{ "id": "content-marketing", "label": "Content & Marketing", "description": "AI for creating, distributing, and measuring content across channels. The most mature creative domain: SEO, copywriting, email, and social media management are established practice. Personalisation at scale and sentiment-driven strategy are advancing but unevenly adopted. Content authenticity and deepfake detection remain bleeding-edge.", "icon": "✍️", "filters": [ "creating", "selling" ], "hasSummary": true, "hasExecSummary": true, "practiceCount": 15, "evidenceCount": 2562, "practices": [ { "slug": "advertising-creative-generation-and-testing", "name": "Advertising creative generation & testing", "tier": "good-practice", "trend": "steady", "blockerType": null, "description": "AI that generates ad creative variants and tests them for performance, iterating toward higher-performing combinations. Includes automated A/B creative testing and variant generation; distinct from image generation in creative media which produces general imagery rather than performance-optimised ads.", "evidenceCount": 163 }, { "slug": "autonomous-content-production", "name": "Autonomous content production", "tier": "bleeding-edge", "trend": "blocked", "blockerType": "trust-safety", "description": "AI that generates and publishes content on a schedule with minimal human editorial oversight or intervention. Includes fully automated blog, social, and newsletter pipelines; distinct from assisted content generation which requires human review before publication.", "evidenceCount": 151 }, { "slug": "brand-voice-workflows", "name": "Brand-voice workflows", "tier": "leading-edge", "trend": "steady", "blockerType": null, "description": "AI content generation constrained to a brand's specific voice, tone, and style guidelines for consistent output. Includes custom model fine-tuning and style enforcement layers; distinct from generic content generation which produces without brand constraints.", "evidenceCount": 155 }, { "slug": "campaign-optimisation-and-performance-prediction", "name": "Campaign optimisation & performance prediction", "tier": "established", "trend": "steady", "blockerType": null, "description": "AI that optimises programmatic ad targeting, predicts campaign performance, and recommends budget allocation. Includes predictive audience modelling and real-time bid optimisation; distinct from marketing analytics which analyses historical rather than optimising future performance.", "evidenceCount": 198 }, { "slug": "content-localisation-and-translation", "name": "Content localisation & translation", "tier": "leading-edge", "trend": "steady", "blockerType": null, "description": "AI-powered translation and cultural adaptation of marketing content for international markets beyond literal translation. Includes transcreation and cultural sensitivity checking; distinct from personal translation tools which support individual communication rather than marketing campaigns.", "evidenceCount": 194 }, { "slug": "content-moderation-and-brand-safety", "name": "Content moderation & brand safety", "tier": "established", "trend": "steady", "blockerType": null, "description": "AI that monitors and moderates user-generated or AI-generated content to ensure brand safety and policy compliance. Includes automated content filtering and brand safety scoring; distinct from content safety in AI governance which governs AI outputs rather than published content.", "evidenceCount": 211 }, { "slug": "content-planning-and-repurposing", "name": "Content planning & repurposing", "tier": "leading-edge", "trend": "steady", "blockerType": null, "description": "AI that generates content calendars, identifies topic opportunities, and repurposes existing content across formats and channels. Includes trend-driven editorial planning and automated content adaptation; distinct from autonomous content production which handles end-to-end publishing.", "evidenceCount": 143 }, { "slug": "influencer-and-partnership-identification", "name": "Influencer & partnership identification", "tier": "good-practice", "trend": "steady", "blockerType": null, "description": "AI that identifies and evaluates potential influencers, partners, and brand ambassadors based on audience fit and engagement. Includes fraud detection for fake followers and ROI prediction; distinct from candidate sourcing in HR which identifies employees rather than marketing partners.", "evidenceCount": 161 }, { "slug": "long-form-content-generation", "name": "Long-form content generation", "tier": "good-practice", "trend": "steady", "blockerType": null, "description": "AI generation of