Part two · The platforms · 08 of 08

Messaging and dark social: where recommendations happen and nobody has measured it

The place most recommendations actually happen: a message, a group chat, a forwarded link. It is the one part of this document where the honest headline is that nobody has measured it properly in over a decade.

  • 2014

    Date of the newest large-scale measurement of dark social with a stated survey sample

    RadiumOne 1

  • ~100

    Participants in the only modern study with a published method, over ten days

    SparkToro, April 2023 2

  • 100%

    Of visits from WhatsApp, Slack, Discord, TikTok and Mastodon in that study logged as "direct"

    SparkToro, April 2023 2

  • 90%

    Of UK online adults reached by WhatsApp. In the US it is around a third of adults

    Ofcom and Pew, 2025 3 , 4

  • 2020

    Year browsers stopped sending the page and campaign in a cross-site referrer, by default

    Chrome 85 5

  • The famous percentages are between eight and fourteen years old

    Every "X percent of sharing is dark" figure in circulation traces to 2014, 2016 or 2018, from vendors selling share buttons. 1

  • Google says Direct is a residual bucket

    Its own documentation lists missing UTMs, redirects, shorteners, offline documents and ad blockers as causes. 6

  • Meta's CEO says WhatsApp is not the US default

    "We're clearly not the primary messaging app in the United States at this point, iMessage is." 4

  • Nobody has validated "how did you hear about us"

    It is recommended everywhere. We found no published study testing whether the answers are accurate. 7

  1. RadiumOne, press release 1 December 2014 and undated report page; GetSocial, 2018. Vendor studies. 2nd
  2. SparkToro, Dark Social Falsely Attributes Significant Percentages of Web Traffic as "Direct" , 27 April 2023. 2nd
  3. Ofcom, Online Nation 2025 , 10 December 2025. 1st
  4. Pew Research Center, Americans' Social Media Use 2025 , 20 November 2025; Meta Q1 2025 earnings call, 30 April 2025. 1st
  5. Chrome for Developers and Chromium blink-dev, July 2020. 1st
  6. Google, GA4 Analytics Help. Undated. 1st
  7. Searched 8 August 2026. n/a

Every dark social statistic in circulation, with its date and its base.

These are the sources behind the numbers you will see quoted in 2026. Ages are calculated to 8 August 2026.

Figure Who produced it Date and age The sample What the percentage is a share OF
59% RadiumOne 2nd 1 Dec 2014
140 months 		Survey of 9,027 consumers across North America, UK, Europe, Australia and France, plus data from RadiumOne's own Po.st share buttons All online sharing activity in the US. The headline is US-specific while the survey spans several regions, and the US sub-sample is not disclosed. The two data sources are not separated
84% RadiumOne 2nd Page carries no date Stated elsewhere as 940 million users of sites carrying RadiumOne's sharing software Outbound sharing from publisher and marketer websites that had RadiumOne's tooling installed. Not all sharing
77% The Drum, reporting the same RadiumOne research 2nd Jun 2016
122 months 		As above Content shares from publisher and marketer sites. Same research programme as the 84 percent, different number
~33% Chartbeat 2nd 4 Dec 2014
140 months 		Chartbeat's publisher network. Pageview counts, network size and geography not stated External referral traffic , not shares. A completely different denominator from every RadiumOne figure, and routinely conflated with them
78% GetSocial 2nd 2018, no exact date
~100 months 		Ten client websites. Country, period, sector and share volume not disclosed Website shares measured through GetSocial's own tools. The same report separately gives 57 percent of referral traffic and 10 percent of overall traffic, three denominators in one document
100% SparkToro 2nd 27 Apr 2023
40 months 		Roughly 100 recruited participants, 1,113 visits, 10 days, mostly US and Canada Visits from each named app that carried no usable referrer. Not a share of all sharing. Messenger was 75 percent, Instagram DMs 30 percent
  1. RadiumOne via PR Newswire, 1 December 2014; RadiumOne report page, undated; The Drum, 9 June 2016. RadiumOne was folded into Rakuten and the underlying method was never published. 2nd
  2. Chartbeat Research, The State of Dark Social , 4 December 2014. 2nd
  3. GetSocial, Dark Social 2018 , undated page. 2nd
  4. SparkToro with Really Good Data, 27 April 2023. 2nd
  5. Searches run 8 August 2026. Retrieval failures stated above. n/a

Why it cannot be measured, and when that happened.

