The State of Play

A living index of AI adoption across industries — where established practice meets the bleeding edge
UPDATED DAILY
← All domains

✍️ Content & Marketing

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.

15 practices: 4 established, 6 good practice, 4 leading edge, 1 bleeding edge

The Headline

Almost every marketing team now uses AI, but it is buying speed, not better results. Meta has paused its terms making advertisers liable for AI it generates, but has not rewritten them.

15
practices tracked
5
at the frontier
0
moving this fortnight
2,949
evidence items

The Picture

Almost every large brand now uses AI in marketing: 96% of major brands, according to the World Federation of Advertisers. Most competitors now use AI for drafting, ad variants and scheduling, so using it no longer sets anyone apart. What sets a small group apart is measured return. Only 6% of marketing organizations tell Bain that AI delivers significant performance impact, and Bain's leaders are 3.7 times more likely to have fully redesigned their workflows rather than added tools to old ones. Most companies sit in the middle: they produce more content faster but cannot show that it sells more. Meanwhile search clicks, attribution and platform-reported conversions, the things everyone measures against, are all moving.

This Fortnight

  • B2B marketers doubled their use of AI agents in a year, but very few credit AI with better content. In the 2027 trends survey from the Content Marketing Institute and MarketingProfs, use of agents (software that acts on its own without being prompted) rose from 28% to 59%, mostly for ideation, scheduling and distribution. Only 17% say AI improved quality, against 78% who cite speed, so if your AI targets measure volume, you are rewarding the part that is not producing better work.

  • Meta says nearly 10 million advertisers now use its AI creative tools, and agencies warn that brands can no longer tell what drove their results. Brainlabs argues that Meta's ongoing AI changes to live ads keep the learning inside Meta. It still thinks the tools are worth testing for brands short on creative, but advises against handing over a whole campaign on day one. Run them against a control rather than in place of your own testing.

  • Google's AI summaries now appear on most searches for brand names in BrightEdge's sample, up from about a fifth in mid-August. Across 282 tracked brand and product names, the share reached 64% by October 24, although BrightEdge does not measure whether those summaries cost clicks. Tools that track AI visibility are moving pricing behind sales calls with attribution still unshipped, and Peec AI's own test showed fabricated pages earning ChatGPT referrals, so treat any AI share-of-voice dashboard as directional rather than proof.

  • France's privacy regulator fined two Fnac Darty companies a total of €7 million because app tracking that fed product recommendations ran without prior consent. The tracking data was also used to train a recommendation algorithm. Fnac Darty disputes the decision, arguing the features were product recommendations rather than targeted advertising. If you sell in Europe, check what consent covers the data feeding your personalization engine, because the regulator treated it like ad tracking.

  • New research is split on whether disclosing AI use costs consumer trust. Three peer-reviewed experiments with 576 participants found that disclosing AI in brand social content reduced consumers' behavioral intentions. A pre-registered LinkedIn study found no trust penalty, and readers spotted AI edits only 49% of the time. The cost appears to fall on content that is recognizable as AI, which favors heavy editing over high volume.

Coming Up

  • Meta's paused advertiser terms could return at any time with the liability clause intact. The terms would make advertisers solely responsible for AI output, including features Meta switches on, and Meta has not said whether you are liable for features you never enabled. Bookshop.org and REI have already had live ads altered by Meta's AI. Have legal review the terms now, and audit which automated AI features are switched on in every ad account before the terms return.

  • Existing custom GPTs on personal ChatGPT accounts retire on December 11, 2026. Teams that keep brand voice or campaign instructions in a GPT built on someone's personal account will lose them. Move them into a written voice document held in a company-controlled workspace before then.

  • EU high-risk AI obligations begin on December 2, 2027, bringing human oversight, approval trails and bias monitoring to legal, medical and safety translation. EU rules requiring disclosure of AI-generated content have applied in every target language since August 2, 2026. If AI translates your product labels, instructions or contracts, budget now for documented review, not just faster output.

