How ChatGPT chooses which business to recommend

When an assistant searches the web to answer a question, it tends to cite pages whose subject matches the question. Asked about electricians, it looks for a page about electricians, and a general page rarely qualifies. That is the clearest pattern in SoudCoh's tests.

Key facts

• 60% of the pages AI assistants cited, when asked who does marketing or websites for a named trade, were pages built for that exact trade.
• Asked the same buyer question three times, only 25% of the websites an assistant cited appeared in all three answers.
• Visits that analytics could attribute to AI assistants were 0.2% of SoudCoh's own website sessions.
• Trade owners shopping for marketing help searched Google by service and place: 41% of non-brand searches that showed our site named a place.

Because AI assistants answer the question they were asked. The pages cited in SoudCoh's tests were built for the trade in the question. A general “we do digital marketing” home page rarely qualified, even for businesses that clearly do the work.

A homeowner's question works the same way. Someone asking an assistant who can replace a hot water system in Frankston, or which electrician does switchboard upgrades near Parramatta, is asking about a specific job in a specific place. The page most likely to be cited is one that is about that job, says where the business works, and answers the obvious next questions.

It is the same discipline good search advertising already needs: one page per service, and a real page for each area you want work from. If you want to be recommended for a trade or a job, publish a page for it.

Because AI recommendations are drawn from a short list, not read off a fixed ranking. When we asked an assistant the same buyer question three times, only 25% of the websites it cited appeared in all three answers.

Two practical conclusions follow:

• Being on the list matters more than being first in any single answer. A page that directly answers the question, a directory profile and independent mentions all help a business stay on the list.
• Never judge AI visibility on one answer. Ask many times, on more than one assistant, and record how often you are named.

The durable approach is the same as for search: pages that answer specific questions clearly, backed by independent mentions.

Only a fraction of them. On our own website, visits that analytics could attribute to AI assistants were 0.2% of sessions. That understates AI discovery badly, because most AI-influenced visits arrive with no referrer. When someone copies a business name from an AI answer into a browser, or taps a link inside an app that drops the referrer, nothing tells analytics where they came from, and the visit is filed as Direct. Some assistants tag their own links, which is why a dedicated AI channel catches part of this traffic.

Three fixes, all cheap:

1. Ask every new enquiry how they found you, and store the answer on the lead. “ChatGPT”, “Google”, “a neighbour” and “saw the van” are all useful answers.
2. Give AI assistants their own channel in Google Analytics, with a custom channel group that matches referrers such as chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com, so the visits that can be identified are not lost in Referral.
3. Watch Direct traffic and searches for your business name together. A rise with no campaign to explain it can include people who met your name in an AI answer.

Yes, but for a different kind of question. On our own site, trade owners shopping for marketing help searched Google by service and city: 41% of non-brand searches that showed our site named a place, while searches phrased as questions barely registered (0.2%). Questions go to AI assistants; Google still gets the short commercial search.

The device tells a similar story. The people who buy marketing for a trade business searched on desktop (85% of the Google search impressions our own site earned), while their own customers' paid clicks were 76% mobile in SoudCoh's research sample. A homeowner choosing a plumber is usually on a phone, often in the middle of the problem.

So a trade website needs both: clear service-and-suburb pages that rank for the short search and load fast on a phone, and plain-English answers to the questions people now ask assistants instead.

The pages AI assistants cite share a pattern, and none of it needs a big budget. It needs specific, accurate content. This is the list we work to, adapted for a trade business:

• A page for each service you want more of, with the service in the address, the title and the first paragraph.
• A real page for each area you want work from: local jobs, local photos and travel times, not a template with the suburb name swapped.
• The answer first. Open with what the job is, whether you do it in the reader's area, and what happens next, in plain language.
• A short list of facts that stand on their own: who it is for, what is included, how it is priced, where you work, your licence number.
• A small FAQ in your customers' words. The questions people ask on the phone are the questions they now ask assistants.
• Evidence: real job photos, licence details, reviews on independent sites, and any figures you can back up.
• Consistent facts everywhere. The same business name, phone number, service area and hours on your website, your Google Business Profile and every directory you appear in.

Across the trade websites we have built, three components recur: an explanation of what happens after someone enquires (80% of builds), clear suburb coverage (75%) and a page for each service (66%). They answer the questions a homeowner has before calling: do you do this exact job, do you come to my suburb, and what happens after I press send?

Two technical checks support the pages without replacing them. If you want to appear in ChatGPT's search answers, make sure your robots.txt does not block OpenAI's search crawler, OAI-SearchBot. And keep a clean sitemap and accurate structured data. Google says no special markup is needed for its AI features; the pages that do well in search are the ones it can draw on.

You can run the same kind of test we do, by hand, in an hour or two a month:

1. Write ten buyer questions in your customers' words: the service and suburb, the problem, the price question. For example, “who installs heat pump hot water in Geelong”.
2. Ask each question several times, on more than one assistant, in a fresh chat each time.
3. Record who is named and which pages are cited in a simple spreadsheet.
4. Study the pages that win. What do they say, plainly and early, that yours does not?
5. Fix one page at a time, then ask the same questions again the following month.

A single answer is noise; the pattern across many answers is the signal. Our white paper Marketing for home services in Australia puts AI discovery alongside search, measurement and growth, and From keyword to revenue covers how to measure where enquiries really come from. For the pages themselves, see our local SEO and SEO work, our website design for electricians, plumbers and cleaning businesses, and our marketing for electricians, plumbing businesses and cleaning businesses. Or talk to us about the services you want to be recommended for.

How does ChatGPT choose which businesses to recommend?

When it searches the web, it tends to cite pages that answer the exact question asked. In SoudCoh's tests of AI answers, 60% of the pages AI assistants cited were built for the exact trade asked about, while home pages and directories were cited far less. A clear page for each service and area gives an assistant something specific to quote.

Why does ChatGPT recommend a different business each time I ask?

Recommendations are drawn from a short list rather than read off a fixed ranking. When we asked the same question three times, only 25% of the websites cited appeared in all three answers. Aim to stay on the list, through clear pages and independent mentions, and measure by asking many times.

How can I tell if customers found my business through ChatGPT?

Ask every new enquiry how they found you, and record the answer on the lead. Analytics sees only part of it: on our own site, identifiable AI-assistant visits were 0.2% of sessions, because most AI-influenced visits arrive with no referrer and are filed as Direct.

Does a small business need a page for every service to be recommended by AI?

For the services you want more of, yes. AI assistants answer the question asked, and a general home page rarely matches a specific question. Each page should lead with the answer, name the areas you cover and explain what happens after someone enquires.

Do reviews and directory listings help with AI recommendations?

They help keep a business on the short list, alongside independent mentions. But in our tests, directories, review sites and social profiles made up only 1.8% of the pages assistants cited; most citations went to pages built for the exact question. Keep your facts consistent across all of them.