A common forecast about AI assistants is that they will absorb customer enquiries, leaving fewer people to phone a business at all. A whitepaper from call analytics company Infinity, built on 41 million calls tracked across its customer base over the past 12 months, points the other way. Call volume grew 24% in the first half of 2026 compared with the first half of 2025, and the share of website visitors who chose to call rose from 1.59% in January 2025 to 2.04% in June 2026.
For business leaders, the more useful question is not whether people still call, but what kind of caller is now on the line.
A buyer who has already done the research
Infinity reads the data as AI search qualifies buyers before they ever reach a website. In the three months from September 2025, the visit-to-call rate jumped 27%, a rise that coincided with the wider rollout of ChatGPT’s search capabilities. A second step-up followed in May and June 2026, settling at a new baseline above 2%. On this view, the call has become the last stage of a journey that an AI tool has already helped to structure.
What the failed calls reveal
The sharpest evidence is in why sales conversations stall. Across installations using its Smart Outcomes tool, Infinity verified 1.3 million calls as carrying genuine purchase intent. Only 349,000 of them, roughly a quarter, ended in a positive outcome.
The reasons have shifted. In the second quarter of 2026, price and budget mismatch was the leading barrier on non-converting sales leads at 31.6%, up from 17.8% a year earlier. Availability ranked second at 18.3%, having barely registered in 2025, and “product not offered” appeared as a new category at 12.8%. Difficulty understanding the product fell from 17.7% to around 8%.
Infinity cautions that its category names changed when it moved between versions of its outcome classification, so the direction of travel is more reliable than the exact percentage-point changes. They explained that buyers consult tools such as ChatGPT or Perplexity before calling, form a view on their options and often on price, then test that view against what the business tells them. That is the paper’s interpretation rather than something call data can prove directly, but it fits a move from calls about understanding to calls about terms.
Where the shift lands inside a business
In your reporting. Infinity argues that reporting which counts only web form completions undercounts conversions, because more buyers research through AI and then phone to commit. The AI channel is already measurable. Across Infinity’s installations, 16,381 calls over 12 months were attributed to visitors arriving directly from AI platforms. Those callers were as likely as organic search visitors to stay on the line beyond a minute, and they generated conversion events at a rate 46% higher than paid search referrals.
Daniel Wilkinson, Chief Customer Officer, makes the wider point in the paper’s closing note: “If you’re a marketer still treating phone calls as an offline black box, you’re leaving your best first-party data on the table.”
One caution on currency: the 1,989% growth in AI referral traffic that Infinity reports covers January 2024 to April 2025 only, and the paper states that a current run-rate is not available from that source. ChatGPT accounted for 78.5% of that traffic. Infinity’s practical suggestion is to check existing referrer data for chatgpt.com, perplexity.ai and gemini.google.com.
On the call itself. Infinity’s view is that agent scripts built to explain a product have not kept pace with callers who arrive with a figure. The paper says the objection ending the most calls in 2026 is that the price does not match what the buyer was told elsewhere, and its advice is to brief agents for that conversation, where the job is to close rather than to educate.
In AI recommendations. Infinity argues that review volume, content depth and mentions in trade publications influence whether AI tools recommend a business, which would make brand investment a lever on call quality rather than a separate discipline. The call data does not measure that link, so it is best treated as a working assumption to test.
Testing it against your own data
These findings come from one provider’s customer base, so they are worth checking against your own call records before they shape budgets. A simple starting point is to sample recent lost calls, count how many cite price or availability, and repeat the exercise on last year’s records if they exist.
If the pattern holds, the customer journey has not shrunk so much as relocated. Discovery happens inside an AI tool, and what reaches your business is a buyer ready to negotiate.
