Years ago (at the beginning of the AI revolution), I had a pretty simple problem.
I was running my own Shopify store, and I knew I should be publishing useful content regularly.
Actually doing it was another matter.
Coming up with topics, researching them, writing the article, finding images, adding product links, sorting out the SEO and then remembering to do the whole thing again the following week was exhausting.
I'm a developer, so naturally my solution was to automate it.
That eventually became autoBlogger, the automated blogging platform I've now spent years building specifically for Shopify.
It started relatively simply, but the idea behind it has always been the same: I wanted blogging to be something a Shopify merchant could set up and then largely forget about.
Today autoBlogger can do a lot more than generate an article. It can use Google Search Console data to help identify content opportunities, research and develop topics, write and edit through multiple AI passes, add product and internal links, handle imagery and metadata, score articles for SEO, schedule and publish them, and then share the finished content socially.
Getting from "AI can write an article" to "this can actually run the blogging process for me" has taught me quite a lot.
And strangely enough, generating the text has become one of the easier parts.
AI writing is the easy bit now
There are hundreds of ways to generate an article with AI.
Ask a decent model:
Write me a 1,500-word SEO article about how to choose hiking boots.
And you'll get an article.
A few years ago that was pretty amazing. Now it's basically the starting point.
The harder questions are everything around it.
What should this particular store actually write about?
Has it already covered the subject?
Is there any evidence people are searching for it?
What products are relevant?
What other articles on the store should it link to?
Does the finished article actually answer the search intent?
Has the AI repeated itself?
Does it contain loads of generic filler?
What images should go with it?
When should it publish?
And what happens after it's published?
If the merchant still needs to manage all of that, you've automated writing an article. You haven't really automated blogging.
That distinction has become increasingly important to how I've built autoBlogger.
Generation and automation aren't the same thing
This is probably one of my biggest takeaways from building with AI.
Generating something is relatively easy.
Building a reliable system that can keep doing the job without someone constantly babysitting it is much harder.
The goal with autoBlogger has always been that a Shopify merchant should be able to tell it about their store, products, audience and goals, configure how they want it to work, and then let it get on with the job.
At a very simplified level, the workflow has gradually become something like:
Store context → content opportunity → research → writing → editing → SEO review → product/internal linking → imagery → scheduling → publishing → social sharing
And all those stages affect each other.
If you pick a rubbish topic, beautifully written content doesn't suddenly make it worthwhile.
If an article is good but completely disconnected from the merchant's products and existing content, you've missed part of the point of blogging for an ecommerce store.
If you can generate 100 articles but nobody is checking their quality, you've just found a faster way of publishing bad content.
For me, the interesting part of building autoBlogger has increasingly become the system around the AI, rather than simply calling the AI itself.
One AI pass wasn't enough
One of the bigger changes I've made over time was moving away from treating article creation as one big AI request.
The simplest version is:
Prompt → AI → finished article
And to be fair, it works.
But I wasn't happy enough with the consistency.
Models repeat themselves. They sometimes settle for obvious explanations. They can produce an article that looks perfectly respectable at first glance but doesn't really say very much.
So autoBlogger has increasingly moved towards multi-pass AI writing and editing.
Instead of asking one generation to research, plan, write, edit and optimise everything at once, different stages can concentrate on different jobs.
One can think about what the article should actually cover.
Another can concentrate on writing it properly.
Another can go back over the result, look for weaknesses and improve it.
There are other checks and pieces around that process too, but the basic lesson for me has been pretty simple: don't ask one AI response to make every important decision.
It takes longer.
It costs me more per article.
But because autoBlogger is publishing content automatically, I'd much rather spend a bit more on generation and be happier with what's going out.
Automation should make quality control more important, not less.
Search Console changed how I think about topic ideas
Topic selection is another area where my thinking has changed quite a bit.
You can ask AI:
Give me 20 blog ideas for a Shopify store selling garden furniture.
You'll get 20 ideas in seconds.
Some might even be good.
But the AI doesn't necessarily know what's actually happening with that store in Google.
That's why I've added Google Search Console-driven recommendations to autoBlogger.
