Mass-Publishing AI Content: What One Two-Site Experiment Showed

OtterlyAI, a company that sells AI visibility tracking, built two new domains about capybaras in 2026: capybaraverse.com and capybara-world.com. Otterly reports loading each site with roughly 1,000 AI-generated blog posts in one click, about 2,000 posts in total.

One variable was meant to differ. capybara-world.com showed an author box under every post, presenting a fictional wildlife biologist with more than 15 years of field experience. capybaraverse.com had no author at all.

Otterly tracked four layers for about four months:

• Google Search Console
• Bing Webmaster Tools
• AI citations across seven engines: ChatGPT, Microsoft Copilot, Claude, Perplexity, Gemini, Google AI Overviews and AI Mode
• AI crawler visits

The write-up was last updated on 2026-08-05 with a fourth month of data. Our breakdown of the experiment's method and limits charts every figure Otterly published. This guide focuses on what the result means for your own site.

Google removed both sites. Otterly reports zero manual actions in Search Console on either domain, so the removal was algorithmic, and Google does not say which system did it. If you are unsure how the two kinds of removal differ, see manual action vs algorithmic deindexing.

The order surprised people. The site with the fictional expert went first, about 14 days in. The site with no author lasted about 90 days.

The collapses were sharp. Otterly's write-up shows capybara-world.com peaking at 2,426 daily Google impressions on April 5, then falling from 1,629 to 15 on April 9. capybaraverse.com built up to more than 21,000 weekly impressions by mid-May, then fell from 859 to 85 daily impressions on June 26.

None of this contradicts Google's written policy. It is the policy. Google's spam policies define scaled content abuse as being "when many pages are generated for the primary purpose of manipulating search rankings and not helping users." Google's guidance on generative AI content is equally clear that the tool is not the problem: "Appropriate use of AI or automation is not against our guidelines."

What it means: the violation is volume without value, not the model that wrote the text. A thousand thin pages on a new domain is the pattern Google describes, and removal can arrive with no notice.

Losing Google did not end the sites' AI visibility. Otterly tracked 100 AI citations for capybara-world.com in April and 1,666 in the first 27 days of July, about 17 times as many, while the site was gone from Google's organic results.

Most of those citations came from one engine. Of the 3,934 citations Otterly tracked for capybara-world.com, 2,718 (69%) came from Microsoft Copilot. Perplexity and Claude cited the sites too. ChatGPT, Gemini, AI Overviews and AI Mode almost never did.

One more figure needs care. Otterly reports that Microsoft's own AI Performance report credited capybara-world.com with 42,817 citations, more than ten times what its prompt-based tracker logged. The time windows may not match, so it is not an exact ratio. It does show that any prompt sample is only a sample.

AI answers that cite the web pass through four stages. A crawler has to reach the page (access). The engine has to store it (index). The engine has to pull it into the answer's context for a given question (retrieval). Then the model decides what to use and cite (generation).

The experiment shows the index stage doing the work. Google states that a page must be "indexed and eligible to be shown in Google Search with a snippet" to appear as a supporting link in AI Overviews or AI Mode. Google removed these pages, and its AI surfaces stopped citing them. Bing kept them, and Copilot, which grounds on Bing, kept citing them.

The ChatGPT result is the second lesson. "ChatGPT uses Bing" gets repeated as a strategy. In this test, ranking in Bing produced almost no ChatGPT citations, even with ChatGPT's crawlers on the site. Being indexed is necessary. It is not enough.

As of October 2026, retrieval is also discretionary on every major platform. Engines decide per question whether to search at all, and they rewrite the question before searching. That is why no index status, on any engine, guarantees a citation.

Otterly ran this in public and published the data, which is more than most AI content advice does. It is still one pair of sites, and the conclusions going around claim more than one pair can carry.

• One site per arm. The author box was not the only difference. The sites had different domain names and everything that comes with a separate domain. Nothing separates the author box from that noise.
• No human-written control. Without a site publishing 1,000 human-written posts at once, the test cannot say whether Google reacted to AI text or to a thousand pages appearing overnight on a domain with no history. Google's policy targets the second, however the content is created.
• The author box is one data point. A fake expert going first is striking. It fits Google's statement that E-E-A-T "isn't a specific ranking factor." It does not show that author boxes hurt.
• A low-stakes topic. Capybara questions are informational with little competition. The result says little about whether Copilot would cite a thin page over established sources in a commercial category.
• Citation counts depend on the prompts. A tracker counts answers to the prompts it asks. Until the prompt list and raw counts are public, nobody outside can separate engine behavior from prompt design.
• A vendor ran it. Otterly sells AI visibility tracking, and so does 99Visibility. Neither fact makes the data wrong. It is why replication matters.