articles, whitepapers, reports, and other long-form written content for marketing and thought leadership. Includes draft generation, outline expansion, and style-matched writing; distinct from technical documentation which targets product information rather than marketing content.", "evidenceCount": 148 }, { "slug": "market-segmentation-and-competitive-analysis", "name": "Market segmentation & competitive analysis", "tier": "good-practice", "trend": "steady", "blockerType": null, "description": "AI that analyses competitive positioning, generates market segments, and creates data-driven buyer personas. Includes messaging gap analysis and segment propensity modelling; distinct from sales ICP refinement which targets individual account fit rather than market-level segmentation.", "evidenceCount": 167 }, { "slug": "marketing-analytics-seo-and-attribution", "name": "Marketing analytics, SEO & attribution", "tier": "good-practice", "trend": "steady", "blockerType": null, "description": "AI that analyses content performance, optimises for search engines, and models marketing attribution across channels. Includes keyword opportunity analysis and multi-touch attribution modelling; distinct from campaign performance prediction which forecasts future rather than analysing past performance.", "evidenceCount": 189 }, { "slug": "personalised-content-delivery-and-recommendation", "name": "Personalised content delivery & recommendation", "tier": "established", "trend": "steady", "blockerType": null, "description": "AI that delivers personalised content experiences to individual users based on behaviour, preferences, and context. Includes content recommendation engines and dynamic content assembly; distinct from personalisation engine design which builds the system rather than using it.", "evidenceCount": 196 }, { "slug": "short-form-social-and-campaign-content-generation", "name": "Short-form, social & campaign content generation", "tier": "good-practice", "trend": "steady", "blockerType": null, "description": "AI that generates social media posts, ad copy, email campaigns, and other short-form marketing content. Includes platform-specific content adaptation and A/B variant generation; distinct from long-form content generation which produces articles, guides, and thought leadership pieces.", "evidenceCount": 170 }, { "slug": "social-listening-and-brand-monitoring", "name": "Social listening & brand monitoring", "tier": "established", "trend": "steady", "blockerType": null, "description": "AI that monitors social media, news, and review platforms for brand mentions, sentiment shifts, and emerging trends. Includes real-time reputation alerting and trend forecasting; distinct from competitive analysis which focuses on rival positioning rather than brand perception.", "evidenceCount": 180 }, { "slug": "specialist-content-events-technical-and-product-documentation", "name": "Specialist content — events, technical & product documentation", "tier": "leading-edge", "trend": "steady", "blockerType": null, "description": "AI that generates event materials, sales collateral, technical documentation, and product content for specific professional contexts. Includes white paper drafting and event programme generation; distinct from general long-form or short-form content which targets broader audiences.", "evidenceCount": 136 } ], "summary": "## Where AI Stands in Content & Marketing\n\nMarketing was the first business function to adopt generative AI at scale, and it is now the first to show what the plateau looks like. Adoption is effectively universal — 91 to 92 percent of marketers in every survey this fortnight — and confidence is going the other way. Orbit Media's twelve-year longitudinal survey of 1,042 content marketers found 92 percent using AI and just 14 percent saying their blog delivers strong results, the lowest figure in the survey's history; AI users and non-users were equally likely to report success. A multi-dataset analysis spanning CoSchedule, 4,200 tracked articles and Ahrefs found 91 percent of teams had increased AI output and 6 percent had seen a meaningful performance improvement. Among advertising buyers, the share who believe AI video delivers a return on investment fell from 93 to 82 percent in a year. Only 41 percent of marketers say they can prove AI ROI, down from 49 percent in 2025. At the enterprise level the picture is starker: a meta-analysis of six major studies puts 95 percent of generative AI pilots at zero P&L return, with 42 percent abandoned before production — up from 17 percent in 2024 — and an average realised return of 5.9 percent. Teradata's survey of 1,000 senior technology leaders found 90 percent planning agentic investment and 37 percent able to point to measurable impact. The industry has bought the tools. It has not, in the main, bought the outcomes.