This is not a mystery. It is four documented changes, each with a date, and it started with browsers rather than with messaging apps.

  1. 01

    Chrome, 85, August 2020

    Chrome changed its default referrer policy to strict-origin-when-cross-origin , so "only the origin is sent in the Referer header of cross-origin requests." 13

  2. 02

    Firefox, 87, March 2021

    Mozilla did the same, and was explicit about the effect: it "will also trim path and query information for all cross-origin requests." 14

  3. 03

    Safari, ITP, March 2020

    Apple's Intelligent Tracking Prevention downgrades cross-site referrers to the origin and caps client-side cookies at seven days. A visitor who arrives, leaves, and returns three weeks later is a brand new direct visitor. 15

  4. 04

    Apple Link Tracking Protection, June 2023

    In Messages, Mail and Safari Private Browsing, "the extra information that some websites add to their URLs will be removed." 16

What that means in one sentence

Path-level and campaign-level referral analysis died in 2020 and 2021, in the browser, before anything a messaging app did. Even when a referrer does arrive, you get https://example.com/ and never the page or the campaign string.

What Apple has not published

Apple has published no list of which URL parameters Link Tracking Protection strips. We searched apple.com, its support site, its developer site and webkit.org and found nothing. Anyone telling you it removes fbclid and gclid but leaves utm_ parameters alone is reporting somebody's reverse-engineering, not Apple documentation. The 2023 announcement also does not name the operating system versions the feature shipped in. 16

And what is assumed rather than known

No messaging app publishes documentation on its referrer behaviour. Whether a referrer arrives depends on the app, on whether the link opens in an in-app browser or hands off to the system browser, on the operating system and on the version. There is no published matrix of this. A blanket statement that messaging apps never send a referrer is a generalisation from small tests, and the newest of those tests is from April 2023.

  1. Chrome for Developers, A new default Referrer-Policy for Chrome , 30 July 2020, and Chromium blink-dev Intent to Ship, 28 July 2020. 1st
  2. Mozilla Security Blog, Firefox 87 trims HTTP Referrers by default , 22 March 2021. 1st
  3. WebKit blog, Full Third-Party Cookie Blocking and More , 24 March 2020. 1st
  4. Apple Newsroom, 5 June 2023, and apple.com/privacy/features, undated. 1st

"Direct" does not mean people typed your address.

Google publishes two descriptions of the same channel. One is the plain English version. The other is the rule the system actually applies.

What Google says Direct is

"Direct is the channel by which users arrive at your site/app via a saved link or by entering your URL."

Google, GA4 default channel group documentation 17

What the rule actually is

"Source exactly matches '(direct)' AND Medium is one of ('(not set)', '(none)')."

Google, GA4 default channel group documentation 17

The description makes Direct sound like bookmarks and typed addresses. The rule makes it the bucket for everything that arrives with no source information at all.

Google's own list of what lands in there

  • "When links to your site lack UTM parameters... traffic source information is lost" 18
  • "Redirects and URL shorteners can also strip tracking details" 18
  • Links "from offline documents such as PDFs and Word documents" 18
  • "Ad blockers can interfere with tracking cookies used to identify the traffic origin" 18
  1. Google, [GA4] Default channel group , Analytics Help. The page carries no publication or revision date; Google does not date Analytics Help articles. Retrieved 8 August 2026. 1st
  2. Google, [GA4] Understand (direct) / (none) traffic , Analytics Help. Also undated. Retrieved 8 August 2026. 1st

WhatsApp is a mass channel in the UK and a minority channel in the US.

Both sides of that are well sourced, one by a regulator and one by an independent research centre. The bases are different, so they are stated here rather than blended.

Measure Figure Base
WhatsApp reach, UK 90% Of UK online adults 18+
Facebook and Messenger combined, UK 93% Of UK online adults 18+
WhatsApp use, UK, aged 16 to 24 81% Of UK social media users
WhatsApp use, UK, aged 75+ 78% Of UK social media users
WhatsApp use, US adults ~33% Of all US adults
WhatsApp use, US adults, 2021 23% Of all US adults
WhatsApp as a news source, UK 24% Of UK adults who use social media for news, down from 28% in 2024

The UK figures come from two different Ofcom studies with two different bases, online adults and social media users. They are not comparable with each other and must never be presented as if they were. 19 , 20 , 21

The sentence that decides a US strategy

"While we have more than 100 million people use WhatsApp in the United States, we're clearly not the primary messaging app in the United States at this point, iMessage is."