What's Hard About This

  • Faster drafting has moved the work into review rather than removing it. Templafy finds 48% of knowledge workers say AI saves them little time once editing is counted, and the pressure for speed is showing up as burnout among B2B content staff. Removing the review step does not solve this, because sites that publish AI content at scale without review usually lose search traffic within a year or two.

  • The same platforms generate your ads, grade them and run the auction. An audit of 240 agency client accounts found platforms over-claiming conversions by between 18% and 31%. Holdout tests and your own marketing-mix modeling are the only checks you control, and most companies do not run them well.

  • Discovery is moving into AI answers that expose very little data. Some 68% of US Google searches ended without a click in early 2026, analytics tools often file AI referrals as direct traffic, and a review of 45 studies of AI-search optimization found no technique with a stable causal effect. For the foreseeable future, decisions about AI search spending will rest on estimates rather than measurement.

2012View this domain's timeline →Today

Practices in this Domain (15)

PRACTICETIERTREND
Advertising creative generation & testingGOOD PRACTICE— Steady
Autonomous content productionBLEEDING EDGE— Steady
Brand-voice workflowsLEADING EDGE— Steady
Campaign optimisation & performance predictionESTABLISHED— Steady
Content localisation & translationLEADING EDGE— Steady
Content moderation & brand safetyESTABLISHED— Steady
Content planning & repurposingLEADING EDGE— Steady
Influencer & partnership identificationGOOD PRACTICE— Steady
Long-form content generationGOOD PRACTICE— Steady
Market segmentation & competitive analysisGOOD PRACTICE— Steady
Marketing analytics, SEO & attributionGOOD PRACTICE— Steady
Personalised content delivery & recommendationESTABLISHED— Steady
Short-form, social & campaign content generationGOOD PRACTICE— Steady
Social listening & brand monitoringESTABLISHED— Steady
Specialist content — events, technical & product documentationLEADING EDGE— Steady
Read the full technical briefing (1,689 words) →

Where AI Stands in Content & Marketing

Content and marketing is where generative AI went mainstream first, and the adoption figures now read like saturation. The World Federation of Advertisers finds 96% of major brands using generative or agentic AI in marketing. Ahrefs puts the share of marketers using AI to create or assist with content at 93%, and Metricool finds three-quarters of social media managers using it daily. The ad platforms have absorbed much of the rest. PPC Land reports Meta putting the number of advertisers on its generative creative tools at nearly 10 million in Q3. Google says Gemini has helped generate more than 1.5 billion ad assets, and its September auto-migration moved more than 500,000 advertisers into AI Max. Automated bidding, brand-safety classification, recommendation and social listening are already core plumbing. In those areas the argument is about proof and control, not about whether to use them.

What the domain has not produced is a return to match. Bain finds only 6% of marketing organisations saying AI delivers significant performance impact today. CMI and MarketingProfs find 78% of B2B marketers crediting AI with speed and 17% with quality. Orbit Media's survey put the share of content marketers reporting strong results at 13.9%, the lowest in its history. The production bottleneck has gone, and three others have taken its place: review capacity, audience trust and measurement. Ahrefs finds 89% of AI users manually checking output. Templafy finds 95% of knowledge workers editing AI drafts, and 48% saying little time is actually saved. Where organisations remove the human gate, the usual result is reversal. Lily Ray's analysis of more than 220 AI-scaled sites found 54% lost at least 30% of peak organic traffic within 12–18 months. Gartner finds only 11% of marketing organisations have agentic AI in live production, and 89% of those piloting still require sign-off before agent output goes live.

Two things set this domain apart. First, the ground it measures is moving under it. Elite Content Marketer reports that 68% of US Google searches ended without a click in early 2026. BrightEdge watched AI Overviews spread across most of the branded queries it tracks in two months. So attribution, SEO and listening are being rebuilt for surfaces that expose little data. Second, the same platforms now generate the creative, grade it and run the auction, while their terms push liability for the output onto advertisers. Momentum is strongest in machine-facing infrastructure: agentic campaign tools that leave changes for a human to approve, MCP connectors across competitive intelligence, scheduling and documentation, and docs platforms where Mintlify says more than three-quarters of traffic now comes from AI agents. Progress stalls wherever the output carries a byline, a factual claim or cultural weight.