In your control

• What you publish, how much, and how fast. Volume without value is the pattern Google's policy names.
• Whether AI crawlers can reach you: per-bot robots.txt rules, and whether your CDN or firewall lets those bots through.
• On-page controls like noindex, nosnippet and data-nosnippet, which keep content out of search and AI answers when applied, including by accident.
• Watching index status in Google Search Console and Bing Webmaster Tools.
• Stating specific facts (a price, a number, a one-line definition) on pages that engines can fetch.

Not in your control

• Which Google system removes a site, and whether it tells you. In this test, it did not.
• Whether an engine searches the web for a given question, and which queries it actually runs.
• Which retrieved pages the model chooses to cite.
• Whether one engine's continued citations last. Copilot citing a deindexed site is an observation from one test, not a fallback plan.

Access comes before indexing, and indexing comes before citation. Work in that order.

1. Confirm crawlers can reach you. A robots.txt that allows a bot while your firewall returns 403 removes you as quietly as a deindex. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away. Remember the rules are per bot: GPTBot governs training, while OAI-SearchBot controls ChatGPT search.
2. Check index status in each engine separately. Look at Google Search Console and Bing Webmaster Tools on their own. One being healthy tells you nothing about the other. The playbook on checking index status engine by engine walks through both. If Google traffic fell off a cliff, start with how to tell if Google has deindexed your site.
3. Gate any scaled content through review. If you generate pages with AI, each one should add something a reader cannot get elsewhere. Use a pre-publish review for scaled AI content before anything goes live in bulk.
4. Publish fewer, more specific pages. A fetchable page that states your price or a clear definition gives an engine a sentence it can quote. See publishing specific, fetchable facts.
5. Measure each AI engine on its own. In this test one engine cited the sites thousands of times while four barely did. A single blended number would have hidden that.

• Treating Copilot citations as a substitute for Google. Google is still the largest search channel. Losing it silently is not offset by one engine that grounds on a different index.
• Adding a fake expert as protection. The site with a fictional biologist was removed first. Invented credentials are a misrepresentation, and nothing in this test suggests they shield scaled content.
• Assuming Bing rankings mean ChatGPT citations. The sites ranked in Bing and ChatGPT barely cited them.
• Waiting for a Search Console warning. Both sites were removed with zero manual actions. Monitor impressions yourself.
• Reading the test as proof that Google penalizes AI text. Google says appropriate use of AI is fine. The test cannot separate AI authorship from volume on a brand-new domain.
• Trusting one tracker's count as the total. Microsoft's own report counted far more citations than the prompt-based tracker. Where an engine publishes its own numbers, read them alongside any sample.

Does Google penalize AI-generated content? Not for being AI-generated. Google's guidance says appropriate use of AI or automation is not against its guidelines. It does act against scaled content abuse, meaning many pages produced to manipulate rankings without helping users. In Otterly's test, Google removed both 1,000-post AI sites from its index, and Otterly reports no manual action on either. The test cannot say whether AI text or sheer volume triggered the removal.

Can AI search engines cite a page Google has deindexed? Yes, if they draw on a different index. In Otterly's test, Microsoft Copilot, which grounds its answers in Bing, kept citing the sites after Google removed them, and Perplexity and Claude cited them too. Google's own AI features, AI Overviews, AI Mode and Gemini, almost never did, because Google requires a page to be indexed and snippet-eligible to appear as a supporting link.

Does an author box protect AI-generated content? Nothing in this test suggests it does. The site with a fictional expert author box was deindexed after about 14 days, and the site without one lasted about 90. With one site per arm, that does not prove author boxes hurt. It is simply no evidence that they help. Google also states that E-E-A-T is not a specific ranking factor, so there is no score an author box raises.

If my site is in Bing, will ChatGPT cite it? Not necessarily. In Otterly's test, both sites stayed in Bing and ChatGPT's crawlers kept visiting, yet ChatGPT citations were effectively zero. ChatGPT search is controlled by OAI-SearchBot access and by the model's own choices about when to search and what to cite. Treat Bing indexing as one requirement for Copilot, not as a route into ChatGPT answers.

How many AI articles is too many to publish at once? No published number answers this, and Google gives none. Its policy turns on purpose and value, not a page count: many pages generated mainly to manipulate rankings without helping users. Otterly's sites published about 1,000 posts each on brand-new domains, which is the clearest version of the pattern. The safer test for any batch is whether each page adds something a reader would miss if it were gone.