Okay, I need to talk about something that's been bothering me for weeks, and I've finally figured out how to articulate it without sounding like I'm yelling at clouds. (Though I am also yelling at clouds. The clouds know what they did.)

There's a quiet contradiction sitting right at the centre of almost every AEO and GEO strategy I'm seeing right now. Businesses are being told — correctly — that they need to structure their content for AI ingestion. Clean schema markup. Authoritative entities. Dense factual signals that give language models something to retrieve and cite. The advice is sound. I've given this advice. You've probably given this advice.

And then those same businesses go and produce pages of technically optimised content that no human being would voluntarily read twice. Some of it you wouldn't read once. Some of it reads like it was written by a committee that was itself written by a committee.

Here's the thing nobody wants to say out loud: the two problems are feeding each other.

We all know what AEO and GEO systems are looking for in the retrieval step. Authoritative sources. Structured data. Recognised entities. Consistent attribution. Topic cluster coverage that demonstrates you actually know what you're talking about rather than having spent forty-five minutes with a chatbot and a vague content calendar.

What's getting dramatically less airtime is the degree to which these systems are also reading behavioural signals. The content that consistently surfaces in AI-generated answers isn't just the most structured content. In many cases, it's the content that has accumulated the strongest engagement record on the open web — the pieces that get cited, shared, linked to, and returned to by actual humans with actual attention spans.

Which means all that beautiful schema markup you've been implementing? It's building a house. But human attention has to pour the foundation first.

(I just used a construction metaphor. My partner would be so proud. They've been renovating our bathroom for six months and I've absorbed construction vocabulary through sheer proximity and resentment.)

Three forces have been quietly eroding that foundation since the generative AI content explosion kicked off. Platform algorithms have gotten better at detecting and suppressing low-quality output. Audiences have developed — without necessarily being able to name it — a gut-level sensitivity to content that feels produced rather than written. And the sheer volume of material competing for attention has raised the implicit bar for what counts as worth reading.

What researchers have started calling "slop" — competent, inoffensive, and forgettable — isn't just a user experience problem. It's increasingly an AI visibility problem. Your beautifully structured slop is still slop. The machines can tell because the humans already told them.

Here's where I had an uncomfortable realisation at about 11pm on a Tuesday (I've been told I need to stop having professional epiphanies at times that suggest I don't have a life, but here we are).

The most common point of failure in AI-assisted content production is not the model. It's the brief. A vague prompt produces generic output. Generic output, published at scale, compounds every structural problem I've just described. The instinct to treat AI as a shortcut — opening a chat window when a first draft is needed, tidying up the result, hitting publish — treats the production step as the problem when the strategy step is where the actual work needs to happen.

Think about it this way. You would not hand a new hire a one-line instruction and expect a polished, on-brand piece with a clear editorial position. (Well, some of you would, and I have questions about your management style, but let's move on.) You'd give them a defined audience. A specific pain point that audience is currently experiencing. The emotional response you're trying to generate. Explicit guidance on what the brand doesn't say as well as what it does.

The quality of AI output is a function of the quality of the brief. AI just accelerates both good and bad outcomes equally. I'd check yours. Right now. I'll wait.

And here's the step that almost everyone skips: the human evaluation checkpoint between generation and publication. Without it, you've built a very efficient loop of mediocrity — fast, scalable, and consistently below the threshold that earns attention. It's like installing a state-of-the-art irrigation system and then planting nothing. Beautiful infrastructure. Zero harvest.

Consumer trust in digital content has been falling for years. AI-generated output didn't cause the decline — it accelerated and concentrated it, like pouring accelerant on a fire that was already doing perfectly well on its own, thank you.

Audiences in 2026 have been exposed to enough AI-generated material to have developed something close to an instinctive detection capacity. They can't always name what feels wrong about a piece of content. But the brain — which is essentially a prediction machine running on coffee and spite — recognises and dismisses what it can easily anticipate. Safe content that follows best practices so faithfully it blends into every competitor's output has become invisible content.

This has direct implications for your AEO strategy. AI discovery systems are retrieving from the same open web that human audiences have already passed judgement on. A law firm that's published twelve technically perfect guides to "understanding employment law" isn't building authority — it's building a monument to competent boredom. A SaaS company with forty product comparison pages that all read like they were generated from the same template is optimising for a search landscape that no longer exists.

Impressions without engagement. Visibility without retention. Citations without trust. These are the outputs of a strategy that has confused the map for the territory. (I'm going to get that tattooed somewhere. Probably my forehead. It'd save time in client meetings.)

The answer isn't more content. It's more considered content. Pieces with a defined editorial position. Written for a specific reader experiencing a specific problem. Content that's willing to actually say something rather than describe everything from a safe distance like a nature documentary narrator who's been told not to pick favourites.

AEO and GEO now operate across text, voice, and visual surfaces. The instinctive response to that expanded surface area is to produce more stuff. More formats. More channels. More everything.

The more productive response is to think more carefully about a single asset before you produce it. Content adapted for platform-native behaviour — rather than syndicated uniformly everywhere like a press release with delusions of grandeur — serves both human audiences and AI retrieval more effectively. Recycling the same asset everywhere serves neither. Audiences have become fluent in the difference between content designed for a context and content dropped into one. It's the digital equivalent of wearing a tuxedo to a barbecue. Technically dressed. Contextually absurd.

Define the core argument first — what is this piece actually trying to say? — then ask what format serves that argument on each surface. That keeps the editorial decision at the centre rather than tacked on at the end like an afterthought with a Canva template.

The metrics that feel like success — impressions, follower counts, aggregate likes — are visibility indicators. They tell you how much content you've distributed. They do not tell you whether audiences are engaging with it in the way that builds the behavioural signals AI discovery systems are reading.

Watch time. Scroll depth. Return visits. Saves and shares. A reader who consumes 85 per cent of an article without tapping a single button is more valuable to your long-term AI visibility than a reader who hits like and bounces in two seconds. The first generates dwell signal. The second generates a number for a slide deck. (Guess which one most teams are reporting on.)

Define what engagement looks like for a specific piece before it's published, not after. Retrofitting meaning to whatever the dashboard shows isn't measurement. It's post-rationalisation with better fonts.

Because I know some of you are already composing a Slack message to your content team and need bullet points. Fine. Here:

Here's the uncomfortable truth that connects all of this: AI can execute a content strategy at scale, but it cannot own one. Ownership — of the argument, the audience relationship, the factual claims, the editorial position — requires a human. That's not a limitation of current technology. It's the condition under which content earns the trust that makes AI discovery worth optimising for in the first place.

The businesses getting this right aren't choosing between AI optimisation and human engagement. They're doing the harder thing: building content that's structurally sound enough for machines to find and distinctive enough for humans to care about.

Everyone else is building beautiful, well-structured monuments to content that nobody asked for, nobody reads, and increasingly, nobody's AI recommends either.

Which is, when you think about it, quite the achievement.

Jo Lambadjieva is an entrepreneur and AI expert in the e-commerce industry. She is the founder and CEO of Amazing Wave, an agency specializing in AI-driven solutions for e-commerce businesses. With over 13 years of experience in digital marketing, agency work, and e-commerce, Joanna has established herself as a thought leader in integrating AI technologies for business growth.

😍It was great, all I needed

🙂Good, not great

🙄It sucked