\n\nThe exceptions follow a rule that has held for a year and hardened this fortnight: AI delivers where it works on data and stalls where it faces an audience. P&G's predictive lifetime-value models targeted 800,000 households through Sam's Club retail media and hit 94 percent prediction accuracy over twelve months of actual purchasing, with a 2.3x marketing lift. Alibaba's multimodal recommender added 3.38 percent to GMV in production; Tencent's generative ranking lifted click-through 3.57 percent. Medicube matched 33,900 creators through an AI platform and generated $102.9 million in US TikTok Shop revenue, up 219 percent year on year. Sprinklr reported 40 percent growth in AI-native product lines and $1.03 billion in remaining performance obligation. On the creative side, the ceiling was independently confirmed. Ipsos tested AI-made advertising with 3,000 consumers across 20 brands and found it scored 14 percent lower on creative effect and 17 percent lower on brand-equity effect than human-made work, with a structural cause: only 30 percent of AI ads used a narrative arc, against a 49 percent human norm. Semrush's analysis of 42,000 top-ranking pages found hybrid content — AI draft, heavy human rewrite — ranked 34 percent higher than raw AI output, and pure AI content earned eight times fewer first-position rankings. On high-difficulty keywords, the gap between pure AI content and human content widened from 14 to 31 percent over sixteen months. The quality gap is not closing with better models. It is widening as the audience and the ranking systems learn to discount what machines produce unaided.\n\nThe third force is that the platforms and the courts have stopped waiting for marketers to work this out. Google's September migration of search campaigns onto AI Max has no opt-out; independent field testing recorded a 35 percent decline in return on ad spend against traditional match types and a 72 percent spike in invalid traffic. Google's 17 August Smart Bidding change cut Target ROAS campaign performance 29 to 35 percent in its first week, driven by a 15 percent drop in conversion rate. Google's Campaign Agent reached general availability on 19 August and Meta's Andromeda now runs autonomous creative testing; the first agency reports describe output as \"competent but generic,\" reliable on accounts with more than a hundred weekly conversions and erratic below that, and IPG-Omnicom has mandated weekly human review. Meanwhile the FTC and 22 state attorneys general charged Amazon with running a soft-reserve pricing algorithm that inflated advertiser auction costs from 30–40 percent of auctions to 80 percent between 2019 and 2024, affecting more than a million advertisers; a Frankfurt court held Meta liable for fraudulent third-party advertising on the ground that algorithmic control defeats Digital Services Act immunity; and Korea's Fair Trade Commission began requiring prior substantiation for AI-generated advertising claims on 3 September. Every practice in this domain held its position this cycle. What changed is that the binding constraints — data quality, review capacity, and the terms platforms impose — are now documented with numbers rather than asserted.\n\n## What's New, 2026-09-05 to 2026-09-19\n\nThe most consequential new evidence concerns the gap between adoption and return, and the degree to which it is a data problem rather than a model problem. Validity's survey of 500 marketers found 78 percent of C-suite respondents had acted on AI recommendations they later discovered were wrong because the underlying customer data was poor, and 62 percent had experienced revenue loss from data issues. Frederick Vallaeys documented the mechanism at account level: Smart Bidding optimising against incomplete conversion signals — no cost of goods, no returns, no fulfilment costs — produced an 11x reported return on ad spend on a campaign losing £0.39 per order. Kiin's benchmark of 263 LinkedIn advertisers and $7.7 million in spend found cost per lead varying 32-fold, from $32 to $1,005, with bidding and audience configuration explaining 80 percent of the variance independent of message quality. On the creative side, the Ipsos study above landed alongside Motion's benchmark of 578,750 Meta creatives and $1.29 billion in spend, which found the winner rate stable at around 5 percent regardless of production value, with text-only formats (11.6 percent) and offer-led hooks (9.29 percent) beating production-heavy work — a finding that rewards iteration cadence over budget and explains why the accounts recovering from Andromeda are those shipping eight to twelve creatives a month. A further signal on the limits of autonomy: Google's multimodal video creation reached general availability on 27 August, and the product still requires manual review of brand identity, claims and disclosures before anything ships. The generation step is automated; the accountable step is not.