Need to know facts
of the month

One in 10 consumers globally say they have been in a relationship with an AI chabot.

Don’t worry, when we say ‘relationship’ we mean ‘friendly’, not ‘more than friendly,’ at least we hope. WARC’s latest Consumer Trends Report 2026 has shown how far AI has moved into our emotional lives.

10% of consumers worldwide admit to being in a relationship with an AI chatbot and 62% of these AI-users say they would rather turn to an AI chatbot than a human friend for personal advice.

The findings point to something bigger than usage, AI is starting to occupy emotional space once reserved for people. It is increasingly used for companionship, emotional support and social fulfilment, with users drawn to frictionless interaction that removes the complexity of human relationships (according to WARC). The shift is in fact driving new categories, from AI companion devices to tools designed for elderly care and safety monitoring, with trust deepening with continued use.

For the marketing world, it seems the implication here is somewhat sensitive (not to mention dystopian) but also quite clear: AI is now operating inside emotional territory the industry is still trying to define, let alone design for. While it opens new engagement opportunities, it also raises serious responsibilities around protecting vulnerable audiences, particularly younger users, and the mental health implications of replacing human connection with synthetic companionship. 

Almost 90% of consumers feel human-made artwork is more meaningful.

While this might seem obvious, the same WARC report has highlighted a growing tension around AI in content creation.

85% of consumers say knowing an artwork is made by a human makes it more meaningful, while 78% say it is “very or extremely important” that AI-generated content is clearly labelled.

AI content is flooding digital spaces and consumers are pushing back on issues of quality, ethics and transparency, with some gaming communities even resisting AI use strongly enough to influence developer decisions.

However, this sentiment may be more mixed than it first appears, as research from Billion Dollar Boy shows that around four in 10 consumers believe AI has improved the quality and diversity of creator content, suggesting a split between caution and curiosity rather than outright rejection. Overall, it is clear that authenticity remains the priority, with most consumers demanding clear disclosure of AI use, particularly in high-stakes areas like healthcare, politics and law, as well as across social media where trust and provenance are increasingly under pressure.

AI is unlocking time for agencies, but not value yet.

According to the industry consultancy Spark AI, and their Agencies Report (Spring 2026), 89% of agency staff are saving one to 10 hours per week through AI, but most of that time is being absorbed into doing more of the same work rather than reinvested into strategy or higher-value output.

‘Shadow’ AI is now the norm.

The same report finds that 52% of AI use in agencies is still informal, with staff operating without formal guardrails or approved tools. This ‘Shadow AI’ trend is also increasing concern, with interest in IP and risk management up 50% in six months, as governance struggles to keep pace with adoption.

The agency workforce is splitting into multiple speeds… and capability is diverging fast.

While 83% of staff say they are capable AI users, only 15% are building real workflows beyond one-off prompting, (according to the Spark AI Agencies report).

At the same time, usage patterns show a broader structural split: 45% of staff use AI daily, 40% use it intermittently, and 15% barely engage at all. Together, this is creating fragmented, multi-speed workforces where small groups are combining productivity and capability, while most remain in light or inconsistent use. The result, says Spark AI, is uneven output quality and a growing difficulty for agencies to standardise AI adoption across teams.

AI is already reshaping the creative business model.

The use of AI in creative generation has risen sharply from 20% to 47% in the last six months, signalling a shift from experimentation to production.

As a result, agencies are beginning to experiment with outcome-based pricing models and invest in proprietary tools and workflows to create defensible advantages, moving away from billing time towards billing output, efficiency and performance, the Spark AI Agencies report claims. The implication is that AI is not just accelerating production but quietly forcing a redesign of how creative value is defined, priced and delivered.

Governance is becoming a client requirement.

The report also notes that having a formal AI policy is now a baseline client expectation but only 25% of agencies are currently updating contracts to reflect AI use, leaving the majority exposed to legal and contractual ambiguity.

For best practice on “what to do about it” - with regards to the issues Spark AI have highlighted above - please refer to the report and Spark AI’s website.

Google is rebuilding search around AI and calling it its biggest change in 25 years

At Google I/O (19 May), Google unveiled a radically redesigned AI-powered search experience, describing it as “the biggest upgrade to our search box in over 25 years.” The company is clearly accelerating beyond traditional search toward something more conversational, multimodal and increasingly agentic, combining “the best of a search engine with the best of AI.” 

The updates include conversational follow-ups inside search, agentic shopping and booking tools, AI-generated dashboards and trackers, and persistent “information agents” that continuously monitor the web on a user’s behalf. Google is also expanding multimodal search, allowing users to search with text, images, files, videos and even Chrome tabs. 

The direction of travel is clear: Google wants search to move beyond retrieving information and toward helping users complete tasks. Instead of simply surfacing links, search is increasingly being designed to reason, monitor, recommend and act. Some of the most ambitious features include AI-generated interfaces, persistent task management and systems capable of handling bookings, shopping and ongoing monitoring inside search itself. 

