Part five

Paid and traditional: what advertising costs, and what the platform is not telling you

One honest chapter. No ad platform publishes what its advertising costs, the most rigorous experiment ever run on paid search found brand ads returning nothing, and the only channel with a published, non-negotiable price is the post office.

  • Zero

    Ad platforms that publish a benchmark cost per thousand or cost per click. Every figure you see is a third party's

    Checked 8 Aug 2026 1

  • +12%

    Meta's average price per ad, year on year, Q2 2026. The only first-party pricing signal that exists, and it is a rate of change

    Meta, 29 July 2026 2

  • Negative

    Average returns on paid search in the largest controlled experiment ever published on the question

    Econometrica, 2015 3

  • 2x

    The amount Google may spend above your daily budget on any given day, capped at 30.4 times daily across a month

    Google Ads Help 4

  • 26c / 32p

    The published floor price to put a leaflet through a door in the US and the UK. Within a few pence of each other

    USPS and Royal Mail 5

  • The $4,800 agency retainer does not exist

    We read the report it is attributed to. 494 respondents, and no dollar figures anywhere in it. 6

  • Google does not count view-through in "Conversions"

    Meta's reporting does include view windows. The two platforms are not counting the same thing. 7

  • The restaurant cost-per-lead everyone quotes is $3.16

    A "lead" there is a form fill. Not a cover, not a booking, not a person who turned up. 8

  • Nobody has compared organic and paid for a small local business

    Every effectiveness study we found measures national media for large brands. 9

Nobody who sells advertising will tell you what it costs.

We checked Google's and TikTok's own help documentation directly. Neither publishes a benchmark cost per click, cost per thousand or cost per result. Meta publishes only how much its prices moved.

The one first-party pricing signal that exists

Meta, year on year Impressions Price per ad
Q4 2025 (quarter) +18% +6%
Full year 2025 +12% +9%
Q1 2026 +19% +12%
Q2 2026 +14% +12%

That is a rate of change, globally, across every advertiser on the platform. It is not a price, and it cannot be turned into one. What it does tell you plainly: Meta's advertising got about 12 percent more expensive in the last year while it served about 14 percent more of it. 10

What Google will tell you about budgets

"The amount of your budget is entirely up to you." And the part worth knowing before you set one: "Your actual spend on any given day may exceed your average daily budget by up to 2 times," with a monthly ceiling of 30.4 times the daily figure. 11

So where do the benchmarks come from

One publisher, aggregating its own clients' accounts. The search figures, updated 1 June 2026: average cost per click $5.42 , cost per lead $66.69 , click-through rate 6.64 percent, conversion rate 8.18 percent. Source described as "thousands of our customers' advertising campaigns." No account count and no data date range are published. 12

The Facebook figures, updated 24 October 2025: traffic campaigns average $0.70 cost per click; lead generation campaigns $1.92 per click and $27.66 per lead. And the row every restaurant article quotes: Restaurants and Food, cost per lead $3.16 . 12

The gap between what a platform reports and what actually happened.

This has been measured properly, twice, at enormous scale, and both times by people with access to the data. The findings are uncomfortable and they are the best evidence in this chapter.

The eBay experiment

"Because search clicks and purchase behavior are correlated, we show that returns from paid search are a fraction of conventional non-experimental estimates. As an extreme case, we show that brand-keyword ads have no measurable short-term benefits. For non-brand keywords we find that new and infrequent users are positively influenced by ads but that more frequent users whose purchasing behavior is not influenced by ads account for most of the advertising expenses, resulting in average returns that are negative."

Blake, Nosko and Tadelis, Econometrica, January 2015 13

A series of large-scale field experiments run inside eBay, published in one of the most demanding journals in economics. Note the nuance almost every citation drops: it found a positive effect on new and infrequent users. The negative average comes from paying to reach people who were coming anyway.

And when Facebook checked its own methods

15 randomised controlled trials on Facebook, 500 million user observations, 1.6 billion impressions. The finding: "observational methods often fail to produce the same effects as the randomized experiments, even after conditioning on extensive demographic and behavioral variables." The gap runs in both directions, which is worse than a simple overstatement, because it means you cannot correct for it. 14

What each platform actually counts

  • Google. "View-through conversions are not included in the 'Conversions' column, only in the 'View-through conversions' and 'All conversions' columns." Default model is data-driven. Default windows: 30-day click, 1-day view, 3-day engaged view 15
  • Meta. Its reporting API exposes 1, 7 and 28 day click and view windows inside its action statistics 16

So the two platforms are not counting the same thing, and neither is counting what the eBay experiment measured.

What businesses spend, and the numbers that turn out not to exist.

