Short answer. Train marketing teams to work with AI-generated content (AIGC) by making the programme structured, role-specific and tied to real workflows, not by handing out a licence and hoping. Formal training is consistently linked to faster adoption and better returns than self-directed learning, and teams that get it right report meaningfully higher productivity and hours saved per week. Generative AI use in marketing is now near-universal; formal, job-specific training on it is not.
Key takeaways
- Generative AI use in marketing has reached 87% of marketers, but only 17% have had comprehensive job-specific training on it.
- Formal, structured training programmes are consistently linked to faster adoption and stronger returns than self-directed learning.
- A programme that pays off has four layers: tool fluency, workflow integration, editorial judgement, and governance and disclosure.
- Only 38% of organisations tie AI training to day-to-day workflows, which is the main reason value stalls after the first layer.
- Measuring a baseline before training and re-measuring at 90 days is what lets a team prove the programme actually worked.
Why is training now the bottleneck, not the tools?
Because adoption is essentially finished and capability is not; the gap between the two is where the remaining value sits. AIGC (AI-generated content) is content — copy, images, video, audio — produced by generative AI tools and reviewed by people before it reaches an audience; it is the material a marketing team is now expected to work with daily.
Salesforce's State of Marketing series tracked generative AI use in at least one marketing workflow rising from 51% in Q1 2024 to 76% in Q1 2025 to 87% in Q1 2026 — a 36-point rise in 24 months, one of the fastest sustained adoption curves recorded in marketing. Enterprise adoption reached 94%, and even teams under ten marketers crossed 73%.
Capability has not kept pace. Only around 40% of companies provide any formal AI training, only 17% of marketers have received comprehensive job-specific training, and 58% name skills gaps, not technology, as their biggest AI challenge. Tool adoption and workflow integration have also pulled apart, with a wide range of organisations still short of full integration despite near-universal tool use.
What returns does good training actually produce?
Large ones, and they compound because early structured adopters keep pulling ahead of teams that wait. Structured training means a defined curriculum with role-specific tracks and measured outcomes, run as cohorts rather than left to whoever picks up the tool fastest — and it is consistently associated with faster adoption and higher realised AI value than ad hoc or self-directed learning.
At team level, 81% of marketing leaders say AI has significantly improved productivity and strategic execution, and companies using AI report publishing markedly more content per month than they did before. AI-assisted campaigns generally run to better return on spend, better conversion through improved segmentation, and lower acquisition costs than the same activity run without AI support, though the exact multiples vary widely by study.
The returns also cluster by application, which tells you what to train on first, based on McKinsey's Global AI Survey data on marketing use cases:
| Application | Reported ROI multiple |
|---|---|
| Content drafting | 3.2x |
| Personalisation | 2.7x |
| Audience research | 2.4x |
| Ad copy | 2.3x |
Teams that adopted early also report a meaningfully larger year-over-year productivity gain than teams that waited until 2026 to formalise training — capability built early compounds, and capability bought late mostly catches up rather than gets ahead.
What should the programme contain?
Four layers, built in sequence — most failed programmes stop after the first.
| Layer | What it covers | Why it matters | Verdict |
|---|---|---|---|
| 1. Tool fluency | Prompting, iteration, tool selection, hands-on practice on live briefs | Courses and structured workshops are the most common formal-training approach among companies that run one at all | Necessary, not sufficient |
| 2. Workflow integration | Where AIGC sits in the actual content, campaign and reporting process | Only 38% of organisations run training tied to day-to-day workflows — the main reason value stalls | The real unlock |
| 3. Editorial judgement | Reviewing, fact-checking, brand voice, knowing when not to use AI | A large majority of companies already edit or review AI output before it ships; buyers expect a human-led minimum | Your quality moat |
| 4. Governance and disclosure | Usage policy, data handling, approvals, labelling obligations | 60% of AI-using organisations still lack an org-wide policy | Prevents the expensive mistake |
Most organisations also report they lack a documented AI roadmap despite high tool adoption, and a majority say they struggle to scale value from AI initiatives — both are governance and process failures, not tool failures. What separates programmes that pay back is structure, specificity and measurement.
| What works | What fails |
|---|---|
| Structured cohorts, run against a curriculum | A tool rollout with no curriculum behind it |
| Role-specific tracks: content, SEO, demand generation, brand | One generic session for the whole department |
| Baseline metrics measured before launch | No baseline, so no provable improvement |
| Training on live briefs, not sandbox exercises | Publishing unreviewed output — the "AI slop" backlash |
| A written usage policy issued alongside the training | English-only enablement in multi-market teams |
Capability is built, not licensed. Prioritise the roles with the most AI-augmentable tasks first, since they see the largest immediate time savings, and set a baseline before you start — most organisations cannot accurately measure their own AI ROI, so without one you cannot prove the programme worked.
The multilingual layer is where teams most often over-trust the tooling. Generative quality and tone degrade noticeably in lower-resource languages and regional variants, so a team trained to review English output will wave through copy that reads as machine-translated in half its markets. Training must therefore include native-speaker review as a defined step, not an optional check. Human-in-the-loop review — a trained person checking, correcting or approving AI output before it ships — is the mechanism that catches this, and what human-in-the-loop review actually does covers it in more detail. Native-speaker review, cultural adaptation and locale-specific quality checks are the human-in-the-loop work Lifewood's AIGC delivery network provides across 50+ languages, drawing on 56,788 registered contributors and a 95%+ accuracy SLA backed by two independent review passes.
What should you do first?
In this order, because the sequence is what produces the compounding returns.
- Measure a baseline before you teach anything. Current tool adoption, comfort levels, hours spent on key workflows. Most organisations cannot measure AI ROI accurately today; start by being one that can.
- Start with the highest-return workflow. Content drafting returns 3.2x on average, ahead of personalisation at 2.7x, audience research at 2.4x and ad copy at 2.3x.
- Run structured cohorts, not self-serve licences. Structured programmes consistently out-adopt self-directed learning; licences alone do not scale capability.
- Teach editorial judgement as a core skill. Reviewing, fact-checking and knowing when not to use AI should be an assessed competency, not an assumption.
- Issue the usage policy with the training. 60% of AI-using organisations have no org-wide policy. The policy is part of the curriculum, not a follow-up.
- Add native-speaker review for every market language. Output quality drops in lower-resource languages, so build the review step into the workflow you are training, following the same quality-control practices used to run AI-generated content at scale.
- Re-measure at 90 days against the baseline. Hours saved, output volume, review pass rate and campaign ROI are the numbers that justify the next round of investment, and the same discipline used when measuring whether an AI content programme is working applies here.
Much of the tool-fluency layer is a briefing problem in disguise: teams that cannot specify what they want get poor output regardless of the model. How to write a brief an AIGC team can produce from covers the input side of the same workflow, and pairing it with Lifewood's AIGC production services or AIGC video production is one way to put a trained team's briefs into a managed pipeline.