Key takeaways
- Google's scaled content abuse policy led to manual actions that deindexed hundreds of sites in 2024-2025, some of which had been pulling millions of monthly visits
- Trust in brands that lean on AI content has cratered: consumers saying heavy AI use would decrease their trust in a favorite brand roughly doubled from 20% in 2025 to 40% in 2026
- AI search engines are pulling back from low-effort formats. ChatGPT's citation share for social posts (a proxy for thin, UGC-style content) fell from 4.4% to under 1% in a single week in August 2026
- Named incidents at Sports Illustrated, CNET, and Gannett show the reputational cost is immediate and expensive: Arena Group's stock dropped 27% in a day after its AI-author scandal broke
- The fix isn't abandoning AI, it's adding a human review gate, refreshing decaying pages, and tracking where you actually show up in AI answers rather than guessing
What actually happened in 2025
A lot of brands read the room wrong. ChatGPT launched, everyone panicked about being left behind in "AI search," and the fastest available move was to point an AI writer at a keyword list and hit publish. Ahrefs analyzed 900,000 newly published English-language pages in April 2025 and found that 74.2% contained AI-generated content of some kind, with only 2.5% being fully AI and about 72% a human/AI mix. A separate survey of 879 content marketers in the same research found 87% were using AI to create content, and only 13% weren't touching it at all.
Volume went up. Quality, mostly, did not. And 2025 turned out to be the year both Google and the AI search engines started pushing back, hard, in ways that showed up in traffic dashboards and brand reputation within months.
I want to be specific here instead of vague, because "AI slop is bad" isn't a useful sentence on its own. What follows is what it actually cost, in rankings, in citations, and in the kind of trust that doesn't come back with an apology post.
The ranking losses: deindexing, not just demotion
Google's March 2024 core update introduced three spam policies aimed directly at this behavior: scaled content abuse, expired domain abuse, and site reputation abuse. Scaled content abuse specifically covers pages "created purely for search engines," and Google was explicit that it doesn't matter whether the content is human-written, AI-generated, or a mix, if the intent is to game rankings at volume, it counts.
The enforcement wasn't theoretical. Google issued manual actions that deindexed hundreds of sites, including some that had previously pulled millions of monthly visits. Independent publisher HouseFresh documented a 91% search traffic loss tied to the update, and its December 2025 follow-up investigation found something even more damning: 81% of the sites Google had publicly held up as "success stories" had actually lost traffic since 2023, and 43% of the independent creators it analyzed lost 80% or more of their Google Search traffic between 2023 and 2025.
That last number is worth sitting with. This wasn't a slow fade. For nearly half the sites in that sample, it was closer to a cliff.
The citation losses: AI search is getting pickier, fast
Here's the part most brand teams missed because they were watching Google Search Console and not AI answer engines. AI platforms have their own version of this correction, and it's happening on a much faster timeline.
Promptwatch's data on Reddit citations shows ChatGPT Search's citation share for reddit.com collapsing from a steady ~3.8% in mid-July 2026 to just 0.5% by mid-August, an 86% relative drop in a matter of days. Google's AI Overviews and AI Mode made similar corrections but far more gradually. Whatever triggered it, the pattern lines up with a broader shift away from low-signal content: Promptwatch's citation type data for August 2026 shows social posts falling from 4.4% of ChatGPT citations to under 1% in the same window, while how-tos more than doubled and documentation nearly tripled its citation share.
There's also a structural squeeze happening that makes every citation slot more valuable and every wasted one more costly. After the GPT-5.3 rollout in March 2026, ChatGPT's average sources per response dropped by roughly 27%, from about 6.4 down to 4.7-4.9, and it never recovered. Promptwatch's average sources per response data puts ChatGPT at around 5 sources per answer, versus roughly 10 for Google AI Overviews and Perplexity. When an AI engine is only citing 5 sources total, publishing generic content that doesn't earn one of those slots isn't neutral, it's a wasted production cost with nothing to show for it.
And the domain-authority story isn't what most brands assume either. Promptwatch's citation share by domain rank data shows DR 61-90 sites holding roughly 42% of ChatGPT citations through August 2026, while DR 91-100 sites actually fell from about 7% to 3% share over the same period. You don't need to be a media giant to get cited. You do need actual substance, which is precisely what mass-produced AI slop skips.
The trust losses: this is the part that compounds
The ranking and citation losses are recoverable, painfully, over months. The trust losses are harder to walk back, and the 2025-to-2026 data on this is the starkest in the whole picture.
Fractl ran an identical consumer survey in Q2 2025 and Q2 2026, and the year-over-year swing is dramatic. The share of consumers saying heavy AI use by a brand would decrease their trust in that brand roughly doubled, from 20% to 40%, with 14% now saying it would decrease trust "significantly." Meanwhile the share of consumers who find AI more helpful than traditional search dropped from 82% to 54%, a 28-point fall in a single year. People got burned, and they adjusted.
The generational split matters for brand strategy too: 54% of Gen Z say trust would decrease if a favorite brand used AI for most of its marketing, compared with roughly a third of Gen X and Boomers. The audience most fluent in AI is also the least forgiving of it being used carelessly on them.
