Call duration has been the standard proxy for call lead quality in Google Ads for years: if someone stays on the phone for 60 seconds, they are probably a real lead. If they hang up in 15 seconds, they are probably not. It is a reasonable heuristic and a terrible optimization signal for service businesses where a 90-second wrong number and a 90-second genuine appointment inquiry look identical to the algorithm.
AI-qualified call leads attempts to move past the duration proxy. The idea is sound. The implementation requires validation before you trust it as a Smart Bidding primary conversion.
The problem with optimizing to call duration
For service businesses where phone calls are the primary conversion mechanism — legal, HVAC, dental, roofing, medical aesthetics — the conversion tracking setup often goes: create a conversion action, set a minimum call duration (60 or 90 seconds), turn on Smart Bidding. The algorithm then optimizes toward generating more calls above that duration threshold.
The problem is that Smart Bidding cannot distinguish between a 90-second call from a competitor doing market research, a 90-second call from someone who cannot afford your service, and a 90-second call from a genuinely qualified buyer who booked an appointment. All three count. All three influence which keywords, audiences, and times of day the algorithm bids up.
What AI-qualified call leads are
Google's AI analyzes the content of calls routed through Google forwarding numbers to determine whether the call represents a qualified lead based on conversation signals — the specificity of the inquiry, the presence of purchase intent language, the nature of the back-and-forth, and call outcome indicators. Calls that clear the quality threshold are classified as qualified and reported separately from total call volume.
What settings and prerequisites matter
- –Google Ads call reporting must be enabled — calls must route through Google forwarding numbers
- –Call recording or transcription must be enabled within Google Ads call settings
- –Sufficient call volume is needed for the AI to classify meaningfully — accounts with fewer than 50 calls per month may see classification with lower reliability
- –Call ads or location extension calls are the primary surfaces; website call measurement works if using Google forwarding numbers on the site
- –Compliance review is required in regulated industries — legal, medical, financial — where call recording has additional requirements
When this is useful vs risky
| Scenario | Useful or risky? | Why |
|---|---|---|
| High call volume service business, 100+ calls/month | Useful | Sufficient volume for model accuracy; meaningful signal improvement over duration |
| Legal or dental practice with regulatory call recording restrictions | Risky without review | Call recording compliance requirements must be verified before enabling |
| Low-volume B2B with 20 calls/month | Risky | Insufficient volume for reliable AI classification; use CRM import instead |
| Businesses where calls include sensitive personal information | Caution | Data handling and retention implications of AI call analysis should be reviewed |
| Accounts where sales team can audit AI classifications weekly | Useful | Human oversight allows quick identification of systematic misclassification |
Call-quality validation framework
- Enable AI-qualified call leads and run for four weeks without changing bidding strategy
- Pull a sample of 30 calls classified as "qualified" and cross-reference against CRM — did these calls result in appointments, quotes, or sales?
- Pull a sample of 30 calls classified as "unqualified" and check whether any were actually real prospects the sales team remembers
- Calculate accuracy rate: if above 75 percent for qualified and below 20 percent false negatives on unqualified, the signal is usable
- If accuracy is acceptable, set AI-qualified call lead as a primary conversion and run Smart Bidding for a minimum eight-week learning period
- Continue monthly accuracy spot-checks — AI classification quality can drift as call content changes seasonally
Service business examples
Legal
A personal injury law firm receives 200 inbound calls per month across Google Ads campaigns. Of those, approximately 60 meet the firm's minimum case value threshold. Using call duration as the conversion signal, Smart Bidding sees 140 or more "conversions" per month and optimizes accordingly. After enabling AI-qualified call leads and running a validation audit, the firm finds the AI classification correctly identifies 52 of their 60 qualified calls and excludes most wrong-number and low-value inquiries. Smart Bidding running on the qualified signal shifts budget toward the keywords and times that produce actual caseload rather than high call volume, reducing cost per viable case by 22 percent over three months.
HVAC
An HVAC company with seasonal call volume uses call duration as the primary signal. During peak summer months, emergency air conditioning calls are very short but extremely high value. Call duration misclassifies many emergency calls as low quality. AI-qualified call leads correctly identifies emergency service intent from the urgency and specificity of the conversation, improving bidding signal quality for the highest-value call type in the account.
eCommerce use case: high-ticket consultation calls
A custom furniture retailer takes inbound calls for large orders, custom design consultations, and trade inquiries. Call volume is modest — around 40 per month — but average order value for phone-initiated purchases is four times higher than online orders. The AI-qualified signal, even at lower volume, helps Smart Bidding identify which campaign types and keywords attract consultation-intent callers versus general product queries that do not convert to phone sales.
How to use it in bidding
- –Set AI-qualified call lead as the primary conversion action — this is what Smart Bidding optimizes toward
- –Keep raw call volume as a secondary conversion for monitoring — watch for divergence between total calls and qualified calls by campaign
- –If using Target CPA bidding, recalibrate the CPA target to reflect qualified call cost, not total call cost — these will be different numbers
- –Supplement with CRM offline imports where possible — AI-qualified call leads is a strong signal, not a substitute for downstream sales data