Best practices — getting value out of an autoblogging plugin

Short answer

How do you use an autoblogging plugin without damaging your site?

Keep draft mode on, read every article, and publish only what you would have signed yourself. Add first-hand experience the AI does not have, set spend ceilings, and publish less but better: thirty reviewed articles beat three hundred ignored ones.

1. Choose the niche

This decides everything else, and it is the one decision the plugin cannot make for you.

Three criteria, in this order:

You can fact-check it. If you cannot spot an error in an article in the niche, you cannot run it. This eliminates more candidates than the other two combined, and it is not negotiable.
The competition is not locked. If the top ten are national brands and institutions, a new domain has no path. Look for queries where ordinary blogs still appear.
Commercial intent exists somewhere. Not on every article, but the niche must contain topics where the reader buys, compares or chooses.

This section is the summary. The full method — the ten-minute competition test, the ten tools, the scoring sheet and the translation strategy — is on the niches page.

The sub-niche test

“Cooking” is not a niche, it is a library. “Sous-vide cooking for beginners” is a niche: narrow enough to cover completely in 60–100 articles, broad enough that 60–100 real topics exist.

Niches to avoid without real expertise: health, medication, personal finance, law, tax, investing, gambling. These are YMYL topics, judged far more harshly — see the risks page.

2. Build the topic cluster

A cluster is a set of articles that cover a subject completely, linked to each other. It is the only structure that lets a new site rank: you do not beat an established authority on one isolated article, but you can cover a question better than anyone.

Write the seed
Specific, not broad. “Espresso at home for beginners”, not “coffee”.
Generate 60–100 topics
The plugin de-duplicates. Budget a minute and a few cents.
Read the list — the step everyone skips
Delete overlapping topics: if one article could answer two lines, it is one line. Delete what you cannot verify. Add what the model missed: the questions your customers actually ask.
Identify the 5–10 pillars
The broad topics that deserve a long, careful article. They will receive links from the others, and they are the ones you write most seriously — possibly by hand.

3. Tune the prompt

The default prompt is a decent starting point, not an endpoint. Three additions change everything:

The reader. “A beginner buying their first machine under $300” produces radically different text from “an interested reader”.
What not to do. Prohibitions work better than positive instructions: no conclusion that restates, no “in today's world”, no invented figures.
Your angle. What you know that the competition does not say. It is the only thing that makes the article defensible.

Available placeholders: {topic} {site_name} {language} {audience} {tone} {word_count}.

4. The first article

Generate one. Read all of it. Then answer honestly:

Would I publish this under my own name?
Does it say one thing not found in the first five results?
How long does it take me to make it good?

The third question matters most: that is your real cost per article. If it is 45 minutes, your sustainable pace is 2–3 a day, not 20 — and the whole plan has to be rebuilt on that number.

5. Review fast and well

An efficient review always follows the same order, worst problems first:

Order	What you check	Time
1	Facts — figures, dates, prices, names, studies. Each against a source.	~50%
2	Structure — is the title's promise delivered in the first two sentences?	~15%
3	Emptiness — delete paragraphs that say nothing. Almost always 10–20% of the text.	~15%
4	Human addition — an experience, a measurement, a screenshot, an opinion.	~15%
5	Links — relevant internal ones, external ones to the real sources.	~5%

The review queue pre-flags the common problems: short articles, statistic mentions, prices, model boilerplate, missing H2s, missing image, no links. It directs attention; it does not replace reading.

6. The human touch — what to actually add

This is what separates a site that lasts from one that vanishes at the next algorithm update. A model can produce none of the following:

Direct experience — “I used this machine for six months; the drip tray leaks.”
Original visuals — your photos, screenshots, diagrams. One authentic image beats ten generic illustrations.
Current figures — real prices checked today, with the date on them.
A definite opinion — “do not buy this” is a sentence a cautious model will never write, and it is exactly what the reader came for.
A real author — an author page, a biography, an identity. Without it there is no expertise signal at all.

The 20% rule

If you cannot add at least 20% content that a model could not have produced, the article has no reason to exist. Publish fewer, but publish them whole.

7. Pace

A realistic plan for a new domain:

Period	Pace	Goal
Weeks 1–2	1 a day, the first 5 heavily polished	Establish the quality bar and your pillars
Weeks 3–8	1 a day, so ~30 a month	Cover the cluster, watch indexation
Months 3–6	30 a month, ~5 of them updates to old posts	Extend, and above all update what exists
After	Whatever the data says	Double down on what works, stop the rest

Thirty articles a month is the serious minimum: below that a new site has no depth and no subject, and you do not get enough data to decide at 90 days. The word that still matters is “reviewed”: if you can only review 15, publish 15.

A new domain has no authority. For the first months almost nothing ranks, at any pace — that is normal and not a failure signal. Publishing more does not shorten that wait; it just increases how much content you have to fix if the niche turns out to be wrong.

8. Maintenance

The part nobody does, which is why it pays:

Monthly — Search Console: which pages get impressions but no clicks? The title or intro is not delivering the promise. Rewrite them.
Quarterly — update the articles that get traffic. A maintained page beats a new one.
Twice a year — delete or merge anything that has had no impressions in six months. It adds nothing and dilutes the site.

Checklist

Niche you can fact-check, beatable competition, commercial intent present
Cluster read by hand, overlapping topics deleted, pillars identified
Prompt tuned: specific reader, explicit prohibitions, your angle
One test article read in full before any schedule is set
Draft by default, daily cap = real reviewing capacity
Facts verified one by one; at least 20% human content added
Real author, author page, transparency about AI use
Affiliate links marked sponsored nofollow and disclosed
Search Console checked monthly, dead content deleted

Monetise → The cost of 10 sites

How to get something out of it