Google's ATLAS report says AI is powering a creative-work boom. Days later, workplace surveys and academic research argued that much of that output is low-quality 'fake creativity' colleagues have to redo.
2026-08-09 · by Moe Ameen
On July 23, 2026, Google published its AI & Economy ATLAS report, an analysis of roughly 15 million de-identified Gemini interactions across hundreds of occupations. Its headline finding was upbeat: creative and other "non-routine cognitive" work shows up in AI usage at nearly double its share of the wider economy (about 65% of work-related prompts versus a 35% baseline), and the vast majority of that use is collaborative — ideation, strategy, drafting, and learning — rather than end-to-end automation. Google's framing positioned AI as a creativity and ideation partner, not a replacement for creative labor.
Four days later, on July 27, a survey from the digital-transformation firm Adaptavist landed with the opposite mood. It reported that around two-thirds of workers said they preferred the pre-AI era of work, roughly four in ten would remove generative AI tools from the world entirely, and — among those who wanted AI gone — about 31% cited reduced creativity as a reason. Many respondents said they now spend more time checking and correcting AI output than the tools save, and that low-quality "AI slop" makes their work feel less meaningful and more repetitive.
The academic case had already been made in the spring. In a paper published April 1, 2026 in the Human Resource Management Journal, University of Bath professor Dirk Lindebaum warned that outsourcing thinking to large language models erodes the kinds of knowledge AI can't hold — hands-on experience, cultural judgment, and analytical problem-solving. "If people begin outsourcing thinking, decision-making or interpretation to AI systems, these critical forms of knowledge wither over time," he wrote, proposing "learning vaults" — protected spaces at work shielded from automated systems — to preserve real skill. The critique echoes the 2025 "workslop" research from Stanford and BetterUp, which described polished-looking AI output that lacks substance and pushes the real work of fixing it onto coworkers.
The controversy isn't that one side is lying. Both readings can be true at once: people are indeed reaching for AI on creative tasks far more than for routine ones, and a large share of what comes out is generic, voiceless, and low-effort — the appearance of creativity rather than the thing itself. That gap between volume and value is what critics mean by "fake creativity," and it is now the central quality question hanging over every AI content tool, Gemini included.
This controversy is really an argument about one line: the difference between content that's merely AI-made and content that's genuinely yours. "Fake creativity" happens when a generic model default speaks in a generic voice and nobody edits it before it ships. [Kompozy](/) is built to sit on the right side of that line. Every output runs through a [Persona Brief](/glossary/persona-brief) — a saved profile of how *you* actually write, with banned words and required phrasing — so the engine isn't imitating a model's house style, it's compounding yours. Images and video carry your look through [HyperFrames](/glossary/hyperframes) brand templates and a face-locked [persona](/glossary/persona-shorts), not a stock AI aesthetic. And nothing has to auto-publish: a per-post review [pipeline](/glossary/output-buckets) puts a human gate between generation and the feed, which is exactly the step the "workslop" research found missing.
Then it does the part the reports don't address — distribution. From one idea Kompozy generates text posts, [persona tweets](/glossary/persona-tweet), carousels, an avatar-narrated short, a blog article, and a newsletter, all held to the same voice, and schedules them across the eight social platforms plus blog and email on [Autopilot](/glossary/autopilot). The takeaway from the "fake creativity" backlash isn't "use less AI." It's "stop shipping the default." Own the voice, keep a human in the loop, and let the machine handle the repetitive fan-out — that's the version of AI-assisted creation the critics aren't complaining about.
It's the criticism that AI tools produce a high volume of polished-looking output that lacks real substance, originality, or a distinctive voice — the appearance of creative work rather than the thing itself. In workplaces it overlaps with "AI slop" and the 2025 "workslop" research: content that looks finished but pushes the real work of fixing it onto colleagues.
Google's AI & Economy ATLAS report, published July 23, 2026, found that creative and other non-routine cognitive tasks appear in AI usage at nearly double their share of the broader economy, and framed AI mostly as a collaborative ideation partner rather than a replacement. Critics counter that heavy usage on creative tasks doesn't prove the output is actually good.
In an April 1, 2026 paper in the Human Resource Management Journal, Professor Dirk Lindebaum argued that outsourcing thinking to large language models erodes knowledge AI can't hold — hands-on experience, cultural judgment, and analytical reasoning — and can dull the skills that separate real creative work from slop. He proposed "learning vaults" to protect first-hand human learning at work.
Stop shipping the model's generic default. Define your own voice once and govern every output with it, keep visuals on-brand, and put a human review step before anything publishes. In Kompozy that's the Persona Brief, HyperFrames brand templates, and a per-post review pipeline — so what goes out sounds like you, not like a generic AI, across every platform.