This plugin can hurt your site.

1. Publishing at scale without reading

This is risk number one, and it is far ahead of the rest.

Google has a spam policy covering scaled content abuse: producing many pages primarily to manipulate rankings rather than to help anyone. The policy says nothing about which tool you used. It is about the outcome. Hand-writing 500 worthless pages falls under it exactly as generating 500 does.

What makes autoblogging dangerous is simply throughput: it becomes trivial to produce the volume that triggers the policy, without ever having the reviewing capacity that avoids it.

How to avoid it: leave “What happens to a finished article” on Draft. Set the daily cap to what you will actually review. Use the review queue — every publication takes a deliberate click.

2. Thin content

A model will happily produce 1,500 words that say nothing: generalities, restatements of the question, “it is important to note that”. It is long without being substantial, and that is exactly what ranking systems devalue.

The test is simple and brutal: does your article say one thing a reader would not find in the first five results? If not, it does not deserve to be published — whatever its length.

How to avoid it: the default prompt demands one concrete specific per section (a number, a step, a named tool, a trade-off). Then add what you know and the model cannot: your experience, your screenshots, your measurements, your real prices.

3. Duplication and cannibalisation

Two separate dangers:

• Internal duplication — two articles covering the same subject under two titles. They split the signals and neither ranks.
• External duplication — the model rephrases what already exists. Nothing justifies ranking you above the source.

How to avoid it: the plugin refuses titles it has already used and tracks written topics. But the real protection is upstream: read the topic cluster before generating and delete the overlapping lines. Two topics one article could answer are one topic.

4. Invented facts

A language model does not know what is true. It produces plausible text, which is a very different thing. It will invent statistics, dates, prices, studies and quotes, in exactly the same confident register it uses for accurate ones.

How to avoid it: the review queue flags mentions of statistics, studies and prices so you can check them one by one. The default prompt explicitly tells the model to describe a mechanism rather than guess a figure. That reduces the problem; it does not remove it.

5. Sensitive topics (YMYL)

Health, finance, law, safety, major life decisions: these are what Google calls Your Money or Your Life topics, and they are judged far more harshly on expertise and trustworthiness.

They are also, almost always, the most lucrative niches — which is precisely why they attract the people who should not be in them.

Direct recommendation: do not automate YMYL writing without real domain expertise. If you cannot spot a medical or tax error while reviewing, you cannot review, and the entire safety loop is broken.

6. Undisclosed affiliate links

Two separate obligations, often confused:

• Legal — advertising must be identifiable as advertising. In the US, the EU, the UK and most other jurisdictions, an unlabelled paid recommendation is a deceptive commercial practice.
• Search engine — a commercial link must carry rel="sponsored" or rel="nofollow". Without it you are passing a ranking signal you were paid to pass.

How to avoid it: list your affiliate domains in the settings — matching links get rel="sponsored nofollow" automatically. And keep the disclosure line switched on.

7. Technical damage

• A drained API balance. “Every 5 minutes” is 288 paid articles a day. The ceilings exist for this; do not set them to zero.
• A sitemap bloated with worthless pages dilutes crawl budget: engines spend their time on filler instead of your good pages.
• An exploding RSS feed. Fifty articles at once is fifty emails to your subscribers. Space them out with “Schedule for later”.
• The external cron URL is a password. Whoever holds it can spend your credit.

What is NOT a risk

To be fair in both directions:

• Using AI to write. Google has said explicitly that appropriate use of AI is not against its guidelines. Quality and usefulness are judged, not the production method.
• Publishing often. A steady cadence is not suspicious in itself. A steady cadence of thin content is.
• Disclosing that AI was used. Transparency is not a demotion factor. It is good for the reader, and that is the whole argument.
• “AI detectors”. They are unreliable, Google publishes none, and optimising to fool them is wasted effort. Optimise for being useful.

The settings that protect you

Signs it is going wrong

Watch Search Console. Warning signs, in order of severity:

1. A manual action. Unambiguous: unpublish, fix, request a review.
2. Impressions falling site-wide after a burst of publishing — not just on the new pages. This is the most worrying signal, because it is hitting what used to work.
3. Pages crawled but not indexed, in a rising proportion. The engine sees them and decides they are not worth the space.
4. Click-through rate collapsing at the same position: your titles promise what the content does not deliver.

---

Search engine policies change. What is written here describes the mechanisms as understood at this page's update date; always check Google's official spam and helpful-content documentation before making a decision that commits a site.

Best practices → Monetise it cleanly