\n\nMeasurement moved in both directions. Google Search Console's generative-AI performance reports rolled out globally on 31 August, a feature mandated by the UK's Competition and Markets Authority, but limited to impressions, with click data promised for December. Similarweb documented why that matters: 91.2 percent of visitors who arrive after a ChatGPT recommendation reach the brand through a non-AI channel, so the influence is invisible to standard attribution; a separate synthesis of five studies found 70.6 percent of AI-referred traffic arrives without referrer headers and is filed as Direct in GA4, so bidding algorithms systematically underfund a channel that converts 4.1 times better. Similarweb's ecommerce report found AI-recommended brands enjoy a two-to-one advantage in purchase decisions and AI platform referrals up 200 percent year on year; Contentsquare found 97 percent of shoppers will reconsider an AI-recommended purchase if the brand's website disappoints, which locates the return on AI discovery firmly in post-click experience. Elsewhere, peer-reviewed research on the DF26 dataset found deepfake detection accuracy on influencer talking-head video collapsed from 94 to 48 percent in a year, while DoubleVerify counted more than 500 million low-quality AI-generated impressions in the first half of 2026 and 48 percent of UK consumers said seeing such content beside a brand damaged their perception of it. TikTok disclosed that 94.1 percent of its Q1 removals were automated while its EU moderation workforce fell from 6,354 to 3,738. In localisation, a UN Office analysis named the systemic blind spot — AI systems deployed across 50-plus languages but safety-tested in English only — as OpenAI built a 14-billion-token Kazakh corpus with a native cultural benchmark and Cohere released an open-weight 218-billion-parameter translation model that outperforms DeepL in non-European regions. OpenAI's ChatGPT for Financial Services reached general availability with Morgan Stanley and Evercore as design partners. Vendor consolidation continued: a live audit of 30 social-listening tools found 12 shut down or acquired; Runway passed $200 million in annual recurring revenue with Chime reporting $200,000 saved per video; TechCrunch's running list of shuttered AI products now includes OpenAI's Sora. No practice changed direction this cycle. The stability is itself the finding: the domain has stopped moving because its constraints are organisational, and organisations move slowly.\n\n## Key Tensions\n\n- **Universal adoption has decoupled from advantage.** When 92 percent of practitioners use the same tools and AI users and non-users report success at the same rate, the tool confers no edge; only 6 percent of marketers have AI fully embedded in workflow, and it is that 6 percent — Neoxra's mid-market client cutting cost per piece from $180 to $65 and revision cycles from 3.8 to 1.2; the boutique agency eliminating six hours of manual overhead per 24-asset cycle — that reports compounding returns. The KOZEC maturity model puts the difference at 5–10x output at 75–85 percent lower cost per article for teams with persistent brand context and governance, against no measurable gain for the rest. The advantage now lies in the operating model wrapped around the model, and that is slow to copy.\n\n- **Platform automation is being imposed, and it optimises whatever it can measure.** Google's AI Max migration has no opt-out; its August bidding change hit budget-constrained campaigns for 80 percent of the impact; 75 percent of practitioners still run exact-match keywords because tight control outperforms broad automation in their accounts, and 53 percent of 1,300 PPC professionals say Google Ads got harder this year. The Vallaeys case — 11x reported ROAS, £0.39 lost per order — and the Amazon soft-reserve charges show the same failure from two sides: automated auctions reliably maximise the metric they are given, and the metric is rarely profit. The operator's remaining leverage is the signal architecture, which is why HyperVerge's 327 percent MQL growth on a fixed $4,000 budget came from Smart Bidding plus weekly human keyword audits, not from either alone.