Find out what 14 million calls tell us about the shift in AI search

Executive summary

Signal 1: the phone call is becoming intentional

Signal 2: paid search is generating high-intent callers

Signal 3: the friction point has moved downstream

Signal 4: AI-referred callers already convert

What this means: four actions

Closing note

Executive summary

+31%

Visit-to-call rate
Jan 2025 to Jun 2026

1.75x

PPC call rate vs SEO
Behaviourally proven, not assumed

1989%

AI referral growth
Jan 2024 – Apr 2025

Four signals from the data point in the same direction:

Call volume is rising, not falling — and visit-to-call rate has increased 31% in 18 months.
Paid search generates calls at 1.75x the rate of organic, with identical quality.
The reason calls fail has fundamentally changed: buyers arrive informed, but businesses can't match AI-set expectations on price and availability.
AI platforms are already a trackable referral channel, and the calls they generate convert at a higher rate than PPC.

Signal 1:
The phone call is becoming intentional

+31%

Visit-to-call rate
Jan 2025 to Jun 2026

41M+

Calls tracked
Last 12 months, all installations

+24%

Call volume growth
H1 2026 vs H1 2025

The prevailing assumption in performance marketing is that AI assistants reduce search volume, website traffic and, in turn, inbound calls. Our data doesn't support this.

Across 41 million calls tracked over the last 12 months, call volume grew 24% H1-on-H1. More significantly, the visit-to-call rate (the proportion of website visitors who choose to call) has risen from 1.59% in January 2025 to 2.04% in June 2026. That's a 31% increase in the share of visitors choosing to pick up the phone.

Two step-changes are visible in the data. The first happened in September–November 2025, coinciding with the broader rollout of ChatGPT search capabilities. Visit-to-call rate jumped 27% within three months. The second step-change arrived in May–June 2026, establishing a new baseline above 2.0%.

The interpretation is consistent: AI search is pre-qualifying buyers before they reach a website. The people who then choose to call have already done their research. So, the call is the last step in a journey that AI has already structured.

What this means for your media strategy

If your attribution model only counts web form completions, you're measuring a shrinking share of your conversions. The proportion of buyers who research via AI and then call to commit is growing. That means bidding strategies that exclude call conversion data are systematically undervaluing paid search.

Signal 2:
Paid search is generating high⁠-⁠intent callers

3.08%

PPC visit-to-call rate
vs 1.76% organic search

1.75x

Higher than SEO
For equal visitor volume

71%

PPC long-call rate
Quality matches organic (73.6%)

Channel-level call behaviour across 11.8 million PPC-attributed calls shows the relationship between paid search and inbound calls directly.

PPC visitors call at 3.08%. SEO visitors call at 1.76%. Display visitors call at 0.46%. Social visitors call at 0.18%. The differences aren't marginal. PPC generates calls at 1.75 times the rate of organic search and 6.7 times the rate of display.

The quality metric is equally significant. Of PPC-attributed calls, 71.1% last more than one minute, which is a relatively substantial conversation. The equivalent figure for SEO is 73.6%. Display sits at 49.8%. The channel that generates calls at the highest rate also sustains conversations at the same quality as organic search.

Display, by contrast, does exactly what awareness traffic should do. It reaches a broad audience, generates occasional calls, and most of those calls are short because the visitor wasn't ready. That's the correct behaviour for a channel optimised for reach, not conversion.

Retrain your bidding model, not your algorithms

Your bidding strategy is already adapting to clicks. If it isn't also adapting to call conversions (by channel, by duration, by outcome), it's working from an incomplete picture. Google Ads and Meta both accept offline conversion imports. Infinity has native integrations with both. The data is available, and so the question is whether it's being fed back.

Find out more about Infinity's integrations

The full funnel: from click to closed deal

Aggregating across Smart Outcome-enabled (SO-enabled) installations for the last 12 months, the funnel from website visit to verified positive outcome looks like this:

121M

Website visits
SO-enabled installations, 12m

1.3M

Sales-intent calls
23% of SO-processed calls

349K

Positive outcomes
26.3% of sales-intent calls

Of 5.75 million inbound calls, Smart Outcomes verified 1.3 million as carrying purchase intent. Of those, 349,000 resulted in a positive outcome. That means for every 350 calls where purchase intent was confirmed, 260 didn't convert.

