SEO AI Blog Writers: How to Use AI for Blogs Without Getting Deindexed

OtterlyAI published a 2026 test that answers the question most AI blog writer users are asking. It launched two new domains about capybaras and loaded each with roughly 1,000 AI-generated posts in one click. One site, capybara-world.com, showed an author box presenting a fictional wildlife biologist. The other, capybaraverse.com, had no author at all.

Google removed both. Per Otterly, capybara-world.com lasted about 14 days in Google's index and capybaraverse.com about 90. Otterly reports zero manual actions in Search Console on either domain. The removal was algorithmic and silent.

The AI side split by engine. Bing never deindexed either site, and Microsoft Copilot, which grounds its answers in Bing, kept citing them. Otterly tracked 3,934 AI citations for capybara-world.com, and 2,718 of them (69%) came from Copilot. ChatGPT, Gemini, AI Overviews and AI Mode almost never cited the sites.

The test has real limits: one site per arm, no human-written control, a low-competition topic, and a vendor running it. Read the method and its limits before you quote it. What it does show is enough to shape a blog plan. A thousand thin pages on a new domain is a pattern Google acts on, and it sends no notice first.

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." The first example Google gives is "Using generative AI tools or other similar tools to generate many pages without adding value for users."

Google's guidance on AI content says the tool isn't the issue: "Appropriate use of AI or automation is not against our guidelines." Put the two statements together and the line is clear. Google doesn't ask who wrote the draft. It asks whether the page adds something for the reader.

The author box result fits here. The site with the fake expert was removed first. That's one data point, so it doesn't prove author boxes hurt. It does match Google's statement that E-E-A-T isn't a specific ranking factor. It gives no support to the idea that a byline protects thin content.

The next move: treat an AI blog writer as a drafting tool, not a publishing pipeline. If your workflow ships posts faster than anyone can add value to them, you're building the pattern Google describes.

A post clears several gates before any engine can quote it. Separating them shows where your effort counts.

• Parametric: what a model learned in training. It's frozen at a cutoff, and nothing you publish this quarter changes it.
• Access: whether a crawler can fetch and store the page. This is yours: robots.txt, firewall and CDN rules, rendering, and noindex or snippet tags.
• Retrieval: whether the page enters the model's context for a given query. The engine decides, and it often rewrites the user's question first.
• Generation: which retrieved pages the model uses and cites. The model decides.

Google states that to appear as a supporting link in AI Overviews or AI Mode, a page must be "indexed and eligible to be shown in Google Search with a snippet." That's why Otterly's sites dropped out of Google's AI surfaces once they were deindexed. No index, no citation.

Rendering matters for the other engines. Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, so text that only appears after scripts run may be invisible to them. Googlebot does render. Vercel's findings, digested, explain which crawlers are affected.

Indexing is necessary but not sufficient. In Otterly's test the sites ranked in Bing and ChatGPT's crawlers kept visiting, yet ChatGPT citations stayed effectively at zero.

Most AI blog advice blurs these two columns. Keep them apart and you'll stop spending time on things nobody can move.

Even a clean setup carries no guarantee. Google's AI features page says indexing and serving isn't guaranteed, and that there are no special optimizations needed to appear in AI Overviews or AI Mode. Your job is to remove blocks and give every post a reason to exist. The rest belongs to the engines.

Do these in order. The early steps remove blocks that make everything after them pointless.

1. Check access first. Confirm robots.txt allows the retrieval bots you care about. As of October 2026 that means OAI-SearchBot for ChatGPT search, PerplexityBot, Claude-SearchBot and Googlebot. GPTBot controls training only. Then confirm your CDN or firewall doesn't return 403 to those bots. The free AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.
2. Render the important text on the server. The post body, key numbers and headings should be in the HTML before JavaScript runs.
3. Pick topics you can add to. Before drafting, write down the specific fact, figure, price or experience the post will contain that a generic draft won't.
4. Draft with AI, then check every claim. Treat each sentence that states a fact as unverified until a person confirms it. A fact-check workflow for AI drafts keeps this repeatable.
5. State specific facts plainly. A price, a number or a one-line definition on a fetchable page gives an engine a sentence it can quote. That's the only on-page pattern with a positive experimental signal, and it still promises nothing. See how to write a post with AI citations in mind.
6. Publish at the pace you can edit. The evidence gives no safe number of posts per month. The constraint is value per page, so let editing capacity set the pace.
7. Check index status engine by engine. Look at Google Search Console and Bing Webmaster Tools separately. In Otterly's test, one index was healthy while the other had dropped both sites.

These are the errors that either break the access gate or waste effort on levers with no evidence behind them.

• Bulk publishing on a new domain. A thousand pages overnight on a domain with no history is the exact shape of Otterly's test. For more on that pattern, see what one mass-publishing test showed.
• Inventing an expert persona. The fictional biologist didn't protect anything. If a real person reviewed the post, say so. If not, don't invent one.
• Rewriting for an "AI-friendly" style. Google's optimization guide says you don't need to write in a specific way just for generative AI search, and that there's no ideal page length. C-SEO Bench found most conversational SEO rewriting methods were largely ineffective and often harmful to ranking in the model's context.
• Chasing long-tail variants. Google says you don't need to capture every variation of how someone might search. The original GEO paper found keyword stuffing lowered its own visibility metric.
• Treating schema or llms.txt as a citation lever. Google says structured data isn't required for generative AI search, and Google Search ignores llms.txt. Existing markup does no harm, so there's no need to remove it. It just isn't a lever.
• Reading one blended score. Copilot cited Otterly's site thousands of times while four other surfaces barely did. Measure each engine separately.
• Letting blog posts carry facts your product pages should own. If engines quote your blog for pricing, read why AI cites a blog post instead of a product page.

Since the tool only drafts, judge it on how well it supports the human work that follows.

• Source visibility. It should show where each factual claim came from, so a reviewer can check it fast.
• Room for your own material. You need to insert first-party data, prices, screenshots described in text and real examples without fighting the editor.
• Clean output. The text should land in your CMS as plain HTML that renders on the server.
• No bulk-publish default. A one-click button for hundreds of posts is a feature that points straight at scaled content abuse.
• No promised outcomes. Be wary of any tool that scores drafts for "AI-friendliness" or promises rankings or citations. Nobody controls what the engines cite.

For a fuller checklist, see what makes a good AI SEO writer tool. The short version: pick the tool that makes checking easy, not the one that makes publishing fast.