AI visibility for local service businesses (plumbers, clinics, salons)

For a local service business, AI visibility is decided mostly off your own website: ChatGPT recommended just 1.2% of business locations, against the 35.9% that surface in Google's local 3-pack, per SOCi's 2026 Local Visibility Index (published 25 March 2026). An assistant naming two plumbers is doing something structurally different from a results page listing ten, and it assembles that shortlist by cross-referencing map, review and directory data long before it opens your site.

Local sites are where we see the widest gap between a page that looks fine to a human and a page a bot can actually read — which is a large part of why Contexta grew a live crawler test in the first place. A booking widget, an opening-hours table drawn by a script, a phone number baked into a header image: all of it invisible to the system deciding whether you get named.

Why is AI visibility harder for a local business than a Google ranking?

Because the assistant has to commit to two or three names instead of listing ten links, and scarcity makes it conservative. A local pack can afford to include a business whose details are a bit inconsistent; an answer that recommends one plumber cannot, so the systems default to businesses whose facts agree everywhere they can be checked. Ranking is a sorting problem. Being recommended is a confidence problem.

35.9%

of locations surface in Google's local 3-pack

11%

recommended by Gemini

7.4%

recommended by Perplexity

1.2%

recommended by ChatGPT

Read that as a shape, not a scoreboard. It's vendor research — SOCi sells local marketing software — and the page reporting the figures doesn't publish the sample or the method alongside them, so the exact percentages deserve less weight than the gap between the columns. The finding that holds up regardless is that traditional local visibility and AI recommendation are different achievements, and the first does not deliver the second.

The uncomfortable consequence is that most of the deciding evidence sits on property you don't own — maps listings, review platforms, industry directories, forum threads. That's the same dynamic behind why Reddit, reviews and directories decide AI visibility more than your blog, except a local business feels it harder, because there's no product catalog for an assistant to fall back on.

If the assistant doesn't cite your website, what is your website for?

Corroboration. Your site is rarely the source of the recommendation, but it is routinely the source that confirms or contradicts it — and a contradiction is expensive, because it gives the model a reason to hedge or to pick the competitor whose details line up. You're not writing to be quoted here; you're publishing a reference copy of the facts other sources are asserting about you.

Google states the supply chain plainly in its own guidance: "Using products like Merchant Center (such as Merchant Center feeds) and Google Business Profiles can help your products and services to be visible in both AI responses and other Google Search results" (Search Central, AI features optimization guide, page updated 2026-07-10, checked July 2026). For a service business with nothing to feed a Merchant Center, that sentence reduces to one instruction: the Business Profile is the primary input, and the website's job is to agree with it.

This inverts the usual advice. Most of the answer-first, get-quoted work that pays off elsewhere — the sentence-level moves in writing for GEO instead of SEO — earns you long-tail question traffic ("how much does a boiler service cost in Leeds"), not a slot in the "best plumber near me" shortlist. Both are worth having. Only one of them is what people mean by AI visibility for local, and confusing them is how a year of blogging produces no recommendations.

Which facts have to match everywhere, and what breaks when they don't?

Six, and they're mundane: legal business name, address, phone number, service area, opening hours, and the actual list of services. When those disagree between your site and your listings, the model either hedges the recommendation or skips you for a business it can state confidently.

• Legal business name, spelled the same way everywhere — no "& Sons" on one and "and Sons" on another
• One phone number, as text in the page source, not inside an image or a click-to-call script
• Service area named as places people say — towns and neighbourhoods, not a radius or a map polygon
• Opening hours as readable text, including the seasonal or holiday exceptions
• Services listed by name, in the words customers use, not as marketing categories
• Licence, registration or credential numbers where your trade has them

The most common breakage we see isn't wrong information, it's information that only exists in a form a machine can't read. Hours rendered by a booking plugin. A service area shown as a shaded map. A phone number in the logo. To a person the page is complete; to a retrieval system the page is silent on all three, so it falls back to whatever the directories say — which may be two years out of date.

Structured data helps here, but as backup rather than substitute: a LocalBusiness block restates the same facts in a form nothing has to interpret, which is confirmation for the prose, never the delivery layer. If the visible page and the markup disagree, you've made the problem worse, not better.

Why do local service sites fail machine-readability more often than shops?

Because their most important facts are exactly the ones most likely to be drawn by a script or a picture. An ecommerce page has prices and specs in HTML because a catalog put them there; a salon site has its hours in a widget, its booking behind an iframe, its phone in the header graphic and its address in the footer image, all added by whoever built the theme. The information is present and none of it is text.

Confirming this takes a fetch, not an opinion, which is what Contexta's AI Visibility test does: it requests your pages as GPTBot, OAI-SearchBot and PerplexityBot, reports the status each one gets, flags a Cloudflare or firewall block sitting in the way, and shows what survives with JavaScript switched off. On local sites the JavaScript check is usually the one that stings — the crawler gets a clean 200 and reads a page with no hours, no phone and no services on it.

Two blocks worth ruling out early, because both are common on agency-managed and shared hosting: a security plugin that blanket-blocks unfamiliar user-agents, and a CDN rule doing the same at the edge where your robots.txt can't see it. Either produces a site that looks perfect in a browser and returns nothing to the systems making recommendations, the same class of failure covered in content that's invisible because it needs JavaScript.

What's overhyped here, and what actually moves it?

The overhyped part is content. You cannot write your way into a "plumber near me" shortlist, and any retainer sold on that promise is selling you blog posts against a problem blog posts don't touch. Content earns the question queries around your trade; reviews, listing accuracy and consistency earn the recommendation itself, and no amount of the first substitutes for the second.

What actually moves it, in the order we'd do it: make your Business Profile complete and current, since that's the input Google names explicitly. Get your six core facts identical across your site and your main listings. Make those facts readable as text without JavaScript. Then keep reviews arriving steadily rather than in bursts — recency is doing real work in these systems, and a wall of five-star reviews from eighteen months ago reads as a business that may not be trading.

One honest limit on all of it: as of mid-2026 nobody outside the labs knows the weighting, and the assistants change their local sourcing without announcing it. Anyone quoting you a guaranteed placement is describing a system they can't see. The defensible position is to be the business whose facts are boringly consistent and machine-readable everywhere — that survives whatever the weighting turns out to be this quarter, which is more than can be said for any tactic aimed at a specific engine.