Nearly a week into the rollout, however, we're starting to get a clearer picture. Independent tracking across thousands of websites is beginning to identify recurring characteristics among some of the sites losing visibility. Scaled and repetitive content is appearing in those observations, along with programmatic pages, thin content and sites where the underlying entity may not be particularly well established.

We need to be careful with that evidence. Google has not confirmed these as the specific targets of the September update, the rollout is not yet complete, and a site losing rankings does not tell us which part of Google's systems caused the loss. But we're no longer looking at the first few hours of an update either. Enough information is beginning to emerge to start asking better questions.

Ecommerce sites are appearing among those reporting losses too — and the timing could hardly be worse.

For Ecommerce, the Timing Could Hardly Be Worse

We're entering Q4. Black Friday, Cyber Monday and Christmas are approaching, and for many online retailers the next few months represent a disproportionate share of annual revenue. A significant loss of organic visibility in March is painful; the same loss heading into the busiest trading period of the year can become a serious commercial problem very quickly.

There is no evidence at this stage that Google has specifically targeted ecommerce, and we shouldn't pretend otherwise. But if you're running an online store that has just lost a substantial amount of search visibility, that distinction may feel fairly academic. The immediate problem is working out what disappeared, what the affected pages have in common and, most importantly, whether there is something genuinely wrong that can be improved.

This is where the emerging evidence becomes interesting. If affected areas contain large volumes of repetitive informational content, thin programmatic pages, poorly differentiated information or unclear entities, responding by commissioning another fifty Christmas articles is unlikely to address the underlying problem.

And neither is installing an AI Skill and asking Claude to fix your GEO.

An AI Skill Won't Magically Fix This Either

There has been a rapid growth in AI Skills, agents and prompts designed to audit websites for SEO, AEO and GEO. Some are useful. They can inspect pages, identify missing metadata, analyse structured data, check crawlability, suggest content changes and make technical audits considerably faster.

There are even Claude Skills specifically designed to optimise websites for AI citations and generative search. Some of the better ones sensibly acknowledge their own limitations: strong underlying SEO still matters, llms.txt isn't a magic bullet, and an automated audit can only work with the information available to it.

That last point is important.

If your problem is a missing meta description, an AI agent can identify it. If your schema is malformed, software can detect it. If important pages are difficult to crawl, an automated audit can probably tell you.

But if your website contains 1,500 pages saying substantially the same things as everybody else, another AI cannot manufacture the missing expertise that makes those pages worth having.

Nor can a Claude Skill automatically know what your sales team has learned from ten years of customer conversations, why customers keep returning a particular product, what your technical staff know that isn't in the manual, or which apparently minor product attribute turns out to matter enormously in real-world use.

The problem isn't that the AI isn't clever enough.

The information isn't there.

That's the distinction between using AI to optimise what has already been expressed and discovering what the organisation actually knows.

The Content Problem Isn't Necessarily Content

For years, one of the standard routes to increasing organic visibility has been straightforward: publish more. For ecommerce businesses that has often meant building a substantial informational layer around the store — buying guides, how-to articles, comparisons, FAQs and articles targeting questions somewhere above the eventual purchase.

There is nothing inherently wrong with that. Useful content remains useful. The problem begins when producing content becomes a substitute for contributing information.

Google's existing spam policies explicitly address scaled content abuse: generating many pages primarily to manipulate rankings rather than help users. Importantly, Google's policy is not simply about AI-generated material. It can apply regardless of whether content was produced by AI, humans or a combination of the two.

A business can therefore have 1,500 articles and still expose surprisingly little of what the organisation actually knows. The important question isn't simply how much content exists, but what useful information it adds.

Ecommerce Businesses Know More Than Their Websites Do

Think about what an established ecommerce business learns simply by operating. It knows which products customers return and why, which questions repeatedly appear before somebody buys, which specifications customers misunderstand and which products work well together. Its support staff encounter problems that manufacturers may never document, while buyers and product specialists accumulate knowledge through years of experience.

Remarkably little of that knowledge necessarily reaches the website.

Instead, the digital presence can contain hundreds of articles answering broadly the same questions that hundreds of other websites already answer. That's a very different problem from simply needing "better content".

Producing more content doesn't create more expertise. Making better use of your expertise can create more useful content.

That's one of the ideas Citewell was built around.

AI Search Makes That Difference More Important

There is another change happening alongside Google's spam systems. Informational content no longer guarantees the click it once did because AI-assisted search increasingly provides an answer before the traditional journey through several search results takes place.

Research from Pew found that Google users clicked a traditional search result on 8% of visits where an AI summary appeared, compared with 15% when one didn't. That doesn't mean informational content has stopped mattering. It means the commercial value of information is beginning to extend beyond attracting the traditional click.

