When everyone produces the same thing, the only scarce resource is being different
Half the articles online are written by AI and the models are homogeneous. What the data says about brand, trust and visibility.
- Date
- August 11, 2026
- Category
- Brand
- Reading
- 10 min read
- Byline
- Outsmart Studio
AI does not lower the average quality of content: it raises it. The problem is that it lowers variance, and variance is exactly what a brand lives on.
Since the start of 2025, roughly half of all articles published online have been predominantly AI-generated: 50.9% in the first quarter of 2026, 49.9% in the next, on a stable plateau (Graphite, 55,400 articles from Common Crawl, three independent detectors). But only 14% of Google search results and 18% of citations from ChatGPT and Perplexity are AI content. Producing more does not produce visibility: scarcity has moved from content to distinctiveness.
The problem is not AI's quality: it is variance
There is a misunderstanding that makes half the discussions on this subject pointless. Generative AI does not lower the average quality of content: it raises it. It raises the floor. The problem is that it lowers the ceiling on variance, and variance is exactly what a brand lives on.
Doshi and Hauser, in Science Advances (16 July 2024): 300 participants write micro-stories, 600 evaluators judge them. With AI assistance, individual novelty rises by 8.1% and usefulness by 9%, and the less creative gain the most. But the assisted group's texts are 10.7% more similar to each other.
The originality of a single output is comparable; it is population variance that collapses. And it holds across every model family. In market terms: if all your competitors use the same tools, everyone gets a little better and everyone gets a lot more alike.
The sameness is documented
Alex Murrell calls it “the age of average”: luxury wordmarks now indistinguishable, car logos flattened until they coincide, twenty-seven different brands using the same “Find Your…” claim. Meanwhile Merriam-Webster made “slop” its word of the year for 2025, defining it as “low-quality digital content usually produced in quantity by artificial intelligence”. In twelve months the market has equipped itself with a derogatory term for recognising what you are about to publish.
What happens when consumers see the label
Here is the most useful finding in the whole literature, and it cuts against both easy narratives. Bynder showed 2,000 consumers two 300-word articles, one by ChatGPT and one by a copywriter, without saying which was which: 56% preferred the AI one. But 52% say they disengage from content they suspect was generated, and when they suspect it they associate words like “impersonal” and “lazy” with the brand.
The penalty is not on quality. It is on attribution.
Recent data confirms the direction. DoubleVerify, 22,000 consumers across 22 markets, July 2026: 42% say a low-quality AI advert would worsen their opinion of the brand, but 40% view polished ones favourably. Klaviyo with Datalily, December 2025: only 7% say visible AI increases trust, 31% say it decreases it. And over 40% of ads with AI integrated invisibly reach the highest levels of brand recognition.
On top of this comes a regulatory fact: from 2 August 2026, Article 50 of the European AI Act makes transparency on synthetic content and deepfakes mandatory. The perception penalty, which until now could be avoided by staying quiet, is becoming structural.
The case that explains everything: Coca-Cola, November 2025
Coca-Cola's 2025 Christmas spot was made by five AI specialists, from over 70,000 generated clips, in thirty days. On the internet it was torn apart. Then System1 tested it on a real audience: 5.9 stars out of 5.9, the top of the scale. The viewers' spontaneous associations? The animals, Christmas, Santa Claus, the trucks, the music. In other words, the distinctive assets built over decades.
Coca-Cola got away with it because it owns a distinctive estate strong enough to absorb a mediocre execution.
The counter-example came a month later: McDonald's Netherlands pulled its entirely AI-generated Christmas spot after the backlash. And a third case deserves mention for honesty: Duolingo announced an “AI-first” strategy in April 2025, took a heavy backlash, and in August closed a quarter with daily active users up 40%. Anyone writing that “AI destroys brands” is overstating it.
Why brand matters more right now
Ahrefs correlated dozens of factors with the visibility of 75,000 brands in AI answers. At the top: mentions on YouTube, around 0.74, and brand mentions across the web, 0.66-0.71. At the bottom, backlinks, with minimal impact. Brands in the bottom half for mentions received 0-3 citations; those in the top quartile, 169 on average.
Edelman reaches the same conclusion from an entirely different direction: 88% consider trust in the brand important or decisive when buying, on a par with quality and price. And it notes that “what others say, unpaid by the brand, counts far more than what the brand says about itself”. Two different methodologies, one conclusion: what counts is how present your name is in the world, not what you declare on your own site.
