# I Produced Too Much AI Content. Here's What Changed My Mind

> Mass-producing AI content felt productive until the data showed it ranks fast and decays fast. Here's the quality-first content approach I use now.

Canonical URL: https://adityamallah.com/blog/i-produced-too-much-ai-content
Author: Aditya Mallah
Published: 2026-10-07T00:00:00.000Z
Updated: 2026-10-07T00:00:00.000Z
Topics: AI content, content strategy, SEO, content pruning, Google updates, B2B marketing

I shipped a lot of AI-assisted posts. For a stretch there, publishing felt like progress, drafts in hours, a calendar full of green squares, output climbing every month. Then I started watching what actually happened to that work, and it changed my mind about AI content quality.

Volume without a point of view stopped being an advantage. Scalable AI text can rank briefly and decay fast, while a small number of useful, opinionated pieces carry the results. That's the pattern I watched firsthand, and it's now exactly what the published research shows.

## What I did (and why it felt productive)

I treated publishing volume as the strategy itself, and it felt productive because every metric I checked was an output metric. Posts per week, drafts per session, keywords covered, all of them went up and to the right.

The workflow was seductive. Build a topical map, generate outlines, draft with a model, edit for grammar and flow, publish. Each post was decent: structured, readable, on-topic.

I put work into it. Drafts were structured, meta descriptions written, internal links placed. I did everything the playbooks said except the one thing they can't teach you: say something.

**AI slop** is mass-produced, low-effort content generated to capture search demand rather than help anyone. I told myself mine wasn't that, because a human touched every draft. But a polish pass isn't a point of view, and most of those posts had nothing on the page a competitor couldn't produce the same afternoon.

## What the evidence says

Mass AI content isn't banned, and it isn't automatically punished. It's something worse for anyone chasing volume: it mostly doesn't last.

In March 2026, Search Engine Land published a 16-month experiment run with SE Ranking's research team. They published 2,000 fully AI-generated, unedited articles across 20 brand-new domains and touched nothing. Early on, it worked: over 70% of the pages got indexed in the first month, impressions climbed past half a million, and plenty of pages cracked the top 100.

Then the floor gave way. By month three, only about 3% of the pages remained in the top 100, and roughly three quarters of all the visibility those sites ever earned came in the first 2.5 months. Over 16 months, 2,000 articles collected about 1.09 million impressions, and 1,381 clicks. That's the tollbooth pattern: seen from the road, never entered.

Two details from that experiment stuck with me. Indexation was worst in YMYL niches (the finance site kept just 9 of its 100 pages in Google's index, the health site 14) so what you write about is as consequential as how fast you write it. And when Google's August 2025 spam update hit, 10 of the 20 sites briefly spiked, with top-100 presence jumping from 3% to about 20%. Even the crashes weren't clean, which tells you these systems are noisy rather than surgical.

Researchers call this **content decay**, the gradual slide in rankings and traffic a page suffers as it ages, competitors refresh, and the topic moves on. The SE Ranking data shows decay in fast-forward.

A second 16-month study, published by Digital Applied in March 2026, tracked 4,200 articles across 140 domains in matched groups: pure AI, AI-assisted, and fully human. Pure AI articles ranked 23% lower than human ones at six months, with the gap widening from 14% at three months to 31% by month 16. They also earned 61% fewer editorial backlinks, nobody voluntarily cites filler. The AI-assisted group, where people did substantial rewriting and added original data, performed within 4% of fully human work.

Then there's the perception gap. A Semrush study covered by Search Engine Land in April 2026 analyzed around 42,000 blog pages and found human-written content took the #1 position 80% of the time versus 9% for purely AI pages, roughly an 8x difference. Meanwhile, 72% of surveyed SEOs believed AI content performs as well as or better than human content. Most of my peers, and past me, were grading our own homework.

Before I sound anti-AI: Ahrefs analyzed 600,000 top-20 ranking pages in July 2025 and found essentially no correlation between AI usage and rankings. Google doesn't punish content for being machine-made. But the composition of those top pages is the real finding, 81.9% are a mix of human and AI, just 4.6% are pure AI, and 13.5% are pure human. AI is everywhere in the winners; it's just rarely the whole winner.

