According to a benchmarking report published by Invoca on July 13, calls forwarded by ChatGPT are more likely to qualify as leads than calls from any other channel. However, once answered, these calls convert at about the average rate.
The report states that the lead rate of calls directed to ChatGPT is 49%, which is almost 10 percentage points higher than the average of the seven channels tracked by Invoca and 6 points higher than Google Business profiles at 43%. The conversion rate from these leads is 40%, compared to an average of 42% for all channels. Invoca considers this to be roughly equivalent.
All figures represent estimates from Invoca customers, based on more than 70 million calls and 600 million minutes of conversation in 10 industries. Invoca sells the call tracking and chat analytics that generate this data.
Invoca says this is the first year it has enough data to measure AI-driven search calls that are productive at all.
What the Data Shows
Across all industries, approximately 56% of calls to businesses are answered by a human. If the call lasts longer than 15 seconds, the response rate increases to about 65%, and for calls longer than 30 seconds, it rises to about 71%. Of the calls answered, about 38% qualify as leads, and about 42% of those leads convert during the call.
ChatGPT sits above that baseline in the first number and below it in the second.
Paid search continues to generate more calls, leads, and conversions between paid channels in the dataset. For multi-location businesses, Google Business Profiles are a top source of organic content. Invoca emphasizes that channel efficiency and scale are separate factors, and percentages alone do not reveal which channel brings in the most business.
What the Report Can Say
Invoca does not publish how many calls directed to ChatGPT 49% are accounted for, it only notes that Invoca does not publish how many calls directed to ChatGPT 49% is accounted for, it only notes that the overall volume caused by AI generated remains very low. When the rate is obtained from a small base, it tends to be less reliable compared to the same rate calculated from a much larger paid search volume.
The report does not specify the measurement window. The methodology explains that the statistics are based on calls tracked and analyzed on the Invoca platform across 10 industries and seven marketing channels, but does not specify a specific start or end date. Gemini, Claude, and Perplexity are not included in the channel classification. Invoca notes that this is a measurement limitation rather than commenting on those assistants, stating that ChatGPT is the only major language model that produces measurable call volume in their dataset.
How Invoca installs calls
Invoca lists the calls as ChatGPT-referred, but the report does not contain details on how this feature works, such as if callers clicked through to ChatGPT, tracked numbers used, or contacted the business in other ways. It only counts calls that are directly caused by ChatGPT and not those from users who researched a business in the assistant and later called via untracked methods.
I compiled a version of that benchmark in June, where the same web data linked ChatGPT product recommendations to a 2.5x higher chance of site visits within seven days. Most of the affiliate traffic came as branded searches rather than direct referrals, which limits what a typical referral report can show. The calls add another attribution problem because the report doesn’t specify what digital method Invoca used to connect them to ChatGPT.
Why This Matters
Calls attributed to ChatGPT qualify as leads more often than calls from other Invoca track channels, by about 10 points. When someone collects, they convert almost as much as businesses are treated with everyone else. That makes it difficult to read what has been building around AI transfers over the past year.
I wrote in May about Adobe’s finding that the conversion rate had turned on AI-directed traffic to US retailers. In twelve months, it went from being the worst performing channel to converting 42% better than others. The explanation given is that research has already taken place within the assistant. Invoca Data is the first half of that. A person comparing options with an assistant and then making a phone call may proceed with a purchase decision, which is how Invoca learns, too.
The second part of the data is not quite the same. Although a high lead rate is seen, it does not translate into a high call conversion rate when looking at Invoca’s average. In this dataset, differences appear in the fitness phase but disappear after that.
Looking Forward
Invoca believes this is a major signal to watch out for in an investable channel, backed by a volume caveat. The main metric that influences this view is call count, a non-descript report. Another question is whether the 40% is moving. If AI-referred callers continue to qualify at the top of the list while converting, the focus shifts from increasing call volume to understanding what happens during those calls.
The report also notes that 64% of businesses do not ask callers to purchase or schedule an appointment, which is a problem on the business side.