Industry reaction suggests this is less the start of a new era than the acceleration of one already underway. As Stephanie Wong, SEO director at Assembly APAC, told Campaign: “People stopped searching for links a long time ago, they are searching for answers.” Wong added bluntly that “the search bar, as we knew it, is gone.” 

 At the same time, much of what Google presented still feels more like a strategic direction than an overnight transformation. Many of the more advanced agentic capabilities are launching gradually through AI Mode, subscription products and limited regional rollouts, meaning the traditional Google experience is unlikely to disappear immediately. What matters more is the long-term shift in intent: search is steadily becoming more conversational, proactive and embedded into ongoing tasks rather than functioning purely as a destination for retrieving links. 

The implication is not that GEO suddenly matters, but that Google is now structurally rebuilding search around AI-native behaviour already being shaped by platforms like ChatGPT and Perplexity. Visibility is increasingly about becoming a source AI platforms choose to reference and trust. Or as Ash Terry, ecommerce director at Jaywing, told Campaign: “Traditional SEO gets you ranked, generative engine optimisation (GEO) gets you cited.” 

OpenAI’s latest releases are a big deal, even if they are getting harder to neatly explain.

OpenAI has shipped a stack of meaningful upgrades, including GPT-5.5, Image 2.0, and new ‘Workspace Agents’ that connect across software and can be built in natural language.

For adland, GPT-5.5 improves how models interpret briefs, reason through tasks and generate usable outputs across strategy, copy and code, meaning the models have become faster and more consistent at turning inputs into structured, on-brand work.

As Ethan Mollick (an AI professor) observes, the progress of OpenAI’s tools are now “increasingly hard to quickly demonstrate,” but GPT-5.5 “intuitively feels better, faster, and more efficient.”

The shift is less about new features and more about smoother, more reliable workflows from brief to output.

Image 2.0 also appears to have made a bigger leap in image generation than it first appeared. As Sean Betts (Chief AI and Innovation Officer at Omnicom Media UK) points out, it is the added “thinking” capability that matters. The model can plan, research and assemble complex visual outputs in one go, not just generate images. This means producing not just a visual, but the logic around it (instructions, systems, data-led outputs), effectively turning image generation into a multi-step problem-solving tool. What makes it powerful for Betts, is not just better visuals, but the ability to “go off, plan, do some research… and then string together the visual expression it needs to answer the prompt.”

Together, these advances from OpenAI point to a real shift away from standalone headline features and towards workflow-level intelligence. The takeaway for adland is that technology is collapsing the number of steps between thinking and making, steadily compressing the creative pipeline inside agencies.

Anthropic halting the release of its latest model due to AI safety fears is a big reality check for frontier AI.

Anthropic’s decision to hold back the release of its latest model, Claude Mythos, reflects the growing concern over how easily advanced systems could be misused, with the model reportedly capable of autonomously identifying and exploiting “zero-day” vulnerabilities - previously unknown software flaws that could be used before fixes exist.

As The Guardian notes, Anthropic itself warned the system could “turn computers into crime scenes,” discovering and chaining weaknesses to effectively take over systems, raising the risk that such tools could “scale cyber-attacks and defences alike.”

Releasing models with this level of capability risks lowering the barrier to sophisticated cyberattacks, while also accelerating defensive discovery. For adland, it is a stark reminder that as AI becomes embedded across campaign delivery, data and production systems, the same security risks apply, making trust, governance and safe deployment just as critical as creative performance.

OpenAI adds another layer to the rise of GEO by trialling cost-per-click (CPC) ads.

OpenAI is testing CPC ads inside ChatGPT, shifting the platform from visibility-based ads to intent-driven performance. Instead of paying for impressions, advertisers would only pay when users actually engage, moving closer to search-style marketing where value is tied to action rather than reach. In practice, this turns conversational AI into a high-intent space where prompts act as queries and commercial moments happen in real time rather than being pre-set.

For ad agencies, this signals a shift toward planning around what people ask inside AI chats, not just keywords or feeds, tightening the link between prompting, discovery and conversion, and reinforcing GEO as a marketing discipline.

OpenClaw has ripped the AI dam wide open.

The emergence and rise of the agentic system OpenClaw is yet another sign of the ever-growing shift from AI acting as an assistant to AI acting as an autonomous operator that’s capable of executing multi-step tasks across tools and systems. As Jasleen Carroll, Director of AI and Product at Anything Is Possible, warns: “This will and should scare you. Humanity and our economy are not ready for this.” Carroll points to real-world uses such as instructing agents to “complete platform checks and share an optimisation task list by 8.30am,” and potential structural change inside agencies, including leaner teams and pressure on roles like project management.

In his TEDx talk, OpenClaw creator Peter Steinberger describes the “holy shit moment” when the platform he created first began to independently find tools, convert files and stitch APIs together. He added that “bots give up, agents improvise” and despite fears around agentic AI, “the lobster is loose, and it’s not going back into the tank.”