A counterpoint: Ahrefs' June 2025 survey of 879 marketers found AI users publish 42% more content per month, and their sites grew about 5% faster year over year (29.08% versus 24.21% median). That's correlation, not causation, Ross Hudgens' read in the same report was that the extra volume is barely creating more performance. AI-assisted volume isn't poison. It's just not the engine.

Scale is the other thing that changed my mind. Graphite's October 2025 analysis of 43,000 articles found AI-generated articles overtook human-written ones in November 2024. Ahrefs found 74.2% of new webpages contained AI content by April 2025. When everyone can ship at machine speed, shipping isn't a moat.

Google's own words land on the same conclusion. Its February 2023 guidance said ranking systems reward helpful, reliable content however it's produced, and automation used to help people is fine. Then in March 2024 it named a new spam policy: scaled content abuse, targeting many pages generated primarily to manipulate rankings, with a specific example about using generative AI tools to generate many pages without adding value. The current spam policies page (updated August 2026) stresses the policy applies no matter how content is created, and violators may rank lower or not appear at all. The people-first guidance page (updated October 2026) lists warning signs like producing many topics hoping some rank well, and heavy automation across many topics. That's my old workflow, described by name.

## What I changed my mind about

Four beliefs quietly drove my production push, and all four turned out to be wrong or backwards.

1. **Velocity as a strategy.** **Content velocity** (how many pages you ship per week or month) is an output, not a strategy. It's a byproduct of having more to say than time to say it, not a dial you turn for rankings.
2. **Editing as a fix.** Light editing repairs grammar, not decisions. The AI-assisted articles that performed in the Digital Applied study had 30%-plus substantive rewriting, original data, and expert attribution, that's ownership, not proofreading.
3. **Topics chosen by volume.** Search volume tells you where the crowd is, not where you can win. I used to pick topics because a tool said they were big; now I pick topics where I can make a claim nobody else is making.
4. **Publishing cadence as a goal.** "Two posts a week" is a schedule, not an objective. Cadence is a constraint to plan around, not a KPI to hit.

## How I triage an AI-heavy library now

The fix for an AI-saturated blog isn't more publishing. It's an audit that concentrates effort on the pages that can actually earn their keep.

**Content pruning** is the process of auditing a site and keeping, merging, or removing pages based on the value they create for search, the business, and the reader. It sounds destructive, but the results say otherwise: a Search Engine Land case study published in October 2026 cut one-third of a B2B site's pages (most of which had earned almost no clicks) and Search Console clicks rose 37% and organic sessions 55% within months. The author's headline finding was blunt: "Nothing broke."

Here's the decision table I use now.

| Decision | Signs it belongs here | What I do |
|---|---|---|
| **Keep and refresh** | Ranks or has ranked in the top 20; has backlinks or conversions; covers something I actually know firsthand | Rewrite with current facts, my own data or experience, and a defensible opinion; update dates honestly |
| **Merge** | Overlaps a stronger page; answers a near-identical question; has a few links or small traffic | Fold the unique parts into the strongest page, redirect the rest with a 301, expand the survivor |
| **Kill** | Zero traffic, links, and conversions; duplicates something done better elsewhere; I can't state its value in one sentence | Delete or 410; if it might matter later, noindex and revisit, don't feed the index thin pages |

Part of that case study's traffic lift came from switching paid search off, so prune for focus, not for a guaranteed percentage. And don't delete anything with real links or conversions, the same study documents the mess that lazy redirects and "redirect debt" can leave behind.

The criteria that study used are a good sanity check for the table above. For each URL, ask: was this page ever meant to answer a search query, is it thin now but useful later (noindex it and revisit), does the business need it live, and does a closer-matching page already exist? If the answers point the wrong way in every direction, you've found a kill.

## Where AI still earns its place in my workflow

AI handles structure, speed, and variations. It never handles the point of view.

Concretely: outlines from my messy notes, first drafts I rewrite rather than edit, repurposing one strong piece into five formats, tightening sentences, and synthesizing research so I can check it against what I've seen in practice. What it doesn't do is choose the argument, supply the experience, or decide what's true.

The data backs this division of labor. Those 81.9% of top-ranking pages that mix human and AI work are the winning pattern: AI-assisted, human-led. Ahrefs' Ryan Law put the caution well in that June 2025 report, Google is not agnostic to bad content or bad actors, and AI makes it very easy to make bad content.

The Semrush survey found 70% of teams cite faster production as AI's top benefit, while only 19% say it improves quality. Everyone gets faster; almost nobody gets better. That's exactly why the edge moved to judgment.