A business therefore needs to ask more than, "How do we create another page that ranks?" It also needs to ask, "What useful information do we possess that deserves to be understood, retrieved and potentially cited?"

Those are different questions.

Shopify Isn't Suddenly Broken

It would also be a mistake to conclude that Shopify itself is the problem. Large themes, multiple apps, unnecessary JavaScript and complicated templates can certainly make ecommerce sites heavier than they need to be, but Google can render JavaScript and we haven't seen evidence establishing that September's update is specifically penalising Shopify architecture.

The more useful examination is underneath the storefront. Can a search engine clearly determine what a product is, what it costs, whether it is available and which attributes matter? Is that information consistent with the structured information describing it? Does the business provide useful evidence for its claims? And is valuable knowledge about the product missing from the digital presence entirely?

Those are much more interesting questions than whether somebody has installed the correct Shopify SEO plugin.

This Is Where Citewell Looks at the Problem Differently

Citewell starts with the digital knowledge the organisation already exposes and asks what is clear, what is missing, what is repetitive and what the business knows that hasn't made it online.

Then we can look outside the organisation.

Gap-Cite examines the differences between the information you make discoverable and the information competitors expose. A competitor may explain an attribute you don't, answer a buying question you've ignored, provide evidence you haven't, or simply make information visible that remains buried inside your organisation.

Sometimes that identifies something the organisation genuinely doesn't know. But sometimes the result is more frustrating: your competitor has published useful information that your own people already know. You simply haven't published it.

That's not necessarily a content-generation problem. It's a knowledge-discovery problem.

And Then There's Take-Cite

AI-assisted discovery introduces another opportunity because the sources currently being cited aren't necessarily the best sources that could exist.

A citation isn't sacred. If an answer engine is relying on a weak source, that source has earned visibility, not permanent ownership of it.

Take-Cite looks for those opportunities and asks: if this is the source being cited today, can we create the source that deserves to be cited tomorrow?

If the existing source provides a superficial answer and your organisation genuinely knows more, can provide stronger evidence or can take the subject further, the objective isn't to rewrite what's already there. It's to build the better source.

That isn't manufactured authority. The objective is to earn the citation.

And This Is Where the Citation Line Matters

There is a hard limit to what any SEO tool, AI agent or Claude Skill can discover from your existing digital presence.

Eventually it reaches the end of the documented evidence.

We call that the Citation Line: the point where accessible documented evidence ends and organisational knowledge that has never been expressed begins.

On the other side of it might be the product specialist who knows why one component repeatedly fails, the person dealing with returns who knows exactly what customers misunderstand, or the buyer who has spent fifteen years learning which specifications actually matter.

An AI cannot discover information that has never been expressed. It can reorganise existing knowledge, analyse it, compare it and help communicate it, but eventually a human has to provide the knowledge that isn't there.

That is why simply adding another AI optimisation layer cannot solve every visibility problem.

Sometimes the opportunity isn't another keyword, another schema block or another prompt.

It's knowledge sitting beyond the current Citation Line.

What September Is Starting to Tell Us

Google still hasn't told us exactly what its September systems are detecting differently, so we aren't going to pretend we've reverse-engineered the update. But independent observations are now beginning to show patterns involving scaled and repetitive content, programmatic scaling, thin pages and poorly established entities.

Those observations broadly overlap with problems Google has been warning about for some time. What makes September particularly interesting is that we're now seeing several of those issues appearing together among sites experiencing substantial movement.

For ecommerce businesses, the timing makes understanding those patterns considerably more urgent. A serious visibility problem entering October isn't merely an interesting SEO problem when Black Friday, Cyber Monday and Christmas are approaching.

This is also why we're watching the update closely at Citewell. Not because we want to produce a "September Update Optimiser", but because the areas beginning to surface are closely related to questions we've already been asking: what useful knowledge does an organisation actually expose, how well is the organisation itself understood, where is information repetitive rather than useful, what are competitors explaining better, and what valuable knowledge remains unpublished?

Don't Fix the Algorithm. Fix the Underlying Problem.

Citewell hasn't changed its underlying strategy because Google released another update.

Search engines will continue to change. AI systems will continue to change. There will be another Google update after this one, and another fashionable optimisation tool promising to deal with it.

Trying to predict every algorithm change is an exhausting way to build a digital strategy.

Building a clearer, better evidenced and more useful representation of what your organisation genuinely knows is a rather more durable one.

The September update is still rolling out, and we're continuing to examine what's being affected as more evidence becomes available.

We don't know everything yet.

But the picture is becoming considerably clearer.