\n\n- **The measurement layer is blind to the highest-value channel.** 68 percent of US Google searches now end without a click; 70.6 percent of AI-referred traffic is misfiled as Direct; overlap between top-ten organic results and AI citations fell from 76 to 38 percent in seven months. Platform-native attribution inflates its own channel — an audit of 109 Klaviyo accounts found email revenue overstated by more than 25 percent — and best-in-class multi-touch tools capture only 30–60 percent of actual conversion activity. Only 18 percent of practitioners rate their attribution as highly accurate, yet 89 percent of B2B demand-generation leaders use models they do not trust for budget decisions. The three-layer stack of marketing mix modelling, incrementality testing and tactical attribution is the sector's answer; it is thicker, not more precise.\n\n- **Generation cost is heading to zero while verification labour is not.** A practitioner's costed pipeline puts token spend at $0.23 per article against $100 for human fact-checking, a 100:1 ratio that no model improvement changes. Marketing Dive's Gartner-sourced analysis notes one manager can generate 50 blog posts in an afternoon that legal and compliance cannot review in a week. Enterprise deployments show 23 percent brand-voice drift on novel topics with only 12 percent of agents holding persona on unfamiliar ground. Specialist content shows citation fabrication rates of 5 to 90 percent depending on model. The consequence is visible in moderation, where TikTok halved its EU human workforce as automated removals passed 94 percent, and in fraud, where detection accuracy fell by half in a year of generative progress. Every practice that works has a person at the gate, and the gate is where the cost has moved.\n\n- **Liability is migrating to whoever controls the algorithm.** The Frankfurt ruling that algorithmic control defeats platform immunity, Meta's $17.1 billion settlement with 51 states over engagement-optimised recommendation, the Amazon auction charges and Korea's prior-substantiation rule for AI advertising claims all point the same way: courts and regulators are treating the optimisation system as the responsible actor. EU AI Act Article 50 applies a materiality test — incidental AI use is exempt, deepfakes need visible labels, penalties reach €15 million or 3 percent of global turnover — and YouTube has begun demonetising generic template video. For marketers, disclosure now carries a measured 31.5 percent click-through penalty, and non-disclosure carries a legal one. That trade-off is permanent, and the organisations treating it as a budget line rather than a surprise are the ones still scaling.\n\n## Top 10 Evidence Items\n\n1. **We asked 1,042 content marketers if their blog works. Just 14% said yes.** (adoption-metric) — Orbit Media's twelve-year longitudinal survey is the single clearest data point for the domain's central finding: adoption is universal (92 percent) but confidence just hit its lowest recorded level, and AI users report success at the same rate as non-users. https://www.linkedin.com/pulse/we-asked-1042-content-marketers-blog-works-just-14-said-crestodina-xvzrc\n2. **Enterprise AI Deployment Outcomes: 95% Pilots Zero ROI, 42% Abandoned Pre-Production, 5.9% Average ROI** (industry-report) — the meta-analysis behind the summary's starkest enterprise figure, showing abandonment before production nearly tripled from 17 percent in 2024, which grounds \"the industry has bought the tools, not the outcomes\" in a multi-study number rather than a single anecdote. https://intuitionlabs.ai/articles/enterprise-ai-deployment-outcomes\n3. **P&G Predictive LTV Modeling: 94% Accuracy Over 12 Months on Retail Media** (case-study) — the counter-example that proves the rule: AI delivers where it works on structured purchasing data (800,000 households, 2.3x lift), not where it faces an audience, and this is the strongest verified deployment number in the scan. https://corporate.walmart.com/about/samsclub/news/2026/09/turning-membership-into-momentum\n4. **AI Generated Ads vs. Human UGC: What the Data Shows** (research-paper) — underlies the Ipsos finding that AI-made advertising scores 14-17 percent lower on creative and brand-equity effect, evidence that the creative quality gap is structural (narrative arc usage) rather than a model-capability problem that better models will close. https://billo.app/blog/ai-generated-ads-performance/\n5. **Google's Forced AI Max Migration: