The most emotional and thought-provoking piece for me was the panel session ‘Molly vs The Machines: What Should the Industry Do Next?’, where Ian Russell shared his very personal story of how his 14-year-old daughter Molly became the victim of social media algorithms that reinforced her vulnerabilities, leading to her death by suicide in 2017.

Eight years on, the Molly Rose Foundation’s own research (August 2025) found suicide and self-harm content still being recommended “at industrial scale” on TikTok and Instagram: 97% of the Reels and 96% of the TikTok videos algorithmically served to an account registered as a 15-year-old girl were harmful. Once again, trust and governance came to the fore.

With the UK announcing a ban on social media for under-16s (Prime Minister Keir Starmer, 15 June 2026, covering Snapchat, TikTok, YouTube, Instagram, Facebook and X, though not WhatsApp, Signal or YouTube Kids, and not due to come into force until spring 2027) the panel saw this as insufficient protection. Like putting a plaster over too big a wound. It gives platforms an excuse to shed some of the responsibility and skip the investment in restriction, monitoring and safeguarding, on the logic that those most at risk from the content will no longer be there, if the ban works as intended. But kids are resourceful and digitally savvy enough to find ways around a ban. That makes their activity more hidden from parents, not less, than the alternative of building in-platform safeguarding measures. It arguably heightens the risk rather than removing it.

And the panel’s scepticism about platforms marking their own homework isn’t a hunch. Ofcom’s own regulatory record backs it up. In March 2026, Ofcom issued four demands to six of the UK’s most-used platforms, including effective age enforcement and safer algorithmic feeds. By its May 2026 update, Meta, Snap and Roblox had all agreed to new protections. TikTok and YouTube had not: both declined to commit to any significant change to their recommendation feeds, maintaining their feeds are already safe, which prompted Ofcom to issue TikTok with a legally binding information request.

Meanwhile, more than a third (35%) of 11 to 17-year-olds report encountering harmful content simply by scrolling their main feed, against just 1% who went looking for it. Algorithms, not searches, are how children find harm online. That is precisely the panel’s point: the recommendation engine is the product, not a neutral pipe.

The reason this content is present is because it gains a huge amount of attention, and that attention is rewarded financially: advertisers pay for it. In reality, we know brands would not want to be associated with detrimental content. Advertisers want meaningful attention, not harmful attention. But right now you cannot differentiate between the two, so both are rewarded equally. This isn’t a hypothetical. The Molly Rose Foundation’s research found advertising, including from major UK fashion retailers, fast food brands and universities, running adjacent to harmful material for roughly one in every ten posts watched. That is advertising revenue, literally, funding the distribution of content that recommends self-harm to children.

The panel argued this needs to change. Teen time spent should not be a success measure. We should lead with safety by design, not incentivise the behaviour we say we don’t want. There is early recognition of this outside the room too: as teen-harm litigation puts platform design under scrutiny in the US, analysts are starting to flag that media plans still tied to time-spent KPIs may not hold up to the same scrutiny for much longer.

Unfortunately the solution isn’t as simple as “advertisers should just make a stand”. The intention is right, but two recent legal moves complicate it. GARM, the last serious cross-industry brand safety body, was dissolved in 2024 after X’s antitrust suit against it. In April 2026, the FTC went further, taking action against WPP, Publicis and Dentsu, alleging they had unlawfully colluded to impose common brand safety standards across the industry. So a coordinated, industry-wide advertiser boycott of harmful content isn’t just a courage problem now, it’s a live legal one, at least in

the US*.

The Conscious Advertising Network made exactly this case at Cannes Lions last month, calling on advertisers to refuse to fund addictive design and demand safety by design from Meta and TikTok, but that call increasingly has to be answered brand by brand and through accredited, independent standards, not through advertisers sitting in a room agreeing a shared line.

There are social platforms offering a genuinely different model. BlueSky (42 million registered users and growing fast) and W Social, an 18-plus, identity-verified platform built on the same open protocol, are both attracting usage on the promise of a healthier environment. However, both are currently ad-free with no monetisation model live at all, so there is no ad spend option to redirect there yet, safer or not. Growing usage is real. Proof that safety earns advertisers a premium is still missing, because neither platform has built the market where that premium could be paid.

