How can I tell whether AI search is bringing us customers?
A customer can find you through an AI answer without clicking its link. Here is how to connect the evidence you have to inquiries, sales and spending decisions.
Picture a new client telling your intake coordinator that ChatGPT suggested your firm. Your analytics shows no visit from ChatGPT that week. The client read the answer, searched your name on Google and called the number listed there. Her explanation supplies a connection the traffic report cannot show.
To understand whether AI search is bringing you customers, follow the inquiries through to sales and keep the evidence of how those buyers found you. Visits from AI assistants and customers' own accounts can establish where AI search played a part. They cannot by themselves tell you which customers you would have won anyway. That distinction matters when you decide whether the work is paying off.
What your analytics can see
Google Analytics now has an AI Assistant channel for visits from recognized sources such as ChatGPT, Gemini and Copilot. Google's AI Overviews and AI Mode are an exception: their visits count as organic search. The channel is a useful place to start, but its total is not a count of everyone who encountered your business in an AI answer.
Other reports answer a different question. Search Console's generative AI report shows impressions in Google's AI features, while Bing Webmaster Tools reports how often supported AI experiences cite your pages. These reports describe appearances rather than the customers those appearances produced.
A visit becomes commercially useful evidence when you can connect it to a qualified inquiry and eventually a sale. That means preserving the source information available with a form submission or tracked call, then matching the inquiry to the outcome in your customer records. An assistant referral that leads to a signed client tells you more than an increase in visits alone.
It still would not capture the client in the opening example. She never clicked a ChatGPT link.
Ask the people who bought
Give new customers a way to tell you about AI assistants when you ask how they found you. If someone mentions ChatGPT, a follow-up about what they asked and whether they already knew your business can reveal its role in the decision.
A remark such as "ChatGPT listed three firms and you were the only one that handled construction cases" would tell you something a traffic source cannot. The answer helped the person distinguish between providers. That is different from using an assistant to look up the phone number of a firm they had already chosen.
Keep those details alongside the customer's inquiry and eventual purchase. Where a referral and a conversation describe the same customer, count that person once. Recollection can be incomplete, so preserve the customer's explanation without turning it into a more definite claim than they made.
Credit is a different thing from cause
Advertising offers a useful lesson. In a large field experiment, eBay stopped buying search ads on its own brand name on Yahoo and MSN. Almost all of the click traffic still arrived through ordinary search results. In the MSN test, after accounting for seasonal effects, total clicks from the search engine fell by only about half a percent. The ads had been receiving credit for visitors who could readily find another route to eBay.
That experiment concerns paid search, not AI recommendations. The useful distinction is between appearing on a customer's path and changing their decision. A buyer who first discovers you in an answer and a buyer who asks an assistant to evaluate an existing shortlist have different histories. Both answers could influence a sale. Neither the referral label nor the wording of a question settles what would have happened without it.
Your records can show an assistant's documented role. Estimating how much additional business the work produced requires looking beyond those records to what else changed and what was happening before the work began.
Read the changes together
Before new work begins, record how AI answers describe your business for relevant buyer questions. Note inquiry levels, customer sources and sales over a comparable period. Agree what the work is meant to change and when to review it, allowing for how long customers take to buy.
If relevant answers improve and more customers describe discovering or evaluating you through an assistant, that is encouraging evidence. A new campaign, seasonal demand or a change in tracking could also affect the numbers. A before-and-after increase needs that context before it is credited to the AI search work.
Flat inquiries deserve the same care. Better sampled answers do not establish that more real buyers saw them. The questions sampled may have little demand, the buying cycle may still be underway, or the tracking may miss later contact. Those are reasons to investigate rather than immediately declare success or failure.
Flaregraph's published work with a law firm illustrates the boundary. Across three sampled answers per question, ChatGPT went from recommending the firm in none of the answers to recommending it in all three for two selected service topics. The dated case records visibility progress; it does not measure inquiries, signed cases or revenue.
Deciding whether it pays
For inquiries that have not closed, estimate their potential using a realistic conversion rate and the value of a customer after the cost of serving them. As sales close, replace estimates with actual results. Keep that value separate from the amount you can reasonably attribute to the work.
Then ask how much additional business would be needed to cover the investment. If the spending only makes sense when every associated sale is treated as new business caused by AI search, the case is fragile. A more credible decision considers the customer accounts, the earlier baseline and other activity that could explain the results. Test whether the spending still makes sense under a more conservative estimate of its contribution.
Promising evidence may warrant continuing to the agreed review point; it does not automatically justify increasing the budget. If a fair review period produces no credible connection to commercial results, revisit the scope or stop the work. The purpose of measurement is to improve that decision, not to make every visibility gain look like a financial return.
Where to start
Flaregraph's AI visibility report shows how AI answers represent your business, where important information is missing or inaccurate and what to improve first. It gives you a clearer view of the problem before you decide which work is worth funding. Customer and sales records remain the evidence for what happens commercially.