Independent Testing Shows 35% ROAS Decline and 72% Invalid Traffic** (opinion) — direct field evidence for the \"platform automation is being imposed\" tension: a no-opt-out migration with independently measured harm, not a vendor claim of improvement. https://www.davydovconsulting.com/post/google-ads-ai-max-migration-has-started-what-business-advertisers-should-do-now\n6. **Marketing Funnel Fragmentation: How AI Broke Attribution** (opinion) — Similarweb's own account of why 91.2 percent of ChatGPT-referred visitors convert through a channel invisible to standard attribution, the mechanism behind the summary's claim that measurement is blind to the highest-value channel. https://aisearch.similarweb.com/blog/marketing-funnel-fragmentation/\n7. **Smart Bidding Failure Mode: Algorithms Optimize Against Incomplete Signals, Destroying Business Value** (opinion) — Vallaeys's account-level walkthrough of an 11x reported ROAS on a campaign losing money per order, the clearest illustration that automated auctions reliably maximise the metric they're given, and the metric is rarely profit. https://www.linkedin.com/pulse/ai-could-ruin-your-ads-account-27-minutes-frederick-vallaeys-ir7sc\n8. **FTC Charges Amazon with Algorithmic Auction Manipulation Affecting 1M+ Advertisers** (news-coverage) — the regulatory action naming an algorithmic system, not a person, as the mechanism of harm (soft-reserve pricing pushed 30-40 percent of auctions to 80 percent), central to the \"liability is migrating to whoever controls the algorithm\" tension. https://digiday.com/marketing/ad-tech-briefing-regulatory-storm-cloud-gather-and-scatter-over-big-tech-and-indies-count-the-cost/\n9. **German court holds Meta liable for fraudulent ads on Facebook, Instagram** (case-study) — the Frankfurt ruling that algorithmic control defeats Digital Services Act immunity, a legal precedent the summary treats as evidence courts are done waiting for platforms to self-police. https://www.aa.com.tr/en/science-technology/german-court-holds-meta-liable-for-fraudulent-ads-on-facebook-instagram/4060030\n10. **The AI graveyard: a running list of projects and startups that didn't make it** (news-coverage) — TechCrunch's running failure ledger, now including OpenAI's Sora, is the uncomfortable-truth counterweight to the fortnight's product-launch news and a concrete marker of the \"vendor consolidation continued\" thread (12 of 30 social-listening tools alone shut down or acquired). https://techcrunch.com/2026/09/15/the-ai-graveyard-a-running-list-of-projects-and-startups-that-didnt-make-it/", "execSummary": "**The headline:** Nearly every marketing team now uses AI, and confidence in it is falling. The gains still available come from data quality and human oversight, not from buying more tools.\n\n### The Picture\n\nMost companies have finished adopting AI in marketing. A twelve-year survey of 1,042 content marketers found 92 percent using it — and only 14 percent saying their content program delivers strong results, the lowest reading in the survey's history. Teams using AI and teams not using it reported success at the same rate. A small group is pulling ahead, and they share a pattern: AI works on data behind the scenes while a person signs off on anything a customer sees. P&G predicted customer value across 800,000 households with 94 percent accuracy; a Korean skincare brand matched 33,900 creators by machine and booked $102.9 million in US TikTok sales. The rest are producing more content, faster, into a market that now measures the difference: AI-made ads score 14 percent lower on creative impact than human-made ones, and pure AI articles earn eight times fewer top search rankings. The window to move from the second group to the first is closing, because platforms are now forcing automation on advertisers and courts are assigning liability for what the algorithm does.\n\n### This Fortnight\n\n- **Google moved search advertisers onto its AI campaign system with no opt-out, and independent tests recorded a 35 percent drop in return on ad spend.** A separate Google bidding change in mid-August cut performance on return-targeted campaigns by roughly a third in its first week. Find out who in your organization holds the campaign settings, and treat any account with fewer than a hundred conversions a week as one that needs a person checking it — the largest agency holding company now mandates weekly human review.\n\n- **The clearest evidence yet that poor data, not poor AI, is what goes wrong.