What's New, 2026-10-31 to 2026-11-14

The fortnight's evidence mostly widened the gap between use and value, and the clearest new numbers came from B2B. In the 2027 B2B trends survey from CMI and MarketingProfs, AI agent use rose from 28% to 59% in a year, mostly for ideation, scheduling and distribution. Accuracy and quality control tied as the top challenge, each at 54%. MarTech Edge's reading of the same research found 65% of B2B marketers under more pressure to produce faster, burnout among 48% of individual contributors, and no AI disclosure policy at 52% of organisations. Contentoo found 80% of the 203 content leaders it surveyed regularly publish work they know is not good enough. Infosys's survey of 1,503 marketing leaders found the share reporting AI value in content creation rising from 35% to 49%, with the gains in speed and cost rather than revenue. On search, the Russian agency AINET ran an eight-month test on ten identical new sites. Human-written pages ended with 71.4% of Yandex visibility and 67.3% of Google's, after the AI pages competed early and then decayed. With five pages per arm and a single query, the result is indicative only. EY's global content strategist said the firm keeps AI out of final thought-leadership writing. Evidence on disclosure split. Three peer-reviewed experiments (n=576) found that disclosing AI use in brand social content cut behavioural intentions. A pre-registered LinkedIn test (n=214) found no trust penalty, and readers spotted AI alteration only 49% of the time.

Platforms and measurement supplied the rest. Brainlabs' Liz DeAngelis warned that Meta's continuous AI creative changes leave brands unable to isolate what drove their gains. BrightEdge found AI Overviews on 64% of 282 tracked brand names by 24 October, up from 19% on 16 August. Semrush and Profound moved pricing behind sales conversations, and Brandlight still lists attribution as "coming soon". Peec AI's own test showed 50,000 fabricated statistics pages drawing more than 10,000 ChatGPT referral sessions over seven months. France's CNIL fined two Fnac Darty companies €7m over consent-free tracking that fed product recommendations and trained a recommendation algorithm. Blue Yonder found 46% of shoppers who bought on an AI recommendation had returned an item in the past year. TikTok now automates over 97% of video removals while former London moderators contest their redundancies in court. Separately, University of Bath researchers saw more than 95% of their reports on drug-supply accounts cleared, each after exactly 30.5 minutes. Sprout Social launched AI sentiment scoring in NewsWhip on 10 November without publishing accuracy figures. Nimdzi's notes from LocWorld 56 recorded Salesforce burning its annual localisation budget before mid-year once it moved from three releases a year to monthly. Nimdzi also reported default translation engines redacting "Taliban" and "IRGC" from Rainbow Railroad's intake messages. No practice changed position. The new evidence deepened existing patterns rather than breaking them.

Key Tensions

  • Faster output, unchanged quality, more review. Drafting time has collapsed, but the saved hours go into fact-checking, voice correction and approvals. That is why CMI finds 78% crediting AI with speed against 17% with quality, and why Templafy finds 48% of knowledge workers saying they save little time once editing is counted. The pressure lands on staff: 48% of B2B individual contributors report burnout linked to the push for faster output.

  • Platforms that generate, grade and disclaim. Meta's generative tools reach nearly 10 million advertisers, yet its updated terms would make advertisers solely responsible for AI output, including features Meta switches on. By 11 November Meta had paused those terms with the wording unchanged. Bookshop.org and REI have already had live ads altered without approval. Agencies warn that continuous in-platform optimisation keeps the learning inside Meta, and a Visionary Marketing audit of 240 accounts found platforms over-claiming conversions by 18% to 31%.

  • Mandated labels that cost engagement. EU transparency obligations for AI-generated content have applied since 2 August 2026, and NYU Stern research cited by the IAB found AI labels cut click-through by 31.5%. This fortnight, peer-reviewed experiments found disclosure lowered perceived authenticity and purchase intent. A pre-registered LinkedIn study found no trust penalty when readers could not spot the AI. The cost seems to fall on AI that is recognisable as such, which rewards heavy editing over volume.