If a Shopify store is already appearing in Google for particular searches, that's useful information.
Maybe Google is showing the store for a query but the merchant doesn't have a particularly good page answering it.
Maybe an existing article is starting to get impressions around a cluster of related searches.
Maybe there are subjects where the store already has a little bit of visibility that could be worth building on.
None of that means "write this article and you'll rank #1".
I don't think anyone can genuinely guarantee that.
There are far too many variables in SEO.
But Search Console gives us real information from that particular store rather than relying entirely on generic keyword suggestions.
It's another useful piece of the puzzle.
And I'd much rather build automation around real store signals where they're available than just generate endless lists of keywords because an AI model thinks they sound relevant.
Shopify content needs to understand that it's part of a store
This sounds obvious, but I think it's quite an important difference between general AI writing and blogging specifically for ecommerce.
A Shopify blog doesn't exist on its own.
It's attached to a store.
There are products.
Collections.
Existing articles.
Different audiences.
Different priorities.
If you're writing an article about caring for merino wool for a merchant who sells merino clothing, the products shouldn't be completely disconnected from the content.
Likewise, if the merchant already has a genuinely useful article that expands on something being discussed, there may be a sensible internal link to make.
That's why product linking and internal linking have become important parts of autoBlogger over the years.
I don't just want it producing another URL every few days.
The aim is to gradually build a useful, connected body of content around the Shopify store.
SEO isn't something anyone can guarantee
This is another area where I've become fairly cautious about some of the claims made around AI content.
Can useful content help with organic search?
Of course.
Can I promise a merchant that installing autoBlogger will increase their organic revenue by a particular percentage?
No.
And I don't think anyone serious about SEO can make that guarantee.
There are just too many variables: the age and authority of the site, competition, backlinks, technical SEO, products, demand, existing content, site quality and plenty more.
I tend to think of autoBlogger as helping provide some important pieces of the SEO puzzle.
Publishing useful content consistently is one.
Building internal links is another.
Connecting informational content to relevant products matters.
Finding opportunities in Search Console can help.
Structuring articles properly matters.
Keeping a blog alive instead of publishing three articles and abandoning it for 18 months matters.
None of those things guarantees rankings.
They're simply useful things to be doing.
That's also why I've added SEO scoring into autoBlogger's workflow. The system can review an article before publication and look at whether it's doing the things we'd expect from useful, search-focused content.
Again, a high score isn't a promise that Google will rank something.
It's another quality-control step.
"Set and forget" is surprisingly hard to build
autoBlogger has always been intended as a set-and-forget system.
That sounds simple from the merchant's side.
It isn't particularly simple from mine.
Manual software can stop and ask the user what they want whenever something unexpected happens.
Automation needs to make sensible decisions by itself.
What happens when there isn't an appropriate product to link?
What happens when the merchant changes their schedule?
How do you avoid repeating similar topics after months or years of publishing?
How much control do you give someone before "set and forget" turns into another complicated dashboard they have to manage?
How do you introduce a new feature without suddenly changing the style of content an existing merchant has been happily publishing?
There are loads of little questions like these.
I've probably spent more time thinking about them over the years than I have about the actual "send a prompt to an AI model" part.
That's also why I think there's a fairly big difference between an AI blogging demo and an established automated blogging product.
Getting something working once is one problem.
Getting it to keep working sensibly for real Shopify merchants is another.
Years of direct merchant feedback have probably shaped autoBlogger more than anything else
One of the benefits of being a solo developer is that there isn't much distance between the people using autoBlogger and the person actually building it.
There is no support department passing feedback to a product team which eventually passes something to engineering.
If you email support, you're talking to me.
Over the years, a huge amount of autoBlogger has been shaped that way.
A merchant asks for more control over something.
Someone finds an edge case I hadn't considered.
Another person has a particular workflow that exposes something the app could handle better.
Sometimes several seemingly unrelated support messages make me realise they're all pointing towards the same missing feature.
And quite often, that ends up becoming an update.
Features have been added, prompts have changed, workflows have been rebuilt and little annoyances have been removed because of conversations I've had directly with merchants actually using autoBlogger on their Shopify stores.