That number, 260 out of 350 verified sales-intent calls failing to convert, isn't a traffic problem or an intent problem: it's a closing problem, and the next signal identifies where that closing problem has emerged.

For every 350 verified sales-intent calls, 260 didn't convert, and the barrier is closing, not traffic.

Signal 3:
The friction point has moved downstream

31.6%

Price/budget barrier
Q2 2026 - #1 reason for non-conversion

+77%

vs Q2 2025
17.8% → 31.6% in 12 months

-55%

Product understanding
17.7% Q2 2025 → ~8% Q2 2026

Comparing Smart Outcomes barrier data between Q2 2025 and Q2 2026 across non-converting sales leads reveals a structural shift in where calls are failing.

In Q2 2025, the primary barriers were:

Decision-making (20.8%)
Pricing uncertainty (17.8%)
Product understanding (17.7%)

Buyers weren't ready, and they needed educating, so the sales agent's role was informational.

In Q2 2026, that picture has changed materially. Price and budget mismatch is now the leading barrier at 31.6%, up 77% in 12 months. Availability issues, which barely registered in 2025, are now the second-largest barrier at 18.3%. Product not offered has emerged as a new category at 12.8%. Product understanding, meanwhile, has more than halved.

The explanation is consistent with what we know about AI search behaviour. Buyers are now arriving at the call having already used ChatGPT or Perplexity to research their options, compare products, and in many cases, get an indication of price. They arrive informed, and they arrive with expectations.

When the price the agent quotes doesn't match what the AI described, the call fails, not because the lead was poor quality, but because the business couldn't match the expectation AI had created.

Retrain your agents, not your algorithms

Your bidding strategy is already adapting to clicks, but your agent handling hasn't adapted to a caller who arrives already sold. The objection ending more calls than any other in 2026 is not "I need more time to think" but "your price doesn't match what I was told." Agents who can handle that conversation are your top performers. The ones who can't are losing sales that were already won.

See Smart Outcomes in action

Your brand matters more, not less

AI synthesises options and recommends one. The brands that appear in ChatGPT results earn that presence through review volume, content authority, and brand signal. Brand investment is now a pre-call performance lever, not a separate strategy from performance marketing. The brand that AI recommends gets the high-intent call.

Signal 4:
AI-referred callers already convert

1989%

AI referral growth
Jan 2024 – Apr 2025

68247

Landings/month
Apr 2025 sustained baseline

78.5%

ChatGPT share
Of all AI referral traffic

AI platforms are already generating measurable, trackable referral traffic to websites tracked by Infinity. This is already happening, and the conversion behaviour of those visitors, when they call, is already visible in the data.

Between January 2024 and April 2025, direct referrals from AI platforms grew 1,989%. ChatGPT.com accounts for 78.5% of all AI referral traffic. Perplexity.ai and Gemini account for a further 20%. The acceleration isn't gradual: a single month in August 2024 saw a 167% month-on-month increase, coinciding with the expansion of ChatGPT's web browsing capability.

What happens when AI-referred visitors call

Infinity tracks channel type "ai" across all active installations, using the same mechanism as PPC, SEO, and referral channels. When a visitor clicks directly from a known AI platform to a tracked website and then calls, that call is attributed to the AI channel.

In the last 12 months, that produced 16,381 calls. Their behaviour is striking:

73.6%

Long-call rate (>1 min)
Identical to SEO; higher than PPC (71.1%)

1.45%

Smart Match event rate
1.46x higher than PPC (0.99%)

16381

AI-attributed calls
Last 12 months, all installations

AI-referred callers stay on the line at the same rate as organic search visitors, and generate offline conversion events at a rate 46% higher than paid search. They're high-intent buyers arriving via a channel that most attribution models don't track.

Measure the moment before the call

AI referrer traffic is trackable today, and no special tool is required. Check your referrer data for chatgpt.com, perplexity.ai, and gemini.google.com because It's probably already there. If you aren't attributing it as a channel, those conversions are being given to whatever touchpoint comes last, usually paid search or direct, so you're giving PPC credit for leads that AI actually delivered.

What this means:
four actions

Each of the four signals points toward a practical change. You don't need a new platform or a new budget, but you do need to look at the data that already exists in a slightly different way.