The solution keeps circling back to trust and governance, and it needs all four players in the room together: government, regulator, advertisers and platforms. That is a deliberately different shape to the industry-only alliance that got GARM into trouble. A standard convened by government, or built through powers Ofcom already has, does not carry the antitrust risk that advertisers agreeing a shared line among themselves does. It can still deliver the same outcome: platforms that can prove, independently, what they recommend and why; advertisers who can tell the difference between attention that is meaningful and attention that is harmful, and fund the former without needing a cartel to hide behind; and a regulator willing to use the powers it has, as Ofcom is now starting to, rather than accept “our feed is already safe” as an answer. Until that differentiation exists, the ban is a plaster, the business model is the wound, and the industry that funds it is implicated whether it means to be or not.

---

Anita Caras is founder and principal consultant at Caras Consulting, an independent research and insight consultancy specialising in marketing measurement and effectiveness.

*NB - Whilst I have quoted US examples, similar legal powers exist in the UK, should the CMA choose to use them. The Competition Act 1998’s Chapter I prohibition covers agreements, decisions by trade associations, and “concerted practices” (including informal, nod-and-wink coordination) that restrict competition, which is the same broad category the FTC used against WPP, Publicis and Dentsu.

**Sources**

- Molly Rose Foundation, *Suicide and Self-Harm Content Still Recommended “At Industrial Scale” by TikTok and Instagram, Eight Years After Molly’s Death* (August 2025): 644 videos analysed on accounts registered as a 15-year-old girl; 97% of Instagram Reels and 96% of TikTok videos algorithmically recommended were harmful; advertising (major UK fashion retailers, fast food brands, universities) appeared adjacent to harmful content roughly once in every ten to 9.5 posts watched. https://mollyrosefoundation.org/suicide-and-self-harm-content-still-recommended-at-industrial-scale-by-tiktok-and-instagram-eight-years-after-mollys-death/

- NPR, *Britain will ban under-16s from social media apps, including TikTok and YouTube* (15 June 2026): PM Keir Starmer’s announcement; platforms covered (Snapchat, TikTok, YouTube, Instagram, Facebook, X); WhatsApp, Signal and YouTube Kids excluded; implementation expected spring 2027. https://www.npr.org/2026/06/15/nx-s1-5858644/britain-social-media-ban

- Ofcom, *Update: tech firms’ responses to our call for action to protect children* (Project Mercury, 21 May 2026): March 2026 demands to six platforms; Meta, Snap and Roblox committed to new protections; TikTok and YouTube declined to commit to recommendation-feed changes, prompting a legally binding information request to TikTok; 35% of 11 to 17-year-olds encountered harmful content while scrolling their main feed, versus 1% who searched for it. https://www.ofcom.org.uk/siteassets/resources/documents/online-safety/protecting-children/project-mercury/update_tech-firms-responses-to-our-call-for-action-to-protect-children.pdf

- Conscious Advertising Network, Cannes Lions session *A Sense of Impending Doomscrolling: Rebooting AI and Social Media* (June 2026): call for advertisers to refuse to fund addictive design and demand safety by design from platforms. https://www.consciousadnetwork.com/

- AdExchanger, *The End of GARM Is a Reset, Not a Setback*, and Federal Trade Commission, press release (April 2026), *FTC Takes Action to Restore Competition in the Digital Advertising Ecosystem*: GARM dissolved in 2024 following X Corp’s antitrust suit; FTC action against WPP, Publicis and Dentsu alleging unlawful collusion on brand safety standards. https://www.adexchanger.com/data-driven-thinking/the-end-of-garm-is-a-reset-not-a-setback/

https://www.ftc.gov/news-events/news/press-releases/2026/04/ftc-takes-action-restore-competition-digital-advertising-ecosystem

- Business of Apps, *Bluesky Revenue and Usage Statistics* (2026): 42.3 million registered users; ad-free, no monetisation tools live as of 2026. https://www.businessofapps.com/data/bluesky-statistics/

- W Social, platform overview: 18-plus, identity verification via the W Identity app, built on the open AT Protocol. https://wsocial.eu/ · https://cybernews.com/tech/europe-social-media-w/

A growing share of people no longer browse their way to a decision. They ask a machine. According to Ofcom’s Online Nation 2025, 54% of UK adults now use AI tools like ChatGPT, Copilot, Gemini or Claude; up from 31% a year earlier. Around 30% of searches return an AI overview, and 53% of adults say they see those summaries often. When the answer arrives pre-assembled, the human doesn’t scroll the options. The machine has already chosen.