  • Measurement losing sight of discovery. Attribution was built on clicks and referrers, while discovery is moving into AI answers that expose neither. GA4 files much AI-referred traffic under Direct, AI Overviews now cover most tracked brand queries in BrightEdge's sample, and a review of 45 GEO studies found no technique with a stable causal effect. The tools sold to fill the gap are moving to quote-only pricing with attribution unshipped, and Peec's fabricated pages show the citation signal can be gamed.

  • Automating the judgement calls. The work AI handles least well is the work organisations most want to automate: moderation context, cultural adaptation, regulated drafting and simulated customers. Pew's AI digital twins missed real panel answers by 12.4 points on average, and Strat7 found synthetic respondents contradicting real price ordering 68% of the time. A RAPS analysis found only 7% of AI-generated references in regulatory writing both authentic and accurate. Nimdzi's LocWorld advice to route work by risk rather than by line is the pattern holding up across the domain.

Top 10 Evidence Items

  1. Faster and Cheaper, But Not Better: What the 2027 B2B Content and Marketing Trends Report Reveals About AI, Speed, and Trust (industry-report) — The sharpest new number on the use/value gap the briefing opens with: agent adoption nearly doubled while quality gains stayed flat. https://contentmarketinginstitute.com/b2b-research/b2b-content-marketing-trends-research?exid=167982
  2. Brainlabs warns Meta's AI creative could leave brands unsure what worked (news-coverage) — Captures the platform tension directly: the same tool generating creative also removes the ability to attribute what worked. https://ppc.land/brainlabs-warns-metas-ai-creative-could-leave-brands-unsure-what-worked/
  3. AI Is Defining Your Brand Before You Get the Chance To (industry-report) — Quantifies how fast the ground is moving under measurement, the domain's defining complication. https://www.brightedge.com/resources/weekly-ai-search-insights/ai-overviews-brand-searches
  4. Human-written content outperforms AI-generated articles by up to 2.5 times in search visibility (case-study) — A controlled (if small) test showing the Lily Ray pattern in action: AI content wins early then decays against human-written pages. https://webiano.digital/human-written-content-outperforms-ai-generated-articles-by-up-to-2-5-times-in-search-visibility/
  5. How consumers respond when brands disclose using AI in creating social media content: examining brand personification and communication strategies (research-paper) — Peer-reviewed evidence for the disclosure-cost tension, showing AI labels damaging perceived authenticity. https://www.emerald.com/jpbm/article-abstract/35/8/1236/1386277/How-consumers-respond-when-brands-disclose-using?redirectedFrom=fulltext
  6. Perceived Trust and Transparency in AI-Altered vs. Human-Written LinkedIn Posts (research-paper) — The necessary counterpoint: disclosure only costs when readers can actually detect it, which the briefing flags as the real mechanism. https://arxiv.org/html/2611.06660
  7. Fnac Darty fined €7 million: when digital personalisation crosses the line on data privacy (news-coverage) — A concrete regulatory cost for the unconsented data pipelines feeding personalisation and recommendation training. https://matomo.org/blog/2026/11/fnac-darty-cnil-fine-tracking-consent/
  8. TikTok is cutting human moderators in favour of AI. Fired workers are fighting back (news-coverage) — The clearest case of automating a judgement call that still fails, with the human cost playing out in court. https://www.bigissue.com/news/employment/tiktok-cuts-human-moderators-ai-fired-workers-fight-back/
  9. Synthetic Users and Synthetic Respondents vs. Real Consumers (opinion) — Shows why automating the judgement calls organisations most want to delegate is exactly where AI is weakest. https://rwazi.com/blog/synthetic-respondents-vs-real-consumers
  10. What The Room Said At LocWorld 56 Vancouver (opinion) — Supplies the one practice-level fix the briefing credits: routing by risk rather than by line, from a function dealing with real budget blowouts. https://www.nimdzi.com/what-the-room-said-at-locworld-56-vancouver/