Skip to content

Benchmarks measured on real accounts.

Most paid-media benchmarks are survey data. Someone asks a few hundred marketers what they pay, averages the answers, and publishes a number nobody can trace. These are different. Each report is read out of live accounts I hold credentials on, in a fixed window, and published under rules that survive an adversarial read: no client named, N stated next to every figure, and ratios rather than revenue.

Seven reports published  ·  Refreshed on a schedule, not rolling  ·  N next to a figure is how many accounts it comes from

Home Services Cost Per Lead by Trade

What a Google Ads lead cost the home-services businesses I manage, trade by trade, July to September 2026. The typical one with tracking I trust paid $163 (N = 9), and the roofer paid the most. Two accounts that over-counted in July mostly stopped counting taps on their phone number as leads, and their cost per lead went up (N = 2). The higher figure is the honest one.

  • 11 Businesses (N)
  • $163 Typical cost per lead (N = 9)
  • 2 of 11 Tracking too broken to price (N = 11)

Q3 2026 edition, 1 July to 30 September. The July 2026 figures stay on the page, marked superseded.

Read the CPL report

Google Ads Benchmarks: 33 Live Accounts

What clicks and leads cost across 33 Google Ads accounts I manage in healthcare, home services, local services and ecommerce. Google listed the searches behind 52% of spend for the typical account (N = 11), and of the spend it listed, the typical account put 30.7% into searches with two or more clicks and no conversions (N = 25). 23 of 31 lost more showings to budget than to Ad Rank (N = 31).

  • $8.53 Typical cost per click (N = 33)
  • 52% Of spend with its searches listed (N = 11)
  • 23 of 31 Lost more to budget than rank (N = 31)

Q3 2026 edition, 1 July to 30 September. Refreshed quarterly.

Read the Google Ads benchmarks

Are Google Ads Clicks Getting More Expensive?

The same 12 local businesses, Google search ads, every summer from 2023 to 2026. 11 of the 12 pay more per click now, and the typical one pays 2.27x as much, so the same budget buys about 44 clicks for every 100 it bought in 2023. Ten accounts without that long history were flat over the last year.

  • 11 of 12 Pay more per click than in 2023 (N = 12)
  • 2.27x Typical change, 2023 to 2026 (N = 12)
  • 76% Typical rise in the last year (N = 12)

Fixed window, summers 2023 to 2026. Refreshed quarterly.

Read the CPC trend

Healthcare Google Ads Cost Per Lead

What a Google Ads lead cost eight healthcare practices I manage, from specialist practices to general care. A lead here is mostly a phone call, so cost per lead mostly measures which calls each practice counts. All eight ran search ads only, and each lost more than 28% of its possible ad showings to budget.

  • $39.54 Typical cost per lead (N = 6)
  • 100% Calls as share of leads, typical practice (N = 8)
  • 8 of 8 Ran search ads only (N = 8)

Q3 2026 edition, 1 July to 30 September. Refreshed quarterly.

Read the healthcare report

Meta Ads CPM by Placement

What Facebook and Instagram ads cost across 18 Meta ad accounts I manage, mostly local service businesses, by placement and by what each ad set asked Meta for. Marketplace sold cheap views and drew the weakest clicks. Half the accounts never asked Meta for a lead or sale. The two cheapest website leads came from accounts with 40% or more of spend on Audience Network, where the click numbers look wrong, so I trust them least.

  • $6.03 Per 1,000 views, all spend (N = 18)
  • 0.27% Of Marketplace views clicked (N = 12)
  • $89.33 Typical website cost per lead (N = 6)

Q3 2026 edition, 1 July to 30 September. Refreshed quarterly.

Read the Meta CPM report

AI Answers Field Report

How often AI assistants name a business, and how often they get its facts wrong, from 4,285 graded answers about 9 businesses. Asked by name, the answers named the business 90.2% of the time. On problem questions that named nobody, 2.6%. Pooled across businesses, with the per-business range beside every rate.

  • 90.2% Named when asked by name (9 businesses)
  • 2.6% Named on problem questions
  • 6 of 9 Never named on a problem question

Q4 2026 edition. A new edition ships once at least three new businesses join the pool.

Read the field report

The Ecommerce Tracking Accuracy Benchmark

How far the ad platform, the analytics import and the store disagree on the same purchases, in online stores I manage. On one store, three purchase tags disagreed by 30.1% over the same window, and the account bid toward the one that read highest. On two of three stores, part of what the Conversions column counted was add-to-cart or checkout starts, not sales.

  • 30.1% Gap between 3 tags, one store (N = 1)
  • 16.6 to 17.2% Of conversions were cart or checkout events, 2 of 3 stores (N = 3)
  • 99.4 to 99.8% “All conversions” not purchases (N = 3)

Measured 13 May to 10 August 2026. N = 3 on the ads side, N = 1 on the full reconciliation.

Read the accuracy benchmark

What a number
has to survive.

These five rules are why the reports can be cited. They also cost real findings: several figures I wanted to publish were cut because a sector descriptor or an exact volume count would have identified the account.

  1. 01

    My book, not a survey

    Every figure comes from an account I have managed, read through the platform API in the window stated, or from AI answers I graded myself on locked question panels. No client is ever named. Nothing is sourced from a vendor report, an industry average, or a survey of what marketers say they spend.

  2. 02

    N is stated next to the number

    A finding measured on one account is labelled N = 1 on the page, next to the figure, not in a footnote. Several of the most striking numbers in these reports are single-account observations, and they are marked as such so you can weight them yourself.

  3. 03

    No client is named

    Accounts appear as codes with a broad group or trade, or are pooled with no per-business figure at all. Descriptors stay general on purpose, because a descriptor precise enough to be interesting is often precise enough to identify the account. Where a group would hold a single recognizable business, it is merged into a broader one or left out.

  4. 04

    Ratios, not revenue

    Counts, percentages and ratios get published. Spend, revenue and order values do not. A cost per lead is a unit price and it is publishable in aggregate. What a given brand turns over is theirs.

  5. 05

    The window is fixed, not rolling

    A rolling window quietly rewrites the number every time someone loads the page, which makes the report impossible to cite. These use a fixed window and get refreshed on a schedule, with the previous figures marked superseded rather than deleted.

What lands next.

Two more are in measurement. A seasonal efficiency report on what the fourth quarter does to return on ad spend, which starts from the finding that on the account I have measured across two of them, the highest-volume quarter was also the weakest one. And a home and furniture brand report covering median CPC, Shopping share of spend and Performance Max mix by spend band. Neither is published until the sample is large enough to state an N I am willing to defend.

The first finding out of the seasonal work is already written up: what three years of purchase-only data say about the fourth quarter.

I will tell you which
of your numbers are real.

Add me as a read-only user and I'll write the audit myself: which of your reported numbers describe money and which describe carts, 5 to 7 pages, back within 24 hours of access. No call needed. If you'd rather talk first, a call is open too.

Get a free ads audit Or book a strategy call hi@connercrowe.com Read the Tracking Stack

Ask me anything about working together.

Every word here is from answers I wrote. If yours isn't covered, send it and I'll reply myself.

People usually ask

  • How much do you charge?
  • Is the audit really free?
  • How is this different from an agency?
  • Can you get my business mentioned in ChatGPT?