It seems that in the marketing sphere, more and more execution work will be handled by agents, compressing production cycles and shifting human value toward direction, judgement and strategic control.

That said, agents should not be fully executing or optimising campaigns directly… this is where the industry needs to draw a clear line.

Advertising is entering an agentic phase where AI agents can negotiate, transact and increasingly attempt to execute campaigns across the supply chain. The IAB Tech Lab is mapping existing ad standards into “agentic” extensions and building an Agent Registry for trust and transparency, while AdCP is working to standardise how agents communicate with ad platforms so they can discover inventory and manage media buys across sellers without bespoke integrations.

On paper, this promises major efficiency gains, but as Sean Betts argues in his newsletter article,

“AI agents are probabilistic by design… you cannot guarantee exactly what it will do,”

making them suitable for insight and forecasting, not live optimisation. “The moment an AI agent can place a buy… you are no longer executing the plan you agreed.” As Daniel Hulme (Chief AI Officer at WPP) similarly warns, “an AI agent that generates a flawed media plan… erodes trust in a way that takes years to rebuild.”

The conclusion, at the moment, is clear: agents can support decisions, but execution must remain tightly governed, deterministic and accountable.

Shopping agents are racing into reality.

Mondelez is betting big on agentic commerce by hiring a global lead to prepare for a future where AI agents shop on behalf of consumers, not just assist them. As reported by Digiday, the shift moves commerce away from clicks and channels toward...

“intelligent agents that guide decisions, discovery and transactions,”

with some retailers expecting 30% of traffic to be agent-driven by 2028. According to McKinsey, the global market could reach $3-5 trillion by 2030, with up to $1 trillion in U.S. retail alone. The battleground is shifting from influencing human decisions to influencing how AI agents choose, recommend and buy.

BUT the app gold rush shows how early this still is.

Retailers are rapidly launching apps inside ChatGPT and Claude to capture this shift, with hundreds already live, but adoption remains low. As Dimitri Ewald (chief of staff at Alpic) also told Digiday,

“adoption and conversion are pretty low… people don’t even know that there are apps.”

It seems that the core issue is discoverability as these experiences do not yet surface naturally in AI interactions, limiting their impact.

While agentic commerce is coming fast, today’s executions are still experimental, meaning short-term advantage lies less in building apps and more in ensuring that brands are visible, structured and influential inside the core AI systems themselves.

Three is the magic number when it comes to AI tools.

A recent Harvard Business Review study suggests that there is a clear limit to effective AI tool use before productivity starts to decline. As Marcos Angelides (Head of AI Operations, Publicis Media, Chair of IPA Commercial Leadership Group) notes on LinkedIn,

“three is the magic number”

when it comes to stacking AI tools. Beyond that, output quality drops sharply and the risk of “AI brain fry” increases as cognitive load, switching costs and over-reliance start to outweigh gains.

Less-is-more it seems. Rather than layering multiple tools for the sake of efficiency, the most productive users focus on a small, well-chosen stack and prioritise depth of use over constant tool-hopping.

There will be three types of agency in the AI-era.

Spark AI’s handbook for agencies, Shift, outlines three emerging types of agencies as AI reshapes the industry. First, the “volume game” agencies using AI to produce vast quantities of low-cost, “good enough” content at speed and scale, effectively becoming creative technology platforms competing on efficiency.

Second, the “innovation players” using AI to create entirely new kinds of work. This means adaptive, personalised, real-time experiences and campaign systems that can be tested at scale before production. These agencies will be pushing creative boundaries rather than just outputting more of it.

Third, the “craft defenders” focusing on hand-made, provably human work, positioned as premium and scarce in an AI-saturated market, similar to vinyl in a streaming world.

The best way to train yourself on AI is to play around with it.

The biggest barrier to AI adoption in agencies is not the tech, it is behaviour. As Jules Love (AI coach and co-founder of Spark AI) puts it in Shift, many agencies “say all the right things about innovation but never actually block out time for it,” expecting polished results too quickly and letting anxiety dominate over experimentation.

This anxiety is widespread. Laura Jordan Bambach (founder and Chief Creative Officer of Uncharted) describes a “silent murmur of ‘where does this leave me?’” among creatives, but argues the only way forward is to “immerse yourself in it and play,” because “the creativity still belongs to you.”

The same theme runs through Sean Betts’s view [link Sean Betts IPAi Forum article] that AI is a “general-use technology… like the internet,” meaning it cannot be taught top-down because people need hands-on experience in real contexts to understand its value.

Morten Legarth (Creative Director at VCCP and faith) echoes this idea in Shift: “The number one thing to achieving mastery is just to go and get stuck in… you’re not going to learn it by reading a book, you need to just go and do it.”

And while experimentation is essential, structure still matters. Clear governance, as outlined by Neil Perkin (AI coach), gives people the confidence to explore safely.

The AI experts are all singing the same tune: stop overthinking AI and start using it, because capability will not come from theory, it will come from doing.