Specificity Beats Authority: Winning the Narrow AI Queries Big Competitors Can't

The short version: for "plumber in Toronto" or "dentist near me in Etobicoke," ChatGPT, Gemini and Google's AI Overview name a business and link its home page, and the biggest, most reviewed operator usually takes that link. But across the local-service sites we monitor there is one repeatable case where an interior page beat every front page in the category: the question was narrow enough that no competitor's generic service page answered it, and one focused page did. That is the whole opening for a smaller GTA trade, clinic or firm. You will not out-authority the incumbent on the head query. You can out-answer them on the specific ones they never bothered to write.

Where the homepage rule breaks

Our tracking says the front page wins local commercial intent. Ask any engine for a service in a city and the URL it links is the home page almost every time. We laid that out in why AI cites your homepage, not your blog.

One query broke it.

It was an equipment-specific duct cleaning request, a precise variant naming a piece of gear only some operators handle, rather than the broad "duct cleaning in the city" phrasing the front pages fight over. The large competitors returned their catch-all service pages, which say nothing exact about that equipment. A single focused page that did address it, start to finish, took the citation. No front page had anything exact to offer.

Read the rule the other way round. Authority is the tie-breaker when every candidate page says roughly the same thing. Remove the tie, by asking something the generic pages skate past, and the tie-breaker never fires. The only passage that answers the question belongs to whoever wrote it.

Why narrow wins: the engine has to retrieve something exact

The mechanics sit in how AI search reads your prompt. It does not run one query. It fans the prompt out into a set of related sub-searches and builds the answer from the union of what those return. Seer Interactive measured an average of 10.7 fan-out queries per prompt on Gemini 3, and found that 95 percent of them had no monthly search volume at all, at an average length of 6.7 words. We covered the method in our query fan-out guide.

Think about what that means for a Scarborough HVAC company. The sub-searches are long, conversational and specific. "Can a heat pump run a 1950s bungalow with original ductwork in Scarborough." "Does a furnace in a Toronto semi need a chimney liner after a high-efficiency swap." Nobody types those. No keyword tool lists them. And the incumbent's page, which reads "we install furnaces and heat pumps across the GTA," matches none of them with any precision.

So the engine retrieves nothing exact from the big brand. If you have published a clean page on exactly that, your passage is the best match, and your name goes into the answer. The incumbent's size is irrelevant to a sub-search it has no passage for.

Two caveats travel with this. First, specificity still needs geography. Across more than 470 un-geotagged informational checks in our tracking, the sites we monitor earned zero citations; those went to government pages and forums. A narrow query with "in Toronto," "in Mississauga" or a neighbourhood attached is winnable. The same query with no place attached mostly is not. Second, we have one unambiguous win we can point to by shape, plus smaller echoes of it across the sites we monitor. We will not dress that up as a percentage we do not have.

What "specific enough" looks like for a Toronto trade

Specific is not the same as long. "Everything You Need to Know About Plumbing in Toronto" is long and completely generic. The test is whether the question has a precise object that most competitors' pages do not name. Some query shapes that pass, across verticals we track or have field-tested in Toronto:

• Equipment or material variants. Not "furnace repair North York," but the make, system type or building material that changes the job. Knob-and-tube in a Riverdale Victorian. A tankless unit throwing a particular error code. Asbestos tile under a Leaside basement floor.
• Property-type constraints. The downtown condo where the board controls the HVAC. The Scarborough bungalow with a crawlspace. The Etobicoke semi sharing a wall. Almost no service page says how these change the job.
• Regulatory edge cases. "Is a permit needed to move a kitchen sink in Toronto." Naming the rule in the H1 and citing the primary source wins these, as we showed in our regulatory piece.
• Clinical or professional sub-questions. "Can a Toronto dentist bill a specific CDCP code for this procedure." The head query goes to the most reviewed clinic. The sub-question goes to whoever answered it.
• The should-I decisions with a local twist. "Should I replace or repair a 15-year-old furnace before selling a house in Mississauga." Reddit takes many of these, but a precise local page can still take the ones the threads only half answer.

The pattern underneath every item is the same. A real customer asked it on the phone. The incumbent never wrote it down.

The thirty-second test

Open ChatGPT or Gemini. Ask for your service in your city. Note who it names and which URL it links.