Three of the most-quoted figures on marketing spend were checked at source. One of them is not in the document it is attributed to, and one of the studies does not appear to exist at all.

The figure Status What we found when we opened it
The average agency retainer is around $4,800 a month Not in the report Attributed to a 2026 agency benchmarks report. We read it. 494 respondents, fielded February to April 2026, collected through the publisher's own audience. It contains no dollar figures at all. The 2023, 2024 and 2025 editions all return 404. Unsupported by anything currently readable 17
The agency pricing survey Appears not to exist We retrieved the full blog index of the analyst it is attributed to. Nothing on agency pricing, agency fees or an agency survey 18
Marketing is 9 percent of revenue True, of large companies The cleanest spend benchmark available: 308 respondents, fielded 7 to 29 January 2026, 97 percent VP level or above. Marketing is "9.0% of revenues" and 9.6 percent of budgets. Overall spend grew 1.7 percent, digital grew 8.2 percent, traditional advertising is forecast to "decline a further 1.5%." Only 13 percent of respondents are under $10 million in revenue. A single-site restaurant is not in that sample 19
Restaurants spend 3 to 6 percent of revenue on advertising No sample, no date, no method Published by a restaurant technology company, credited to a name with no link. The same page carries "90 percent of restaurants pay for social media ads," credited to a different unnamed source. Neither has a sample, a country or a date. This is a worked example of a bad statistic, not a data point 20

The referral evidence, which is real

One properly conducted study on referred customers found they show "a higher contribution margin, though this difference erodes over time," "a higher retention rate" that persists, and that they "are more valuable in both the short and the long run." That is a genuine finding about the economics of word of mouth, and it is the only one in this chapter that comes from outside the marketing industry. 21

Print, post, radio and the local sponsorship nobody has ever measured.

Traditional channels have one advantage over everything digital in this chapter: some of them publish their actual prices, in public, and you cannot be charged more than that.

The published floor price for a leaflet through a door

Where Price The rules
United States $0.260 Flats up to 3.3 oz, effective 12 July 2026. Minimum 200 and maximum 5,000 pieces per postcode per day
United Kingdom 32p "Starting at 32p per item for unsorted mail," minimum 1,000 letters or 250 large letters. "As low as 25p" with high sort. No effective date is published, which is worth knowing

The two floors land within a few pence of each other. That is the only genuinely comparable, verifiable, first-party price in this entire chapter. 22

How direct mail is measured in the UK, and it is good

1,100 nationally representative households a month, a 28-day photographic diary where "the household co-ordinator takes a photo of each commercial mail item and records how it is interacted with, by whom, and when," with automated monitoring by a research agency and oversight by a technical committee. Nearly 400,000 mail journeys collected. It is funded by the postal service and the industry bodies, which is worth saying, and it is more transparent about its method than anything else in this chapter. 23

The effectiveness numbers, and who paid for them

Television and video. A study of £1.8 billion of spend across 141 brands reports full profit return of £5.61 per pound for television, £3.86 for online video and £4.11 across all media. It was published 24 April 2024, so it is 27 months old, and it is funded by the television industry's marketing body. 24

Radio. £2.50 short-term and £5.00 full-term return per pound, from 84 audio campaigns, published 2026. It is a re-cut of the same underlying dataset, funded by the radio industry's trade body. That is not independent corroboration, it is the same data with a different sponsor. 25

Six claims checked against the source.

And one honest admission at the end of them, because it is the question everybody actually wants answered and nobody has answered it.

The claim Verdict What the source actually says
The average cost per click on Facebook is X One agency's client book No ad platform publishes a benchmark cost per click or cost per thousand. We checked Google's and TikTok's own help documentation directly. Every figure in circulation is a third party aggregating its own customers' accounts, usually with no account count and no date range. 26
Restaurant leads cost about $3 on Facebook A form fill, not a cover The $3.16 figure is a cost per lead in a lead generation campaign. It is not a booking and not a person who arrived. The same table puts dentists at $76.71, which tells you the variation is mostly about how easy the form is. 27
Bidding on your own brand name protects you Measured, and it did not The largest controlled experiment published on paid search: "we show that brand-keyword ads have no measurable short-term benefits." Average returns across the programme were negative, because most of the spend reached people who were coming anyway. 28
The platform's conversion number tells you what worked Not the same number Google excludes view-through from its Conversions column; Meta's reporting includes view windows. And when Facebook compared 15 randomised trials against observational methods on 1.6 billion impressions, the methods "often fail to produce the same effects," in both directions. 29
The average agency retainer is $4,800 a month Not in the report We opened the 2026 report it is attributed to. 494 respondents, and no dollar figures anywhere in it. Earlier editions all return 404. 30
Traditional media is dead Overstated, and mostly unmeasured Traditional advertising spend is forecast to decline 1.5 percent in the coming year, which is a decline, not a collapse. Postal advertising has a published, verifiable floor price of about 26 cents or 32 pence per item. And local sponsorship has never been measured by anyone we could find, which is not the same as not working. 31

Questions people ask about this chapter

Every answer below is the chapter's own wording, with a link to the page it sits on.