I think this matters particularly with AI products.
It's possible to build an impressive AI demo remarkably quickly now.
What you can't build quickly is several years of watching real people use the product.
You don't yet know all the weird edge cases.
You haven't had years of merchants telling you what works and what doesn't.
You haven't seen how something behaves after a store has been using it week after week.
That experience has become a pretty important part of autoBlogger.
I'm still updating it every week
autoBlogger isn't an old app that I built and occasionally check in on.
I'm still actively working on it all the time.
Generally, I'm releasing updates at least once a week.
Sometimes that's a fairly substantial new feature.
Sometimes it's an improvement to article generation or one of the automation stages.
Sometimes it's something tiny that came from a support conversation with one merchant and made me think, "Yep, that should work better."
The app today is very different from the original version I built, and I'm sure it'll look different again a year from now.
That's partly because AI itself is moving ridiculously quickly, but it's also because the merchants using autoBlogger keep teaching me things about the problem I'm trying to solve.
Still completely solo and self-funded
I'm still the only developer behind autoBlogger.
It's completely self-funded.
So I'm the founder, developer, support person, product manager, marketer and the person who gets to investigate anything that breaks at an inconvenient hour.
There are definitely disadvantages to doing everything yourself.
But I genuinely think the closeness to merchants has made autoBlogger better.
I know why particular features exist because I was involved in the conversation that led to them.
I know the compromises behind different parts of the system.
And if several merchants start asking for the same thing, I don't need three meetings and a quarterly roadmap review before I can do something about it.
autoBlogger has now also been selected as a Shopify Staff Pick twice.
That recognition meant a lot to me, particularly as a solo, self-funded developer.
But I'm probably equally proud of the fact that the product has been around long enough to accumulate genuine merchant feedback and reviews and to have evolved substantially because of them.
The current AI boom makes maturity more important, not less
There are an extraordinary number of AI products appearing right now.
And that's exciting.
I'm obviously building with the same technology, so I'm not going to complain about developers experimenting with it.
But I do think there's an important difference between proving that an LLM can perform a task and building a product that can reliably perform that task for merchants month after month.
autoBlogger is the original automated blogging platform I've spent years building specifically for Shopify.
It wasn't created as a quick wrapper around whichever AI model happened to be popular this month.
It has been built, used, supported and continually improved over years of real Shopify merchant use.
That history matters.
The underlying models will change.
They'll get cheaper.
They'll get faster.
They'll become better writers.
I'll undoubtedly replace parts of the AI stack again as better options become available.
So I don't think the lasting value of an AI product can simply be:
We call a clever model.
The value has to be everything you've built around that model.
For autoBlogger, that's the Shopify-specific workflow, merchant context, Search Console signals, research, multi-pass writing and editing, quality controls, SEO scoring, product and internal linking, imagery, scheduling, publishing, social sharing and the years of merchant feedback that have shaped how all those pieces work together.
That's much harder to recreate than an API call.
If I were starting again, I'd start with the workflow
If I were building autoBlogger from scratch today, I don't think I'd start with the article generator.
I'd start by mapping the whole job.
Everything that happens between a Shopify merchant thinking:
"We really should blog more."
and eventually reaching:
"A useful article has been researched, written, checked, connected to our store, published and distributed, and I didn't have to spend my afternoon doing it."
Then I'd look at each decision in between and ask whether software could make it reliably.
AI would obviously be a big part of the answer.
But AI wouldn't be the product.
The system around it would.
That's probably the biggest thing several years of building autoBlogger has taught me.
For transparency, I'm Ollie, the founder and solo developer of autoBlogger, the original automated blogging platform built specifically for Shopify.
I've been developing it for years, it's been selected as a Shopify Staff Pick twice, and I'm still shipping improvements at least weekly, many of them directly influenced by feedback from the Shopify merchants using it.
If you want to see what it has evolved into, you can find autoBlogger on the Shopify App Store.
And if you're building an AI product yourself, I'd be interested in your experience too: at what point did you realise the difficult bit wasn't the AI generation, but everything you had to build around it?
Top comments (0)