Now ask the narrowest real question a customer asked you this week, with the city in it. If the engine names nobody, or hedges, or links a government page, that question is open. Go write the page.

How to find your under-answered tail

The tail you want is specific, local and unanswered. Four sources, in the order we use them on audits.

Your own call log and inbox. The questions customers ask before booking are the fan-out sub-searches in their natural form. An HVAC dispatcher in Vaughan hears the same six odd questions every month. None of them is on the website.

Google autocomplete, run with the Canadian region. Type the service, then a constraint, then a place, and read the suggestions. Autocomplete is a documented proxy for the demand the fan-out draws on, and it is free.

The engines' own follow-up prompts. After an AI answer, the suggested next questions are the sub-searches it expects to run. Collect them for your category.

The competitor gap check. Open the top three competitors' pages for the head term. If none contains a passage answering the narrow question directly, the question is under-answered. If one does, skip it. You beat a generic page with any precise answer. You do not beat a precise answer with a slightly more precise one.

Rank what survives by how often a real customer asked it and how clearly a place can be attached.

Writing the page that takes the citation

The page has one job, which is to be the best retrievable passage for one specific sub-search.

• Put the specific thing in the H1 and the first sentence, with the place. "Chimney liner requirements after a high-efficiency furnace install in a Toronto semi" is a headline a sub-search can match. "Furnace Services" is not.
• Answer in the first paragraph. State the answer, then explain. A page that makes the engine read 600 words of preamble loses to one that does not.
• Cover the neighbours of the question on the same page. One thorough page on the equipment variant, the permit angle and the property-type wrinkle outscores four thin pages. Engines rarely cite two pages from the same domain, so splitting the topic competes with yourself.
• Name the entities plainly. The equipment, the neighbourhood, the regulation number, your company. A passage that says "this" and "it" matches nothing.
• Keep the technical floor. Content in the HTML rather than loaded by script, a fast response, no accidental crawler block.

Do not expect this page to take the head query. It will not. The front page keeps that job. The specific page earns the links the front page structurally cannot.

What our audits flag, and what we fix first

When we audit a GTA local-service site, the specificity gap has a recognisable shape: one home page, five or six catch-all service pages, and a blog of city-plus-service posts that duplicate them. Nothing answers a narrow question precisely, so the site competes on authority alone, which is the fight it loses.

The first fix is never more city-service pages. It is a list of five to ten real customer questions, each with a place attached, each checked against competitors' pages, each given one answer-first page. Then we track them per engine, because the engines disagree far more than they agree, and a page ChatGPT ignores may be exactly what Perplexity cites.

If you want to see whether the engines can even read your site before you write a word, the free AEO checker covers the technical floor. If you would rather we find the tail and write for it, that is part of our SEO and AEO service.

FAQ

Can a small Toronto business really get cited by AI over a bigger competitor? Yes, on narrow queries. For the head query, "service in city," the engines default to the most established operator's home page, and a smaller business rarely displaces it. For a specific, local question that no competitor's generic page answers precisely, a focused page from a small operator can take the citation, because the engine has to retrieve an exact passage and only one exists. We have watched that happen across the local-service sites we monitor.

How do I know if a query is specific enough to win? Ask it to ChatGPT or Gemini with your city attached. If the answer names a competitor and links their front page, it is a head term and the front page rule applies. If it names nobody, hedges, or links a government or forum page, the question is under-answered. Confirm no top competitor's page contains a direct answer, and the question is yours to take.

Does a narrow page still need to mention Toronto or my neighbourhood? It does. In our tracking, un-geotagged informational queries went to government and forum sources every time, with no local-service site cited across hundreds of checks. The question has to carry a place, in the headline and first paragraph, for a local business to be a candidate at all. Specificity opens the door. Geography lets you through it.

Should I write one page per narrow question or one big page? One thorough page per question cluster. AI search fans a prompt into many related sub-searches, so a page answering the main question and its close neighbours matches more of them than several thin pages would. Engines also rarely cite two pages from the same site in one answer, so fragments compete with each other. Cover the cluster, answer first, and let the front page keep the head term.

• $3,000 Growth CreditFree assessment · credit applied to onboarding. See if you qualify →

Want the optimization playbook, not just the platform overview? Get our free ChatGPT Ads cheat sheet, context hints, bid-floor tests, industry readiness, and the measurement stack from practitioners running real budgets.

Quick answers