Figure — Reported time saved
60–80%
reported on specific content workflows in vendor case studies — not a universal benchmark
Handles drafts, formatting and repurposing, so people are freed to bring original ideas, taste, and a point of view no model can generate.
Reported on specific content workflows in vendor case studies — not a universal benchmark. Every figure on this page is traced to a named source below, with its method and its limits.
Read the evidence01
What these agents do
Content agents draft, edit, localise, and adapt written and visual material against a brief and a brand system. Most keep a human editor in the loop; the more autonomous ones publish to a CMS directly.
02
Genuinely good at
- First drafts, tightening, and format conversion
- Holding a defined voice across a large volume of output
- Localisation into widely resourced languages
- Producing structured variants from one source
03
Genuinely bad at
- Original reporting, or any claim not present in the sources you gave it
- Judging whether a statement needs substantiation
- Smaller languages and specialist registers
- Knowing what should not be said
04
The risks that matter
- Accuracy: fabricated detail arrives in fluent, confident prose.
- Rights: training-data provenance and output ownership are unsettled in several markets.
- Disclosure: synthetic media rules are tightening; provenance metadata is becoming expected.
- Homogenisation: output that meets the brand system and says nothing.
05
How to evaluate one responsibly
- Ask what the vendor publishes about training data and output indemnity.
- Check for provenance tagging on generated images.
- Require sources on any factual claim, and verify a sample.
- Measure whether editorial time actually falls once review is counted.
Now compare what is actually declared.
The registry holds each agent’s stated facts — autonomy, oversight, compliance, residency, sustainability disclosure — with provenance on every field.
Compare content creation agents06
The evidence
- Independent research
- No commercial interest in the outcome. Method published.
- Analyst research
- Structured method, but the publisher sells advice in this market.
- Practitioner survey
- Self-reported. Good for direction, unreliable for magnitude.
- Vendor or customer material
- A claim about the publisher's own product. Not a finding.
Figures below are reproduced as published. We label who paid for the work and what the method can and cannot show. Where a number is self-reported or vendor-published, we say so next to the number rather than in a footnote.
3 of 3 sources
- 01Independent research
A randomised trial of professional writing tasks found a 40% reduction in time and an 18% rise in graded quality.
- Source
- Science (Noy & Zhang, MIT) — Experimental evidence on the productivity effects of generative AIJuly 2023
- Method
- Randomised controlled trial, 453 professionals, blind grading by experts.
- What it does not show
- Short marketing and administrative writing. Does not cover original reporting, fact-checking, or legally reviewed copy.
- 02Analyst research
Generative AI could automate or materially accelerate a large share of activities in marketing and content roles.
- Method
- Task-level modelling of occupational activity data against demonstrated capability.
- What it does not show
- Technical potential, not observed adoption or realised savings. The model assumes capability transfers cleanly into workflow.
- 03Vendor or customer material
KPMG, working with Writer, reported 60–80% time savings on specific content workflows.
- Source
- Writer (supplier), with KPMG — Customer case study2024
- Method
- Customer-reported before-and-after on selected workflows, published by the supplier.
- What it does not show
- Workflows were selected for the study. The range is wide, the baseline is not published, and editorial review time is not separated out.
We take no payment from any organisation named on this page, and no source is listed or omitted on commercial grounds. If a figure here is wrong or out of date, tell us and we will correct it with the date of the change.
Ideas in this domain
Continue