ChatGPT Knows Your Company. Does It Recommend You?

You type your company's name into ChatGPT. It knows your services and gets the city right. That is encouraging, but it leaves the business question unanswered: would it name you when a homeowner asks who to hire?

For owners and marketing managers, recognition and discovery need separate checks. A branded question hands the assistant your company name. An unbranded service question asks it to select from possible providers.

A familiar name is not a place on the shortlist

BrightLocal's October 21 research highlights this gap: confidence that a business appears in AI can come from checking its name rather than the questions prospective buyers ask. Its survey covers 1,000 US small-business owners across industries. It does not establish how often contractors are recommended.

A citation is different again. An assistant might use your article to explain water-heater maintenance while recommending another company, or no company at all. Record what the answer actually does.

Compare three kinds of question

The following are illustrative prompts for a fictional plumbing business, not homeowner query logs or measured search demand:

• Recognition: “Does Example Plumbing install water heaters in [city]?”
• Discovery: “Who should I contact to replace a water heater in [city]?”
• Evaluation: “What should I compare when choosing a water-heater installer in [city]?”

Recognition tests the assistant's description of the brand. Discovery tests whether it selects the business for that question. Evaluation helps you inspect the criteria and sources it presents to a buyer. Keep branded and unbranded results separate in reporting.

Run a repeatable check

1. Define the service and location. Use jobs the company accepts and places it actually serves. Do not imply emergency coverage or credentials the business does not have.
2. Fix the question panel. Save exact wording and identify whether each question is branded or unbranded. If you change wording, label it as a new test.
3. Record the conditions. Note platform, date, location settings, search mode and whether the conversation began fresh. Use comparable conditions for repeat checks.
4. Preserve the evidence. Save the answer and linked sources. Record mentions, citations, explicit recommendations, factual errors and cases where the assistant declines to name anyone.
5. Repeat and report the denominator. “Recommended in two of six observations on this platform during these dates” describes a test sample. It is not market share, search volume or a prediction of leads.

BrightLocal's September local AI visibility study used repeated, localized tests and found variation across platforms and locations. Those were generated test responses, not observed homeowner searches. That distinction applies to your own monitoring too.

Use the answers to find something worth improving

Suppose the assistant describes your business using an old service area. Follow its cited sources and check the current website and listings. Correct inaccurate information where you control it and record correction requests elsewhere. Do not assume a citation caused every part of the answer.

If your service page says little beyond “water heater installation,” explain what the team evaluates, what the quote covers, which options are offered and which questions need an on-site assessment. Add a documented job example when you have permission and sufficient evidence.

If other providers appear, examine the sources supporting their inclusion. Relevant reviews, clear service information or independent coverage may expose a gap to investigate. Their presence does not prove a ranking formula or justify buying directory placements indiscriminately.

Google's AI-feature guidance says normal SEO practices still apply, with no additional technical requirements. Clear evidence and accurate information serve the homeowner regardless of whether a particular assistant changes its answer.

Connect the test to inquiries

Maintain separate records for mentions, citations, recommendations, identifiable referral visits, actual inquiries and booked jobs. Ask new prospects how they found you and preserve what they volunteer. A reported AI recommendation is useful intake evidence, but memory alone will not reconstruct every step.

Do not add every platform's conversion count together. One homeowner may encounter several channels before calling. Reconcile calls, forms and bookings so the same opportunity is not counted repeatedly.

Use our AI search methodology to define the baseline and our search everywhere optimization guide to connect the work across your website, listings and other channels.

The goal is to learn where accurate business information reaches a relevant buyer and what happens next. A favorable answer is one piece of evidence. An inquiry and a booked job belong in their own columns.

Explore an AI search assessment to review a defined set of service questions, the public sources behind the answers and the gaps worth addressing.