Mark Zuckerberg, Meta Q1 2025 earnings call, 30 April 2025 22

On the same call: "WhatsApp now has more than 3 billion monthly actives." That is the cleanest first-party WhatsApp user number available, and it exists only because Zuckerberg said it aloud. 22

  1. Ofcom, Online Nation 2025 , 10 December 2025. Ipsos iris passive panel, UK internet users 18+, May 2025, universe 49.1 million. 1st
  2. Ofcom, Adults' Media Use and Attitudes 2026 , 2 April 2026. 7,533 UK adults 16+, fieldwork 29 September to 28 November 2025. 1st
  3. Ofcom, News consumption in the UK 2025 , 21 July 2025. 1st
  4. Meta Platforms, Q1 2025 earnings call transcript, 30 April 2025. 1st
  5. Meta Platforms, Q2 2026 results, 29 July 2026; Pew Research Center, Americans' Social Media Use 2025 , 20 November 2025, 5,022 US adults, fieldwork February to June 2025. 1st

Answering is free. Broadcasting is not.

There are two WhatsApp products for businesses and they are not the same thing. The pricing model changed on 1 July 2025, and the change is the whole story.

Two products

The WhatsApp Business app is "built with the small business owner in mind." The WhatsApp Business Platform is the API: "This documentation is intended for developers using our APIs." For an independent venue, the app is the relevant product and the platform is not. Note that neither Meta page states the app's price, so we are not asserting one. 24

What the per-message pricing actually charges for

Meta's own documentation: "Conversation-based pricing is deprecated. It was replaced with per-message pricing on July 1, 2025." What is free, in Meta's words: 25

  • "All non-template messages are free" inside an open customer service window
  • "Utility templates delivered within an open customer service window are free"
  • "All messages are free for 72 hours" inside an open free entry point window
  • Service conversations free for all businesses since 1 November 2024

Marketing templates are the paid category, and marketing templates are the thing a venue would actually want to send.

What that pricing model is telling you

Meta has made replying free and broadcasting paid. Answering a customer who messaged you first costs nothing, indefinitely, inside the service window. Sending an unprompted marketing template to a list costs money per message. The pricing is the product design: the channel is built for conversations that the customer starts. 25

Before you broadcast anything, in the UK

The ICO's guidance, dated 28 April 2026, is that unsolicited electronic mail marketing to individual subscribers needs consent or a narrow soft opt-in, while "you can send unsolicited electronic mail marketing to corporate subscribers without consent or a soft opt-in." The soft opt-in requires details collected during negotiations for a product or service, marketing only for similar products, and an opt-out both at collection and in every message. 26

That page discusses email and SMS. It does not state whether messaging apps fall inside the definition of "electronic mail," so we are not going to tell you that they do or that they do not. The safe reading: treat a WhatsApp marketing broadcast to consumers exactly as you would an SMS one. 26

In the US, marketing texts are governed by the TCPA and need prior express written consent. The detail moved during 2025 and we could only source that movement from law firm commentary rather than the primary orders, so we are not going to summarise it. Anyone running US SMS marketing needs a US lawyer, not this document.

  1. WhatsApp Business product page, last modified 22 June 2026, and Meta for Developers, About the WhatsApp Business Platform , updated 4 August 2026. 1st
  2. Meta for Developers, Pricing on the WhatsApp Business Platform , retrieved 8 August 2026. 1st
  3. ICO, How do we comply with the PECR electronic mail marketing rules? , page dated 28 April 2026. 1st

The Updates tab, and the only two money figures Meta has given.

WhatsApp is becoming a place businesses can pay to appear. What has actually been published about that is thinner than the coverage suggests.