Measure the moment before the call

AI referrer traffic is trackable today, and no special tool is required. Check your referrer data for chatgpt.com, perplexity.ai, and gemini.google.com because it's probably already there. If you aren't attributing it as a channel, those conversions are being given to whatever touchpoint comes last, usually paid search or direct, so you're giving PPC credit for leads that AI actually delivered.

Track AI as a channel, not a footnote

Add chatgpt.com, perplexity.ai, and gemini.google.com to your channel taxonomy, which takes ten minutes. The traffic is already arriving. AI-referred callers generate conversion events at 1.46x the rate of PPC. If you aren't tracking it, those conversions are being misattributed.

Fix the closing problem, not the traffic problem

That leaves 260 out of every 350 verified sales-intent calls not converting. The barrier is price and availability mismatch, not lead quality. The caller has already been sold by AI. The job now is to close, not to educate. Brief agents on the new objection landscape. The call that fails in 2026 starts with "the AI told me it would cost..."

Build brand presence where AI looks

AI synthesises from authoritative sources. Review volume, content depth, structured data, and brand mentions in trade publications all influence whether your brand appears in AI-generated recommendations. Brand investment is now a pre-call performance lever. The brand that AI recommends gets the high-intent call.

Closing note

The data in this paper comes from 41 million calls across every major sector. None of it was configured for this purpose. It's the product of passive measurement running in the background while marketing teams focused on clicks, impressions, and form fills.

The signals were there. They were just not being read.

The buyer journey hasn't ended. It's been restructured. AI handles the discovery phase. The phone handles the commitment phase. The marketers who understand that distinction, and build their measurement, attribution and commercial models around it, will have a structural advantage over those who do not.

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.

Daniel Wilkinson, Chief Customer Officer, Infinity

2215743.fs1.hubspotusercontent-na1.nethubfs2215743AI-Feedback-Loop-Cover-2-Apr-17-2025-09-42-08-6723-AM

Find out how to feed the right signals back to your bidding platforms with the AI feedback loop.

Get the guide

Uncover growth opportunities and identify conversion barriers with our Rapid Insights projects.

Discover more

Infinity-Conversation-Analytics-for-Marketers-Guide-MR

Build your brand presence and refine messaging with the insights from your customer conversations.

Find out how

Ready to see what your own call data is telling you?

+44 3892 2600

“Infinity are an innovative market leader that are investing heavily into their product.”
Head of PPC (Senior Partner)

About the author

Daniel Wilkinson is Chief Customer Officer at Infinity. He has spent 20+ years at the intersection of paid media and technology, most recently building JUMP and SafetyNet, Jellyfish's bid management and automated spend-protection products. He joined Infinity in 2024 to bridge call intelligence and paid media, leading the go-to-market for Smart Outcomes as Chief Solutions Officer before moving into his current role.

About Infinity

Infinity is the Call Analytics Platform that helps you turn customer conversations into measurable revenue growth by improving demand quality and conversion performance. Tracking 41 million+ calls per year across automotive, financial services, travel, home services, healthcare, retail, legal and more, Infinity connects paid media spend to phone-based outcomes and feeds first-party conversion signals back into Google, Meta and programmatic platforms.

Data sources and methodology

All data in this paper is aggregated and anonymised across Infinity's active customer base. No individual customer or consumer data is identifiable. Sources used:

QBR Machine v1.2 (ElastiCube): visit-to-call rate, call volume, channel breakdown — last 12 months, all active installations, void calls excluded.
Internal Metrics (ElastiCube): cross-customer channel-type breakdown including AI channel calls, long-call rates, Smart Match event rates — last 12 months.
hub-dashboard-so (Live model): Smart Outcomes intent categories and barrier data — Q2 2025 vs Q2 2026, non-converting sales leads.
AI Chat Referrers (ElastiCube): AI platform referral traffic — Jan 2024–Apr 2025. Pipeline not refreshed beyond Apr 2025; growth trajectory is accurate, absolute current run-rate is not available from this source.

Smart Outcomes barrier theme names changed between SO v1 and SO v2 taxonomy migration. Directional findings are robust; precise percentage-point changes between periods should be read indicatively.

WHITE PAPER What 41 million phone calls tell us about the AI search shift

Contents

Executive summary