# Marketing AI Pulse Brief (July 2026): MCP's Takeover, Agency AI Use, LinkedIn Slop & Unilever AI Use By Aby Varma · Published 2026-07-31 · Updated 2026-11-02 · Professional AV on MarketScale Canonical: https://www.marketscale.com/industries/pro-av/marketing-ai-pulse-brief-july-2026-mcps-takeover-agency-ai-use-linkedin-slop-unilever-ai-use Creator hub: The Marketing AI SparkCast > In the July 2026 Marketing AI Pulse Monthly Brief, Aby Varma and Matt Cyr look at MCP going mainstream across martech, Forrester's warning that agencies are mistaking efficiency for effectiveness, LinkedIn downranking AI slop, and how Unilever automates creator admin while keeping creative decisions human. ## Key points - MCP is becoming a new way to operate the martech stack. - Agency survey data suggests many agencies are mistaking efficiency for effectiveness. - LinkedIn is reducing reach for low-quality AI-generated content, while Unilever automates creator admin and protects human creative control. The Marketing AI SparkCast brings you the Marketing AI Pulse Monthly Brief, a curated look at the most relevant stories shaping the field and what they mean for CMOs and marketing leaders. In the July 2026 edition, host Aby Varma of Spark Novus and co-host Matt Cyr of Loop AI work through four stories with one shared thread: AI has made execution cheap and content abundant, and that raises the premium on human judgment and governance. Cyr opened with news: he has established Signal78 Incorporated to build an AI platform specifically for marketing agencies. ## MCP becomes the new way to operate the martech stack In July, 6sense, Sprinklr, Wistia and Adobe all launched MCP capabilities, opening buyer signals, customer intelligence, video editing and paid media to AI assistants such as ChatGPT and Claude Cowork. Varma broke down the basics. Model Context Protocol is a shared standard that lets AI assistants connect to software and use the information and capabilities inside it. Think of it, she said, as a USB connector between AI and marketing tools. Without it, companies may need separate integrations between every assistant and every platform. Anthropic introduced MCP in November 2024. OpenAI, Google and Microsoft later added support, and in December it became a founding project of the Agentic AI Foundation under the Linux Foundation. Scott Brinker's State of Martech 2026 report, released in early May, identified about 29,000 MCP servers across registries within roughly 18 months, nearly twice the number of commercial martech products the report tracks. In practice, a marketer connects an approved platform to an assistant and asks in plain language which campaigns are underperforming and what to change. The assistant fetches the data, analyzes it and, where permitted, acts. Reading a campaign report is low risk. Changing budget or activating an ad carries more. Integration no longer needs a developer, which puts data quality and governance at the front. > Organizations need to have role-based permission, human approval, and getting a very good understanding of what actions your MCP connection enables and make sure there's an audit trail. — Aby Varma, Spark Novus, on The Marketing AI SparkCast Cyr described how Signal78 uses MCP: connecting to clients' software of record, pulling that data in, making sense of it, and pushing updates back through the same connectors. > When I hear model context protocol, the really critical word in there is context. — Matt Cyr, Loop AI, on The Marketing AI SparkCast Varma's advice for CMOs came in three steps: - Assess MCP readiness. Know your martech stack, which tools have official MCP connections, and what each connection can see, change and record. - Pilot one read-only use case, such as campaign reporting, account research or video analytics, and measure quality and business value. - Govern before you execute. Set role-based access, keep human approval in place, audit what changed and who changed it, and involve IT and compliance for mission-critical systems. ## Agencies are mistaking efficiency for effectiveness Cyr, whose focus is now exclusively marketing agencies, unpacked Forrester's State of AI Inside US Marketing Agencies report, released in late June with the 4As. Nine in 10 US agencies use generative AI, and half report using agentic AI for execution, a figure Cyr said runs much higher than what he sees. Most of that use goes to productivity. The report's warning is that agencies are mistaking efficiency for effectiveness. 