The Updates tab, and what the 1.5 billion figure covers

Meta and WhatsApp both published, on 16 June 2025, that the Updates tab is "used by 1.5 billion people a day" globally. That is the tab, which contains both Status and Channels, not Channels on their own. The two are constantly conflated in trade coverage. Announced at the same time: Channel Subscriptions, Promoted Channels, and ads in Status, all confined to that tab, "away from your personal chats." 27

As of the Q3 2025 call, the ad product was still being rolled out: "we are continuing to gradually introduce ads and expect to complete the roll out next year." 28

The only two money figures Meta has given

Click-to-WhatsApp ads "grew revenue 60% year-over-year in Q3" 2025. That is a growth rate on a base Meta does not disclose. And in Q2 2026, "quarterly Family of Apps other revenue reached $1 billion and grew 73% year-over-year, driven primarily by WhatsApp paid messaging and subscriptions revenue." That line bundles WhatsApp paid messaging with Meta Verified and other items. It is not WhatsApp revenue. 29 , 30

For scale on that line: Family of Apps other revenue was $510 million in Q1 2025 and $690 million in Q3 2025. It is growing fast and it is a small fraction of Meta's total revenue. 29 , 30

Two more first-party numbers worth having

From the Q2 2026 call, 29 July 2026: "WhatsApp just hit an all-time messaging record, peaking at 30 million messages sent per second during the World Cup Final," and "there are already more than 1 million businesses" using Meta Business Agents on WhatsApp and Messenger. 30

  1. WhatsApp Blog and Meta Newsroom, Helping You Find More Channels and Businesses on WhatsApp , both 16 June 2025. 1st
  2. Meta Q3 2025 earnings call, 29 October 2025. 1st
  3. Meta Q1 2025 and Q3 2025 earnings calls. 1st
  4. Meta Q2 2026 earnings call, 29 July 2026. 1st

Instagram says two different things about sends.

"Sends per reach" is the single most repeated piece of Instagram advice in 2026. Here is where it comes from, and what Instagram's own written documentation says instead.

Where the sends doctrine comes from

Adam Mosseri, head of Instagram, in a video post, as reported: "sends are slightly more important for unconnected content." That is the source. It is a spoken remark in an informal video, with no number attached to it and no measurement window. 31

We could not read the original. Instagram's own page for that post returned a robots restriction to our fetcher on 8 August 2026, so we could not check the quote against Mosseri's own words or confirm the post date. What is in this document is reported speech, from a trade publication, dated 22 January 2025. 31

Messenger, for completeness

"Messenger is also used by more than a billion people each month," and "there are now as many messages sent each day on Instagram as there are on Messenger." Both Zuckerberg, 30 April 2025. Meta publishes no Messenger business-messaging metric at all, in any filing or call we checked. 33

What Instagram has actually written down

Instagram document Date Signals it names
Instagram Ranking Explained , by Adam Mosseri 31 May 2023 Seconds spent, comment, like, share, profile photo tap
Meta system card, Instagram Feed Ranking Updated 23 Feb 2022 Comment, like, save, profile photo tap
Instagram creator guidance on reach Undated Liking, commenting, sharing, saving

None of the three mentions sends. Instagram's formal, dated, first-party documentation and the head of Instagram's video remarks describe different things, and the documentation has not been updated to match. 32

Email forwarding

Unmeasured. Every source we found bundles "email and instant messaging" together and gives no separate figure for email. Not small, not large. Unmeasured.

  1. Adam Mosseri, quoted by Social Media Today, 22 January 2025. The original Instagram post could not be retrieved on 8 August 2026. 2nd
  2. Instagram, Instagram Ranking Explained , 31 May 2023; Meta, Instagram Feed Ranking system card, updated 23 February 2022; Instagram creator guidance, undated. 1st
  3. Meta Platforms, Q1 2025 earnings call, 30 April 2025. 1st

Business messaging in the default texting app.

RCS is the successor to SMS and it is the channel with the largest reach of anything in this chapter, because it lives in the app that ships with the phone. Getting a business into it is another matter.