So I produce less and say more. Every piece now has to answer one question before it earns a slot: what does a reader get here that they can't get from the other 74% of new pages that were made the same way? If I can't answer it, the draft gets merged into something better or killed. That's the whole strategy, fewer bets, made loudly, maintained properly.

## FAQ

**Does mass AI content still work in 2026?**
Briefly, sometimes, then it decays. The SE Ranking experiment showed unedited AI articles picking up real impressions for about ten weeks before collapsing to near-zero visibility. Raw volume still gets indexed; it just stops being shown.

**Does Google penalize AI content?**
Not for being AI. Google has said since February 2023 that it rewards quality however content is produced. What gets penalized, under the scaled content abuse policy from March 2024, is generating many pages primarily to manipulate rankings, regardless of whether a human or a model made them.

**How do I fix a blog full of AI content?**
Audit before you publish anything new. Sort every URL into keep, merge, or kill using the table above, refresh the keepers with original data and a real point of view, and redirect or remove the rest. In the October 2026 case study, cutting a third of a site lifted clicks 37%, but note their caveat that some of that lift came from turning paid search off.

**How much human editing does AI content need?**
More than a polish pass. The AI-assisted content that performed nearly as well as human work in the 16-month study got 30%-plus substantive rewriting, original data, or expert attribution. If your edit takes five minutes, you edited the wrong thing, start with the claim, not the commas.

**Is content velocity still a ranking strategy?**
No. John Mueller has said for years that publishing more doesn't make Google think a site is better quality (via Search Engine Journal, October 2021), and the 2026 reality makes it moot: with most new pages containing AI content, everyone is fast. Judgment is the edge now, not speed.

**What counts as AI slop?**
Volume without value. If a page exists to catch a query rather than to help the person typing it (no experience, no original input, no defensible claim) it's slop, whether a model wrote it or a human copied one.

## Sources

- [Google Search's guidance about AI-generated content](https://developers.google.com/search/blog/2023/02/google-search-and-ai-content), Google Search Central, February 2023
- [Spam policies for Google web search](https://developers.google.com/search/docs/essentials/spam-policies), Google Search Central, updated August 2026 (scaled content abuse)
- [Creating Helpful, Reliable, People-First Content](https://developers.google.com/search/docs/fundamentals/creating-helpful-content), Google Search Central, updated October 2026
- [How AI-generated content performs in Google Search](https://searchengineland.com/ai-generated-content-google-search-experiment-472234), Search Engine Land with SE Ranking, March 2026
- [Human content is 8x more likely than AI to rank #1 on Google](https://searchengineland.com/human-content-ai-rank-google-study-473697), Search Engine Land, April 2026 (Semrush study; SEL is owned by Semrush)
- [AI-Generated Content Does Not Hurt Your Google Rankings (600,000 Pages Analyzed)](https://ahrefs.com/blog/ai-generated-content-does-not-hurt-your-google-rankings), Ahrefs, July 2025
- [74% of New Webpages Include AI Content](https://ahrefs.com/blog/what-percentage-of-new-content-is-ai-generated), Ahrefs, May 2025
- [Websites Using AI Content Grow 5% Faster](https://ahrefs.com/blog/websites-using-ai-content-grow-faster), Ahrefs, June 2025
- [AI vs Human Content: 16-Month Google Ranking Study](https://www.digitalapplied.com/blog/ai-generated-vs-human-content-16-month-google-ranking-study), Digital Applied, March 2026
- [Content pruning case study: Cutting 33% of pages](https://searchengineland.com/content-pruning-case-study-497056), Search Engine Land, October 2026
- [More Articles Are Now Created by AI Than Humans](https://graphite.io/five-percent/research/more-articles-are-now-created-by-ai-than-humans), Graphite, October 2025

## Related reading

- [Why Your B2B Content Marketing Gets Traffic but Fails to Generate Pipeline](/blog/b2b-content-marketing-traffic-but-no-pipeline)
- [How to Create Content That Earns Citations in AI Answers](/blog/content-that-earns-citations-in-ai-answers)
- [AI Search vs Google Search: Rethinking Content Strategy for Both](/blog/ai-search-vs-google-search-content-strategy)

Source: https://adityamallah.com/blog/i-produced-too-much-ai-content