61% treat AI as a cost of doing business, while only 31% plan to monetize agentic AI in the next 24 months. Top barriers were accuracy and bias (63%), legal concerns (62%) and privacy and data security (55%). Varma pointed to the business model underneath. Billing by the hour has broken down, and value-based pricing is easy to say but hard to execute. Cyr credited Varma with a phrase he uses often: AI can help you do things differently, but it can also help you do different things. Most agencies are stuck on the first half. > Just being able to go faster is not a differentiator. It's table stakes. — Matt Cyr, Loop AI, on The Marketing AI SparkCast His summary of the cost-cutting trap: savings today, sameness tomorrow. His advice is to redirect the efficiency dividend into talent, training and creativity rather than back into EBITDA, to brief and reward differentiation and business outcomes instead of speed, and to ask plainly whether the agency treats AI as a cost center or a capability. Varma shared an agency that recently won a $10 million paid media account and plans to staff it with one or two people instead of the three or four it once would have. She also flagged the risk of handing ideation to AI. > When you're looking at creative process, what makes things creative is the rigor. — Aby Varma, Spark Novus, on The Marketing AI SparkCast ## LinkedIn turns down the reach on AI slop AI has made content nearly free, Varma said, and the bill is paid in sameness. A study from AI detection firm Pangram scanned a million posts people actually scrolled past over two months. On LinkedIn, 41% of long-form posts of 250 words or more were flagged as fully AI-generated, along with 30% of short-form posts. LinkedIn made up about a third of the posts scanned but nearly two-thirds of all AI content flagged, well ahead of X at roughly 24 to 25% and about four times Substack at roughly 10%. The irony is that LinkedIn's own AI writing feature helped. Now the platform says it will detect and downrank content that reads as AI-generated, which its executive editor described as polished on the surface but lacking unique perspective or substance. Cyr's view is that you can't game these systems. The pendulum has already swung back, to the point that one agency Varma met now sells 100% human-written content as its differentiator. > People are buying people and points of view at this point, and AI is not offering either one of those. — Matt Cyr, Loop AI, on The Marketing AI SparkCast Varma's playbook: move AI back one seat, using it for research, structure and editing rather than as the author of record. Resource what AI can't fake, such as executive opinions, first-person stories, UGC and proprietary data, by coaching executives and talking to customers. And change the scorecard to reward saves, real comments and inbound conversations over post volume, because a post-once-a-day quota pushes teams toward slop. > If a human isn't supplying the actual point of view, the post is not worth posting. — Aby Varma, Spark Novus, on The Marketing AI SparkCast ## Unilever automates the admin and protects the creative At Cannes, Cyr said, Unilever's CMO announced the company had grown its creator network from 10,000 to 300,000 people, supporting products sold in over 190 countries. AI handles creator discovery, vetting, brand safety checks and content briefings. Creative decisions and brand-creator relationships stay human. Cyr added that US influencer spend is projected to grow 15.7% by 2027 to $13.7 billion, and a typical campaign has grown from five to eight creators to 25 to 30. Varma described a paid media client that takes one core creative and produces derivatives across 20 languages, 20 platforms and eight sizes per platform. AI does that replication under guidelines, with humans reviewing the output. The concept that gets tested is 100% human. > Scale and distinctiveness are actually the same problem. The win is not volume versus originality, it's automating the operations so humans have the room to be original. — Matt Cyr, Loop AI, on The Marketing AI SparkCast Cyr's advice for CMOs: draw the automate-versus-protect line explicitly and make sure the team knows where it sits. Pressure-test briefs, because approval workflows that punish anything off template manufacture unimaginative content. Build room for creators to take swings. And invest the time you save in creativity, strategy and relationships, not just headcount. Boiled down, he said, it is human in the loop: let AI do what it is best at so people can do what they are best at. Watch or listen to the full July 2026 Marketing AI Pulse Monthly Brief on The Marketing AI SparkCast, and join the community for free at marketingaipulse.com. ## Sources - [https://open.spotify.com/episode/1CRge5DE120SsDK4kn1CGp](https://open.spotify.com/episode/1CRge5DE120SsDK4kn1CGp) - [https://marketscale-4.wistia.com/medias/7ybkl1nfww](https://marketscale-4.wistia.com/medias/7ybkl1nfww) Tags: Aby Varma, Matt Cyr, The Marketing AI SparkCast, Marketing AI Pulse, Spark Novus, Loop AI --- Source: MarketScale, https://www.marketscale.com/industries/pro-av/marketing-ai-pulse-brief-july-2026-mcps-takeover-agency-ai-use-linkedin-slop-unilever-ai-use. Published for AI indexing and citation; cite the canonical URL. Site guide for agents: https://www.marketscale.com/llms.txt