An article a day, designed like your store.
- It plans before it writes — an outline is composed first, from the sections your theme can actually render. A post can’t be planned into a shape that won’t compile.
- Real search data, not guesses — target terms come from your own keyword profile with real volumes attached. The model selects from them; it never invents one.
- Your products, in the sections that suit them — chosen per section and pinned by handle, so the spotlight and the gallery aren’t showing the same three things.
- Your voice, learned once — tone, formality, sentence length, humour, emoji. Taken from your store at onboarding, not from a dropdown.
- Ten named checks before it’s yours — specific, unglamorous, and each one’s result is stored with the article. They’re listed further down this page.
Six stages. You stand at the fourth.
Most AI writing tools start typing immediately. This one decides what the article is, what it targets and what it will contain — and shows you all three — before it writes a sentence you’d have to read to reject.
- 01
Read the brief
One sentence becomes a product cast, a collection, a season, an angle and a set of targets.
- 02
Claim a target
From your own keyword library — minus every term another page of yours already owns.
- 03
Compose the outline
An ordered section plan, chosen from the menu of sections that actually compile.
- 04
You approve
Reorder it, cut a section, change the products. Then the quote is fixed.
Merchant gate - 05
Write it
Section by section, in your store’s voice, with each section’s own keywords.
- 06
Check and publish
Ten checks, then the article registers its keyword in your ownership ledger.
Automatic check
- the model selects from what was retrieved — products from your catalog, terms from your keyword library, collections from your own list
It can only plan what your theme can build.
There are twenty-six kinds of section an article can contain — a hero, an introduction, a product spotlight, a gallery, a comparison table, an FAQ, a timeline, a quote, a conclusion, and so on. Older versions of this engine used a fixed recipe per article type, which meant every “how to” post came out the same shape regardless of the topic.
Now the planner composes from a menu instead — and the menu is defined by what the renderer can genuinely compile. That constraint is load-bearing rather than tidy: a section type the renderer doesn’t recognise is silently dropped at build time, and an article that loses all its sections is one Shopify refuses to publish at all. So the plan is checked against the renderer before it becomes an article, not after.
- Structure follows the topic — a comparison piece and a styling guide are not the same shape
- Nothing plans what can’t render — the outline is filtered against the compilable set
- You edit it before it’s written — the outline is the cheap moment to change your mind
- Hero
- Introduction
- Rich text
- Heading
- Product spotlight
- Product gallery
- Product grid
- Comparison
- FAQ
- Table of contents
- Timeline
- Statistics
- Quote
- Testimonial
- Image
- Image gallery
- Video
- Video gallery
- Quiz
- Newsletter
- Social links
- Author bio
- Related posts
- CTA
- Divider
- Conclusion
Twenty-six types. Four of them can carry your products — the product spotlight, the product gallery, the product grid and the comparison — and the planner decides which of those this article needs, and how many.
Every section gets its own job.
An article doesn’t target one phrase — it targets a cluster, and different parts of the page are the right place for different parts of that cluster. So the plan assigns terms per section before the writing starts. This is a real plan from a real published article.
Black Sandals 2026: Complete Dress-to-Dinner Styling
published on shopbezza.com · 13 sections · commercial intent
| Term | Monthly searches | Where it lands |
|---|---|---|
| black sandalsthe primary target | 40,500 | throughout |
| platform sandalsfocus term | 74,000 | FAQ |
| comfortable sandalsfocus term | 74,000 | spotlight |
| strappy sandalsfocus term | 14,800 | rich text |
| black wedge sandalsfocus term | 12,100 | introduction |
| how to style black high heel sandalslong tail | low volume | FAQ |
Volumes are real measured search data from the store’s own keyword profile — the writer picks from these, it does not propose terms of its own.
Ten things it is not allowed to do.
Generic AI writing fails in specific, repeatable ways, and vague quality scores don’t catch them. So the gate is a set of named checks with a stated severity, run on every article after it’s written. These are the real ones, with their real names.
no_off_class_framing error
Doesn’t describe a $40 product as if it were a $400 one.
no_cross_brand_accessory error
Doesn’t style your product with a competitor’s.
no_rental_in_comparison error
Doesn’t compare something you sell to something you rent.
lede_names_focal_first error
The opening paragraph names the thing the article is actually about.
no_internal_tag_leakage error
Your internal category codes never appear in prose.
no_banned_platitudes warning
No “in today’s fast-paced world”. No “elevate your look”.
comparison_table_density warning
A comparison table has to actually compare something.
technique_closure warning
A how-to that starts a technique has to finish it.
accessory_price_coherence warning
The accessories it suggests sit in a plausible price range.
language_consistency error
The article stays in one language throughout.
the result is stored with the article — which check, which severity, and what it found
A new article never takes a term you already rank for.
The quiet way a content programme goes wrong is that post eleven starts competing with post four. Both rank a little; neither ranks well; the store is now bidding against itself with its own writing.
So terms are claimed in an ownership ledger. A new article picks from what is unclaimed, and when it publishes it registers its own term — but the claim is deliberately soft: if a product page, a collection or an older article already owns that term, the existing owner keeps it. A blog post can never displace one.
| At publish | What happens | Why |
|---|---|---|
| Term is unclaimednothing of yours ranks for it | Article claims it | new ground |
| A product page owns ityour own commercial page | Owner keeps it | never taken |
| A collection owns ita category page you published | Owner keeps it | never taken |
| An older article owns itsomething you wrote before | Owner keeps it | never taken |
Priced by the piece. Quoted first.
An article is the sum of its parts, and the outline you approve is what the quote is calculated from — not the one the planner proposed.
An article
35 creditsBrief, target, outline, writing and the checksEach extra section
7 creditsWhen you add one at the outline stageA topic idea
3 creditsProposed against your keyword libraryThe cover image
10 creditsGenerated for the article, optional
The honest answers.
Will this rank?
We won’t promise that, and nobody honestly can. What the engine does is aim each article at a term with real measured demand that nothing else of yours already owns, and then measure what happens — that measurement is a separate feature, and it’s allowed to tell you the answer was no.
Does it sound like AI?
It’s written against your store’s own voice profile, and the check list exists mostly to catch the tells — the platitudes, the wrong register, the opening paragraph that takes four sentences to say what the article is about. Read one before you publish it; that’s what the approval step is for.
Can I edit the outline?
That’s the point of showing it. Reorder sections, remove them, swap the products. The price is quoted from the outline you approve, not the one it proposed.
Where do the products come from?
Your catalog, discovered for the topic and pinned to sections by handle. The article links to products you actually sell, at the URLs they actually live at.
What if I already have a blog?
Then the more useful feature may be the other one: refreshing what’s already published and declining, rather than adding to it. They’re designed to be used together.
Publishing weekly is easy.
Publishing weekly, on purpose, is not.
first article in your free trial · approve the outline before anything is written