And there's a labeling demand that's nearly unanimous: 84% of consumers want AI-written content labeled as such, with similar or higher numbers for video, images, and audio. Brands that hid AI authorship in 2023 and 2024, and several did, learned this the hard way.
The named incidents: what it actually cost specific brands
Abstract statistics are useful, but the individual cases are what stick.
Sports Illustrated, published under The Arena Group, was caught in late 2023 running product reviews under fake AI-generated author personas complete with AI-generated headshots. When journalists asked about it, the fake author profiles were quietly deleted with no explanation. Arena Group's stock dropped 27% in a single day, and the CEO was fired weeks later. The same publisher's other properties weren't clean either: Men's Journal had to issue a "massive correction" on its first AI-generated health article.
CNET, owned by Red Ventures, was caught running AI-written finance articles under a vague "CNET Money Staff" byline. In the backlash, CNET ended up issuing corrections on more than half of its AI-generated articles, some of which contained plagiarism.
Gannett paused an AI experiment generating high-school sports recaps in 2023 after errors surfaced, and one of the company's own sports journalists publicly called the output embarrassing. These aren't small local blogs. These are large, well-resourced newsrooms that still got the execution wrong, which tells you the risk isn't really about budget.
The hidden productivity cost nobody budgets for
There's a second layer of cost that rarely makes it into a board deck: the time spent cleaning up after AI content, whether it's published slop or internal "workslop" that looks finished but isn't. Harvard Business Review's September 2025 research coined the term "workslop" for exactly this, and found 40% of employees had received AI-generated work in the prior month that looked polished but lacked real substance. The same research, covered by Axios, estimated each workslop incident costs an average of 1 hour 56 minutes to fix, adding up to roughly $186 per employee per month, which for a large company can top $9 million a year.
Separate research from Freshworks put UK businesses at £11.7 billion a year spent correcting AI slop, with a quarter of work hours wasted on it, and found mid-market companies losing about 25% of their AI budget to what it calls "complexity overhead" before they see any return at all.
What actually works instead
None of this means AI-assisted content is the problem. The Ahrefs data shows 72% of new content is already a human/AI mix, and plenty of that ranks fine. The distinction that matters is whether a knowledgeable human is actually shaping and checking the output before it ships, or whether AI output goes straight from generation to publish with nobody looking at it.
| Approach | What it looks like | Typical outcome |
|---|---|---|
| Raw AI output, no review | Quota-driven publishing, generic "ultimate guides," no fact-check | Ranking decay within 2-6 months, citation share near zero, trust erosion |
| AI draft + human review gate | Subject-matter expert shapes voice, verifies claims, decides what's worth publishing | Retains rankings, earns citations in AI answers, keeps reader trust |
| Human-first, AI-assisted research | Writer uses AI for research/structure, writes and edits themselves | Slower, but highest trust and citation durability |
A few concrete moves that actually move the needle:
Audit and refresh before you prune. Content doesn't just decay on its own timeline; Semrush data cited in recent content-decay research suggests 82% of high-ranking blog posts start losing traffic within 12-24 months without maintenance, and pages updated within the last 60 days are roughly 1.9x more likely to appear in AI answers at all. A 2020 post refreshed from 926 to 2,099 words saw its CTR jump tenfold in one agency case study. Refreshing thin AI pages with real substance is often cheaper than writing new ones and safer than deleting them outright.
Stop guessing at your AI visibility and start measuring it. If you don't know which of your pages are actually getting cited by ChatGPT, Perplexity, or Google AI Overviews, you're optimizing blind. Tools like Promptwatch track citation trends, crawler activity, and content gaps across AI platforms so you can see which pages earn citations and which ones are just sitting there as dead weight.
If you're comparing content optimization platforms more broadly, tools like Surfer SEO, Frase, and Clearscope help ground AI-assisted drafts in real search intent data rather than letting a model free-associate a generic outline.
For teams evaluating a wider set of GEO and AI-visibility tools, the directory at bestgeosoftware.com is a reasonable place to compare options side by side.
Building the review gate that actually holds
A human review gate isn't a vague commitment, it's a specific checklist someone owns. At minimum it should cover: does a subject-matter expert actually agree with every claim in the draft, are the statistics traceable to a real source rather than a model's best guess, does the piece say something the last ten "ultimate guides" on this topic didn't already say, and would this person's name on the byline survive scrutiny if a reader fact-checked it.
That last question is the one most mass-publishing operations skip, and it's the one that determined whether Sports Illustrated's scandal happened or didn't. If you wouldn't put a real editor's name behind a piece, that's the signal it isn't ready, no matter how fast it was to produce.
The brands still standing after the 2025 slop correction aren't the ones that avoided AI entirely. They're the ones that treated AI as a drafting tool, kept a real person making the final call on every published page, and started actually tracking what AI engines cite instead of assuming more pages automatically meant more visibility. The data from the last eighteen months makes it pretty clear which bet paid off.