What is actually live

Apple's own support page, dated 11 May 2026: "To use RCS, you need iOS 18 and a cellular plan from a carrier that supports RCS messaging." Apple describes RCS as a carrier-provided service, which is the constraint that governs everything else. Availability is per carrier, not per device. 34

End-to-end encrypted RCS between iPhone and Android shipped with iOS 26.5 on 11 May 2026, and the GSMA's own announcement carries the same conditional: it arrives "once enabled by supporting mobile network operators," with a padlock icon marking encrypted conversations. 35

And what a business would have to do

Branded, verified business messaging over RCS runs through a partner programme: "To apply to become an RCS for Business partner, fill out the partner registration interest form." That route is for developers, aggregators and carriers. The honest answer for an independent venue is that verified RCS business messaging is not something you sign up for yourself. 36

  1. Apple Support, Turn on RCS messaging on your iPhone , page dated 11 May 2026. 1st
  2. GSMA Newsroom, iOS 26.5 brings E2EE for RCS , 11 May 2026. 1st
  3. Google, RCS Business Messaging: Get started , Google Business Communications developer documentation. Page carries no date. Retrieved 8 August 2026. 1st

Six claims checked against the source.

This chapter has the highest failure rate in the document. That is the finding, not a gap in the research.

The claim Verdict What the source actually says
84 percent of sharing happens on dark social Undated, narrow base The page carrying that figure has no date and the base is outbound sharing from websites that had the vendor's own share buttons installed. The same research programme was reported at 77 percent elsewhere and 63 percent elsewhere again. The vendor no longer exists under that name and the method was never published. 37
Dark social is a third of your traffic Different denominator The one-third figure is Chartbeat's, from December 2014, and it measures external referral traffic across a publisher network, not shares. It is routinely quoted alongside the share-of-shares figures as if they measured the same thing. 38
There is up-to-date research on this There is not Nothing credible from 2025 or 2026. The newest study with a stated method is April 2023, with roughly 100 participants over ten days. The rest is from 2014, 2016 and 2018, produced by companies selling share buttons. 39
Everything in Direct is word of mouth Inference, not documentation Google's list of what lands in Direct is missing UTMs, redirects, shorteners, offline documents and ad blockers. It does not name messaging apps. The link to private sharing is a reasonable inference from the 2023 test, and it is not something Google publishes. 40
WhatsApp is where your customers are True in the UK, not in the US 90 percent of UK online adults, per the regulator. Around a third of US adults, per Pew. Meta's own CEO: "we're clearly not the primary messaging app in the United States at this point, iMessage is." 41
Just ask people how they heard about you Unvalidated, and still the best available We found no published study testing the accuracy of self-reported attribution. The most-cited article on the subject runs no experiment. The known biases are plausible and unmeasured. It remains the only instrument that reaches the channels analytics cannot see, which is an argument for using it and not for trusting it precisely. 42
  1. RadiumOne report page, undated; The Drum, June 2016. 2nd
  2. Chartbeat Research, 4 December 2014. 2nd
  3. SparkToro, 27 April 2023, and searches run 8 August 2026. 2nd
  4. Google, GA4 Analytics Help, undated. 1st
  5. Ofcom, 10 December 2025; Pew Research Center, 20 November 2025; Meta Q1 2025 earnings call, 30 April 2025. 1st
  6. Searched 8 August 2026. n/a

Questions people ask about this chapter

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

Can dark social be measured?

It is four documented changes, each with a date, and it started with browsers rather than with messaging apps.

Read the full answer
Will a UTM link survive being shared in a message?

It does not survive somebody retyping your domain, a share sheet that rewrites the link, an app that strips parameters, or a redirect that drops the query.

Read the full answer
Does Direct traffic mean people typed my address?

The rule makes it the bucket for everything that arrives with no source information at all.

Read the full answer
Does it cost anything to reply to customers on WhatsApp?

Answering a customer who messaged you first costs nothing, indefinitely, inside the service window.

Read the full answer
Should I broadcast marketing messages on WhatsApp?

The safe reading: treat a WhatsApp marketing broadcast to consumers exactly as you would an SMS one.

Read the full answer
Are sends really the most important Instagram signal?

Instagram's formal, dated, first-party documentation and the head of Instagram's video remarks describe different things, and the documentation has not been updated to match.

