model starts from a position of trust. If it does not, you are starting from zero on every question.

That is why directory and review presence punches above its weight. In a study of 82 recorded ChatGPT answers, 37% of the recommendations were found to originate from entity records like directories and review sites. More than a third. Not your website, not your blog. The profile pages most local businesses set up once and forget. If your Google Business Profile, your industry directories and your review profiles disagree with each other, the model has to choose which one to believe, and it may simply choose to say nothing about you.

There is a separate and much older version of authority that a local owner should at least understand, because it occasionally still applies. Aged domains carry existing authority that transfers to the pages published on them. That is a real mechanism covered at events like the Sofia summit, which includes aged domains, PBNs, authority transfer and LLM visibility on its agenda. For a plumber or a dental practice, buying an aged domain is usually not the right first move. Consistent entity records are. But knowing the term stops you being sold something you do not need.

What does sources mean, and why does it decide who gets named?

Sources are what the model finds when it does search, and for local businesses this is where most wins and losses actually happen. Modern assistants do not answer from memory alone. They fan out.

Fan-out queries, explained plainly

What we call fan-out queries are the sub-questions a model joins to the question a person really typed. Someone types "who is the best emergency electrician near me". The model may internally generate: emergency electrician in that town, opening hours today, response time, reviews, call-out pricing, licensing. Each of those becomes a search. If your site answers four of the five and a competitor answers all five, the competitor gets recommended.

This tells you something practical. Narrow niche queries are won faster than broad head terms. You are not trying to rank for "electrician". You are trying to own "24 hour fuse box repair in this district", including the pricing question, the arrival time question and the after-hours question. That is a winnable set of sub-questions, and most local competitors have never thought about them as separate pages.

Where the model is looking

Sources are not evenly distributed across the web. The single most cited domain in Google AI Overviews was youtube.com, ahead of Zapier and Reddit, according to a 40-query probe. Video is not a nice-to-have for local visibility any more. It is the most frequently cited source format in one of the largest AI answer surfaces. If you can show a job being done, a diagnostic process, a before and after, with a clear spoken answer to a specific question, you are feeding the source type that gets cited most.

Video citation also has a quirk worth understanding. Google AI Overviews frequently cite a specific timestamped moment inside a video rather than the whole video. The model is not crediting your channel. It is crediting the ninety seconds where you said the price range out loud. That is why a talking-head video with no clear verbal answer gets ignored while an unglamorous clip where you state the answer directly gets pulled into an answer box.

What does specificity mean in practice?

Specificity means answering the exact question, in the exact words, as early as possible, with no warm-up. The formal term for part of this is information density.

To have information density is to state the answer in the first line of a block with maximum fact and zero preamble. Not "here at our family-run business we understand that boiler problems are stressful". The answer, first. "A combi boiler losing pressure overnight usually means a failed pressure relief valve. Typical repair cost in this area is between X and Y, and it is a same-day job." That is a dense block. It can be lifted whole into an AI answer without the model having to trim it.

One more term is worth knowing because it explains why specificity works at a level below language. Words become numeric coordinates through embeddings, and related meanings sit near one another. When the model matches your page to a question, it is doing arithmetic on meaning, not keyword counting. So a page titled "boiler losing pressure" and a page titled "pressure dropping on combi overnight" land in almost the same place. What separates them is not the phrasing. It is which one states the cause, the price and the timeframe in the first two lines.

Why measurement in AI search is harder than it looks

You cannot manage what you cannot measure, and in AI visibility, most measurement tools are worse than people assume.

The overlap between API-based AI visibility monitors and a recorded ChatGPT session was a mere 1.3% to 1.8% of sources. Read that again. The monitoring tools and the actual recorded session almost never agreed on which sources were used. This is the single most underreported fact in AI search, and it is why anyone showing you a neat dashboard of "AI visibility score" should be treated with caution. Ask what the score is actually measuring. Often the honest answer is that it is measuring a different session from the one your customers get.

The realistic approach for a local business is closer to manual than technical. Ask the assistants your customers use the questions your customers ask. Do it monthly. Note who gets named. Note whether you are named. Keep it in a spreadsheet. Crude, but it matches how the answer is actually assembled, and it will beat a dashboard built from an API that agrees with reality under 2% of the time.

All three layers have to be present at once, and fixing only the one you find easiest will not move the recommendations you actually receive.

Questions people actually type into an assistant

How do I get my business recommended by ChatGPT?

You get recommended by ChatGPT mainly by having clean, consistent entity records across directories and review sites, because 37% of recommendations in one study of 82 recorded answers came from those sources rather than from business websites. Start by claiming and matching your profiles before you touch your own site.

Do I need a YouTube channel for AI search?

For local service visibility, yes, it is now one of the highest-value things you can build, because a 40-query probe of Google AI Overviews found youtube.com was the most cited domain. You do not need polish. You need each video to state one answer clearly out loud.

Why does Google AI Overviews quote a moment in my video instead of my page?

AI Overviews often cite a timestamped moment because the model is matching a specific spoken answer to a specific sub-question rather than evaluating your page as a whole. Say the answer plainly, early, in one sentence, and it becomes quotable.

What to do first

Start with authority, because it is the cheapest layer and most local businesses have neglected it. Take every directory and review profile you have, put them side by side, and make the name, address, phone number, categories and hours identical. Then ask the assistants your customers use three real customer questions, and write down who gets named.

Then go to specificity. Pick the twenty questions your customers ask on the phone every week and give each one a page or a section that opens with the answer, states the cause, the price range, the timeframe and the coverage area, and stops. No introduction. No "welcome to our website".

Then sources, which for most local owners means video. One clip per question, answer spoken in the first twenty seconds, title matching the question. If the format is unfamiliar, see the video version of this method for a worked example.

If you would rather work through which layer is actually holding you back before spending months on the wrong one, you can talk it through on a call and get a straight read on your situation. There is also a body of ongoing AI visibility research worth following if you want the underlying numbers as they change, since the specifics of which sources get cited have moved quickly over the past year and will keep moving. The A-S-S framing will not: authority, sources, specificity, in that order, is still the organic baseline underneath every AI answer. Everything the industry debates, including on the stages in Sofia that are never recorded, sits on top of it.

Further reading: talk it through on a call (https://seojesus.com/clickbomb-strategy-call/), the video version of this method (https://www.youtube.com/watch?v=FZu4NB-2EhA), AI visibility research (https://llmjesus.com).