** In a survey of 500 marketers, 78 percent of senior executives said they had acted on an AI recommendation they later found to be wrong because the customer data underneath it was bad. Before approving another AI initiative, ask what it will be fed and who owns keeping that clean.\n\n- **AI advertising has a measured quality ceiling, confirmed by a 3,000-person Ipsos study.** AI-made ads scored 14 percent lower on creative effect and 17 percent lower on brand effect than human work, largely because only 30 percent of them tell a story. Use AI for volume and variants; keep people on the campaigns that carry the brand.\n\n- **Courts and regulators started holding the algorithm's owner responsible.** A German court ruled Meta liable for fraudulent ads on Facebook because its algorithms chose where to show them, and US regulators charged Amazon with rigging ad auctions against more than a million advertisers. If a platform's automation places your ads or your content, ask your counsel who carries the risk when it places them badly.\n\n- **Tools for spotting fake video are losing.** Peer-reviewed research found deepfake detection (spotting AI-fabricated video of real-looking people) fell from 94 percent to 48 percent accuracy in one year on the kind of talking-head clips influencers post. Verify creators through payment and delivery history, not through how genuine they look.\n\n### Coming Up\n\n- **December brings two measurement deadlines.** Google has promised click data for AI search results in its Search Console by December, after a UK regulator forced it to show impressions from August 31; and the EU's marking requirement for AI-generated images, audio and video takes effect December 2. Make sure your analytics team is ready for the first and your vendor contracts cover the second.\n\n- **Korea now requires advertisers to substantiate AI-generated claims before they run, and other regulators will follow.** The rule took effect September 3 and is the first to require prior approval rather than after-the-fact enforcement. Start keeping an evidence file for every AI-produced claim in market, so you are not building one under deadline.\n\n- **Fully autonomous campaign management is now on sale from Google and Meta.** Early agency reports call the output \"competent but generic\" and erratic on smaller accounts. Pilot it on high-volume, low-stakes budgets only, and decide in advance what result would make you switch it off.\n\n### What's Hard About This\n\n- **Your analytics cannot see your best channel.** Seventy percent of visitors arriving from AI assistants show up in Google Analytics as \"Direct\" with no source, even though they convert around four times better than ordinary search traffic — so automated bidding systematically underfunds them. No dashboard fixes this; it takes a deliberate decision about what to measure and how.\n\n- **Making content is now nearly free, and checking it is not.** One costed production pipeline puts the AI expense at 23 cents per article and the human fact-checking at $100. Every efficiency claim you hear should be tested against that ratio, because the reviewing cost is where the money actually goes.\n\n- **Disclosure costs clicks, and non-disclosure costs more.** Labeling an ad as AI-made cuts click-through by about a third, while failing to label content that needs it now carries fines of up to 3 percent of global revenue in Europe. There is no configuration that avoids both; there is only a decision about which cost you would rather explain.", "headline": "Nearly every marketing team now uses AI, and confidence in it is falling. The gains still available come from data quality and human oversight, not from buying more tools.", "execSummarySections": [ { "id": "the-picture", "title": "The Picture", "body": "Most companies have finished adopting AI in marketing. A twelve-year survey of 1,042 content marketers found 92 percent using it — and only 14 percent saying their content program delivers strong results, the lowest reading in the survey's history. Teams using AI and teams not using it reported success at the same rate. A small group is pulling ahead, and they share a pattern: AI works on data behind the scenes while a person signs off on anything a customer sees. P&G predicted customer value across 800,000 households with 94 percent accuracy; a Korean skincare brand matched 33,900 creators by machine and booked $102.9 million in US TikTok sales. The rest are producing more content, faster, into a market that now measures the difference: AI-made ads score 14 percent lower on creative impact than human-made ones, and pure AI articles earn eight times fewer top search rankings. The window to move from the second group to the first is closing, because platforms are now forcing automation on advertisers and courts are assigning liability for what the algorithm does." }, { "id": "this-fortnight", "title": "This Fortnight", "body": "- **Google moved search advertisers onto its AI campaign system with no opt-out, and independent tests recorded a 35 percent drop in return on ad spend.