Read the full answer

Sources

  1. RadiumOne, press release 1 December 2014 and undated report page; GetSocial, 2018. Vendor studies. 2nd
  2. SparkToro, Dark Social Falsely Attributes Significant Percentages of Web Traffic as "Direct" , 27 April 2023. 2nd
  3. Ofcom, Online Nation 2025 , 10 December 2025. 1st
  4. Pew Research Center, Americans' Social Media Use 2025 , 20 November 2025; Meta Q1 2025 earnings call, 30 April 2025. 1st
  5. Chrome for Developers and Chromium blink-dev, July 2020. 1st
  6. Google, GA4 Analytics Help. Undated. 1st
  7. Searched 8 August 2026. n/a
  8. RadiumOne via PR Newswire, 1 December 2014; RadiumOne report page, undated; The Drum, 9 June 2016. RadiumOne was folded into Rakuten and the underlying method was never published. 2nd
  9. Chartbeat Research, The State of Dark Social , 4 December 2014. 2nd
  10. GetSocial, Dark Social 2018 , undated page. 2nd
  11. SparkToro with Really Good Data, 27 April 2023. 2nd
  12. Searches run 8 August 2026. Retrieval failures stated above. n/a
  13. Chrome for Developers, A new default Referrer-Policy for Chrome , 30 July 2020, and Chromium blink-dev Intent to Ship, 28 July 2020. 1st
  14. Mozilla Security Blog, Firefox 87 trims HTTP Referrers by default , 22 March 2021. 1st
  15. WebKit blog, Full Third-Party Cookie Blocking and More , 24 March 2020. 1st
  16. Apple Newsroom, 5 June 2023, and apple.com/privacy/features, undated. 1st
  17. Google, [GA4] Default channel group , Analytics Help. The page carries no publication or revision date; Google does not date Analytics Help articles. Retrieved 8 August 2026. 1st
  18. Google, [GA4] Understand (direct) / (none) traffic , Analytics Help. Also undated. Retrieved 8 August 2026. 1st
  19. Ofcom, Online Nation 2025 , 10 December 2025. Ipsos iris passive panel, UK internet users 18+, May 2025, universe 49.1 million. 1st
  20. Ofcom, Adults' Media Use and Attitudes 2026 , 2 April 2026. 7,533 UK adults 16+, fieldwork 29 September to 28 November 2025. 1st
  21. Ofcom, News consumption in the UK 2025 , 21 July 2025. 1st
  22. Meta Platforms, Q1 2025 earnings call transcript, 30 April 2025. 1st
  23. Meta Platforms, Q2 2026 results, 29 July 2026; Pew Research Center, Americans' Social Media Use 2025 , 20 November 2025, 5,022 US adults, fieldwork February to June 2025. 1st
  24. WhatsApp Business product page, last modified 22 June 2026, and Meta for Developers, About the WhatsApp Business Platform , updated 4 August 2026. 1st
  25. Meta for Developers, Pricing on the WhatsApp Business Platform , retrieved 8 August 2026. 1st
  26. ICO, How do we comply with the PECR electronic mail marketing rules? , page dated 28 April 2026. 1st
  27. WhatsApp Blog and Meta Newsroom, Helping You Find More Channels and Businesses on WhatsApp , both 16 June 2025. 1st
  28. Meta Q3 2025 earnings call, 29 October 2025. 1st
  29. Meta Q1 2025 and Q3 2025 earnings calls. 1st
  30. Meta Q2 2026 earnings call, 29 July 2026. 1st
  31. Adam Mosseri, quoted by Social Media Today, 22 January 2025. The original Instagram post could not be retrieved on 8 August 2026. 2nd
  32. Instagram, Instagram Ranking Explained , 31 May 2023; Meta, Instagram Feed Ranking system card, updated 23 February 2022; Instagram creator guidance, undated. 1st
  33. Meta Platforms, Q1 2025 earnings call, 30 April 2025. 1st
  34. Apple Support, Turn on RCS messaging on your iPhone , page dated 11 May 2026. 1st
  35. GSMA Newsroom, iOS 26.5 brings E2EE for RCS , 11 May 2026. 1st
  36. Google, RCS Business Messaging: Get started , Google Business Communications developer documentation. Page carries no date. Retrieved 8 August 2026. 1st
  37. RadiumOne report page, undated; The Drum, June 2016. 2nd
  38. Chartbeat Research, 4 December 2014. 2nd
  39. SparkToro, 27 April 2023, and searches run 8 August 2026. 2nd
  40. Google, GA4 Analytics Help, undated. 1st
  41. Ofcom, 10 December 2025; Pew Research Center, 20 November 2025; Meta Q1 2025 earnings call, 30 April 2025. 1st
  42. Searched 8 August 2026. n/a