What should an ad campaign cost me?

If you are going to spend money on ads, your own first month of data is a better benchmark than anybody else's table.

Read the full answer
Is a 3.16 cost per lead a booking?

It is not a booking, not a cover, not a person who walked through the door.

Read the full answer
Should I bid on my own business name?

The negative average comes from paying to reach people who were coming anyway.

Read the full answer
How can I test whether my ads are working?

The one test that costs nothing is to turn a campaign off for a fortnight and watch what happens to bookings.

Read the full answer
Does sponsoring the local team work?

If you sponsor the local football team, you are doing it on judgement, and there is nothing published that will tell you whether it works.

Read the full answer
Is organic or paid better for a small local business?

The honest bottom line of this chapter: there is no published study comparing organic and paid effectiveness like for like for a small local business.

Read the full answer

Sources

  1. Google Ads Help and TikTok Ads help centre, checked directly. 1st
  2. Meta Platforms, Q2 2026 results, 29 July 2026. 1st
  3. Blake, Nosko and Tadelis, Econometrica 83(1), January 2015. 1st
  4. Google Ads Help, average daily budget. 1st
  5. USPS Notice 123, effective 12 July 2026; Royal Mail Advertising Mail. 1st
  6. AgencyAnalytics, Agency Benchmarks Report 2026 . 2nd
  7. Google Ads Help and Meta Marketing API documentation. 1st
  8. LocaliQ and WordStream, updated 24 October 2025. 2nd
  9. Searched 8 August 2026. n/a
  10. Meta Platforms earnings releases, Q4 2025, Q1 2026 and Q2 2026. Denominator is always the prior-year period, global, all advertisers. 1st
  11. Google Ads Help, average daily budget documentation. Retrieved 8 August 2026. 1st
  12. LocaliQ and WordStream search benchmarks, updated 1 June 2026, and Facebook benchmarks across 20 industries, updated 24 October 2025. Self-selected client sample. 2nd
  13. Blake, Nosko and Tadelis, Consumer Heterogeneity and Paid Search Effectiveness: A Large Scale Field Experiment , Econometrica 83(1), January 2015, pp 155 to 174. Quote from the authors' NBER working paper w20171, 29 May 2014. 1st
  14. Gordon and others, comparing randomised experiments to observational methods on Facebook advertising. 1st
  15. Google Ads Help, conversion windows and attribution model documentation. 1st
  16. Meta Marketing API documentation, action attribution windows. 1st
  17. AgencyAnalytics, Agency Benchmarks Report 2026 , 494 respondents, Typeform survey of the publisher's own audience. 2nd
  18. SparkToro blog index, retrieved in full, 8 August 2026. n/a
  19. The CMO Survey, Spring 2026, fielded 7 to 29 January 2026, n=308, 14.6 percent response rate. 2nd
  20. Toast, restaurant marketing statistics page, credited to "RestaurantGrowth" and "MenuTiger". 2nd
  21. Published study of referred versus non-referred customers. 1st
  22. USPS Notice 123, effective 12 July 2026, and USPS EDDM documentation; Royal Mail Advertising Mail pricing. 1st
  23. JICMAIL, methodology documentation, retrieved 8 August 2026. Funded by Royal Mail, the DMA, Whistl, the IPA and ISBA. 1st
  24. Thinkbox, Profit Ability 2 , 24 April 2024. £1.8bn of spend, 141 brands. Only television, online video and the all-media average are published; per-channel figures for press, outdoor, display and search are not. 2nd
  25. Radiocentre, High Gain Audio , 2026, a re-cut of the same dataset, 84 audio campaigns. 2nd
  26. Google Ads Help and TikTok Ads help centre, checked 8 August 2026; LocaliQ and WordStream benchmarks. 1st/2nd
  27. LocaliQ and WordStream Facebook benchmarks, 24 October 2025. 2nd
  28. Blake, Nosko and Tadelis, Econometrica, January 2015. 1st
  29. Google Ads Help; Meta Marketing API; Gordon and others on Facebook advertising. 1st
  30. AgencyAnalytics, Agency Benchmarks Report 2026 . 2nd
  31. The CMO Survey, Spring 2026; USPS Notice 123 and Royal Mail; searched for sponsorship measurement 8 August 2026. 1st/2nd