** A separate Google bidding change in mid-August cut performance on return-targeted campaigns by roughly a third in its first week. Find out who in your organization holds the campaign settings, and treat any account with fewer than a hundred conversions a week as one that needs a person checking it — the largest agency holding company now mandates weekly human review.\n\n- **The clearest evidence yet that poor data, not poor AI, is what goes wrong.** In a survey of 500 marketers, 78 percent of senior executives said they had acted on an AI recommendation they later found to be wrong because the customer data underneath it was bad. Before approving another AI initiative, ask what it will be fed and who owns keeping that clean.\n\n- **AI advertising has a measured quality ceiling, confirmed by a 3,000-person Ipsos study.** AI-made ads scored 14 percent lower on creative effect and 17 percent lower on brand effect than human work, largely because only 30 percent of them tell a story. Use AI for volume and variants; keep people on the campaigns that carry the brand.\n\n- **Courts and regulators started holding the algorithm's owner responsible.** A German court ruled Meta liable for fraudulent ads on Facebook because its algorithms chose where to show them, and US regulators charged Amazon with rigging ad auctions against more than a million advertisers. If a platform's automation places your ads or your content, ask your counsel who carries the risk when it places them badly.\n\n- **Tools for spotting fake video are losing.** Peer-reviewed research found deepfake detection (spotting AI-fabricated video of real-looking people) fell from 94 percent to 48 percent accuracy in one year on the kind of talking-head clips influencers post. Verify creators through payment and delivery history, not through how genuine they look." }, { "id": "coming-up", "title": "Coming Up", "body": "- **December brings two measurement deadlines.** Google has promised click data for AI search results in its Search Console by December, after a UK regulator forced it to show impressions from August 31; and the EU's marking requirement for AI-generated images, audio and video takes effect December 2. Make sure your analytics team is ready for the first and your vendor contracts cover the second.\n\n- **Korea now requires advertisers to substantiate AI-generated claims before they run, and other regulators will follow.** The rule took effect September 3 and is the first to require prior approval rather than after-the-fact enforcement. Start keeping an evidence file for every AI-produced claim in market, so you are not building one under deadline.\n\n- **Fully autonomous campaign management is now on sale from Google and Meta.** Early agency reports call the output \"competent but generic\" and erratic on smaller accounts. Pilot it on high-volume, low-stakes budgets only, and decide in advance what result would make you switch it off." }, { "id": "whats-hard-about-this", "title": "What's Hard About This", "body": "- **Your analytics cannot see your best channel.** Seventy percent of visitors arriving from AI assistants show up in Google Analytics as \"Direct\" with no source, even though they convert around four times better than ordinary search traffic — so automated bidding systematically underfunds them. No dashboard fixes this; it takes a deliberate decision about what to measure and how.\n\n- **Making content is now nearly free, and checking it is not.** One costed production pipeline puts the AI expense at 23 cents per article and the human fact-checking at $100. Every efficiency claim you hear should be tested against that ratio, because the reviewing cost is where the money actually goes.\n\n- **Disclosure costs clicks, and non-disclosure costs more.** Labeling an ad as AI-made cuts click-through by about a third, while failing to label content that needs it now carries fines of up to 3 percent of global revenue in Europe. There is no configuration that avoids both; there is only a decision about which cost you would rather explain." } ], "url": "https://www.thestateofplay.ai/domain/content-marketing", "license": "CC BY 4.0", "licenseUrl": "https://creativecommons.org/licenses/by/4.0/", "generatedAt": "2026-10-01" }