I’ve spent 30+ years measuring where human attention goes. What’s keeping me up now is what happens to attention when a machine sits between the person and the answer.

Here’s the shift. When discovery happens inside a chat interface, the currencies we’ve spent a decade standardising simply stop applying; no impression, no viewability, no dwell, no attention-second. The question every marketer will soon be asking is: were we recommended, and who is measuring it? Right now, that question has no neutral answer.

Being seen is not the same as being recommended.

And this is a growing concern. EY’s global study of more than 850 consumer-products executives (May 2026) found that 47% believe influencing algorithmic recommendations will be essential to staying competitive within five years, but only 21% think they can do it today. That gap is what we need to focus on. The industry can feel the currency moving and has no agreed way to trade in it.

So a cottage industry has rushed to fill the void. Generative Engine Optimisation, Answer Engine Optimisation, GEO, AEO, take your pick of acronym. The intent is reasonable enough: help brands understand their visibility inside AI answers. The problem is the proof. These tools are overwhelmingly vendor-built and self-reported, and not one of them carries third-party accreditation; no MRC, no independent audit to validate what it claims.

I’ll be blunt, because this is the part the industry would rather not say out loud. If the platform shaping the recommendation also measures the recommendation, that’s not measurement, that’s marketing. We’ve seen this play out before. Marking your own homework is not a new problem. It has simply found a new home which is a far more opaque place to hide.

And opacity is the direction of travel. As discovery, decision, conversion and measurement collapse into single AI-managed environments, value concentrates in the platforms that bundle all of it and away from the intermediaries who used to connect buyers and signals separately. McKinsey’s research into the agentic advertising economy (2026) found that 42% of US advertisers already name reliance on “black box” optimisation as a key risk AI brings to their media investment. They can feel the floor moving. What they don’t yet have is an independent way to stand on it.

None of this makes the platforms villains. It means the market needs what every functioning market eventually demands: a currency that is defined, governed and audited by someone other than the people selling it. We did it for TV viewing, radio listening and print circulation. We did it for impressions and viewability. We are still building it for attention. Recommendation is simply next and it is the hardest one yet, because the data lives inside the very systems we’d need to hold to account.

So here’s where I land. Trust and governance, not optimisation, are about to become the real competitive differentiators in an AI-mediated media. The job for our industry is to define recommendation as a currency, agree how it’s validated, and build the independent audit before the AI platforms define it for us.

Let’s shift our focus in 2026 on measuring being chosen.

---

*Postscript, 1 July 2026.* Four days after I wrote this, OpenAI’s ads boss David Dugan told Digiday that third-party measurement is “a natural step,” conceding that bidding and outcome data doesn’t prove a real human saw an ad, and that grading your own homework won’t build lasting trust. This is encouraging to see, but it is about verifying the ads within the AI platform. The harder question, as Digiday itself concludes, is how you measure influence inside a conversation in the first place, something that in Digiday’s words is “still being invented.” That’s the gap this piece is about.

---

*Anita Caras is founder and principal consultant at Caras Consulting, an independent research and insight consultancy specialising in marketing measurement and effectiveness.*

**Sources**

- Ofcom, *Online Nation 2025* (published 10 December 2025): UK AI-tool usage 54% vs 31% prior year; ~30% of searches show AI overviews; 53% see summaries often. https://www.ofcom.org.uk/media-use-and-attitudes/online-habits/from-apps-to-ai-search-how-the-uk-goes-online-in-2025

- EY, *AI is reshaping consumer products selection, accelerating brand consideration risk* (11 May 2026): 47% of CP executives say influencing recommendations will be essential within five years; 21% can deliver today. https://www.ey.com/en_gl/newsroom/2026/05/ey-report-ai-is-reshaping-consumer-products-selection-accelerating-brand-consideration-risk

- McKinsey, *The agentic advertising economy: From attention to action* (2026): 42% of advertisers cite reliance on “black box” optimisation as a key AI risk (US sample, 182 advertising/marketing leaders). https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-agentic-advertising-economy-from-attention-to-action

- Digiday, *OpenAI ads boss David Dugan on third-party measurement: ‘it’s a natural step’* (Krystal Scanlon, 30 June 2026): Dugan on third-party verification as “a natural step”; Digiday’s conclusion that measuring influence inside a conversation “is still being invented.” https://digiday.com/marketing/openai-ads-boss-david-dugan-on-third-party-measurement-its-a-natural-step