The counter-argument, which has to be said
It would be dishonest to stop here. A June 2026 preprint, around 23,000 API calls across GPT-4o-mini, Claude Sonnet and Gemini 3 Flash, shows that at identical specifications the models recommend the real brand almost every time, but that product parameters explain 82.4% of the ranking and brand identity only 1.2%: an advantage of 0.075 stars is enough to overturn everything. It is a non-peer-reviewed preprint, on synthetic products. But the message is useful: brand is an extremely powerful tie-breaker between equal products, not a substitute for substance.
- System1 with Effie Worldwide
- 56 brands, over 4,000 creative assets. Those using distinctive assets for a year score 2.4 stars for effectiveness; after four or five years of consistent use, 3.5. And recognition speed is 86% with owned distinctive assets against 81% with a celebrity: an expensive endorser performs worse than something you own.
- IPA, Binet and Davis
- Budget explains 89% of the variation in profit, ROI only 11%. And yet 65% of marketers prioritise ROI. Since Covid, average ROI has risen by 4% while real incremental profit has fallen by 11%.
- Kantar BrandZ 2026
- In an environment where “machines increasingly surface and weigh content”, standing out as meaningful and different has become more important for brands, not less.
And for a Swiss company?
Regular use of generative AI among 15-64 year olds in Switzerland went from 34.8% at the end of 2025 to 37.8% in the first quarter of 2026, against a world average of 17.8%. It is one of the markets most exposed to saturation. Provenance, craft, signed accountability: in Switzerland these are not brand rhetoric, they are measurable assets. And they are, by definition, what a generative model cannot produce in your place.
- Define and protect your distinctive assets
- Colour, shape, voice, character, sound. Do not reinvent them every two years: multi-year consistency is what the System1 figure measures.
- Use AI where it replicates, not where it invents
- Unilever applied digital twins to product image production: −55% costs and −65% time in Beauty & Wellbeing. AI industrialises a distinctiveness already defined, it does not create it.
- If the AI shows, it has already lost
- The dividing line Kantar measures is not whether you use it: it is whether it is invisible in the result.
- Move budget from bought mentions to earned ones
- It is what both people and models read.
- Do not hide the label: make it irrelevant
- From 2 August 2026 transparency is mandatory in Europe. The only sustainable answer is that, underneath the label, there is something only you could have made.
- Is AI really saturating the web with content?
- Around half of published articles have been predominantly AI-generated since the start of 2025. But only 14% of Google search results and 18% of chatbot citations are AI content: volume does not translate into visibility.
- Do consumers recognise generated content?
- Badly: around half of people fail to identify it in a blind test. But when they suspect it, their judgement of the brand drops sharply.
- Is it worth declaring the use of AI?
- In Europe, from 2 August 2026, for deepfakes and synthetic content it is no longer a choice. The lever you can work on is quality: ads with AI integrated invisibly perform far better.
- Does brand help you get recommended by AI assistants?
- Brand mentions across the web show the highest correlation with visibility in AI answers, far more than backlinks. But between equal products: brand is a tie-breaker, not a substitute for substance.
- How long does a distinctive asset take to produce effects?
- According to System1, average effectiveness goes from 2.4 to 3.5 stars between the first and the fourth or fifth year of consistent use. It is a compounding return, not a campaign result.
Sources: Graphite, Q1 2026 · Axios, 14 October 2025 · Doshi & Hauser, Science Advances, 16 July 2024 · Wenger & Kenett, PNAS Nexus, 24 March 2026 · Alex Murrell, Branding Strategy Insider, 8 July 2026 · TechCrunch, 15 December 2025 · Bynder, April 2024 · DoubleVerify, 29 July 2026 · Klaviyo & Datalily via eMarketer, December 2025 · Kantar · Ahrefs, 12 December 2025 · Edelman, Trust Barometer 2026 · Chu & Hou, arXiv:2606.17443 (preprint) · System1 × Effie, 18 February 2026 · IPA, Binet & Davis, 8 October 2025 · Kantar BrandZ 2026 · System1, 20 November 2025 · Unilever, 18 March 2025 · Risiko-Dialog Foundation, Digital Barometer 2026 · Microsoft, Global Diffusion, May 2026.
