Ghost's GEO tools: A promising idea in search of evidence
Making content AI-friendly is one thing. Finding evidence that it improves visibility is another.

Ghost's GEO tools: A promising idea in search of evidence

In July, Ghost introduced new tools designed to make published content easier for AI systems to discover, read, and understand. Called GEO, for Generative Engine Optimisation, the features promised to help publishers become more visible in AI-generated answers.

I enabled them on Studio6, my Ghost(Pro) publication. It took little more than flicking a switch in the settings.

Then I started wondering what difference it would make.

Ghost's announcement, Optimize your site for AI search, included an appealing claim:

Much like SEO, these new GEO tools help AI agents discover, read and understand your content more easily and efficiently, making it more likely to be referenced when responding to user queries.

That sounds worthwhile. But how would I know whether my articles were becoming more likely to be cited? And what evidence supported that expectation?

What began as a simple question led me into an interesting exploration of what GEO currently offers independent publishers, how AI systems use published content, and the difficulty of measuring whether new technology delivers its promised benefits.

Making content easier for AI

Ghost's GEO tools do two main things. They generate an llms.txt file, effectively a guide to a publication's content for AI systems, and provide simpler Markdown versions of public articles, removing much of the surrounding website design and navigation.

Both make technical sense. But AI systems can already read conventional webpages, so the question is what additional benefit these alternative formats provide.

💡
Making content easier for a machine to process might improve efficiency. Whether that translates into greater visibility, more citations or additional readers is another matter.

So I asked Ghost

I contacted Ghost's support team to ask what evidence existed for the expected benefits and how publishers could measure them.

The responses were helpful and candid. The team explained that Ghost cannot promise increased citations because AI providers decide which sources to reference. They couldn't point me towards published Ghost research demonstrating improved citation rates.

They also provided an important clarification about measurement. Ghost's analytics can identify some visits referred by AI services, but Ghost(Pro) doesn't currently provide separate reporting on AI crawler activity or citations.

In other words, I can't tell from Ghost's analytics whether AI agents are accessing my special GEO files, much less whether that leads to more citations.

The support team directed me to Ghost's media team for an official perspective. I sent questions about the supporting evidence, Markdown's advantages over HTML, measurement and future analytics. I followed up, but haven't received a response at the time of writing.

I'd welcome Ghost's contribution and would be happy to update this article.

Looking beyond Ghost for answers

With Ghost unable to provide much evidence of measurable outcomes, I turned to someone who knew the platform well.

Jon Hickman, formerly Ghost's head of support, had been particularly helpful during my move from WordPress to Ghost in early 2025. He now works independently with publishers using the platform, and I wanted his perspective.

Jon subsequently published his own thoughtful assessment, GEO in Ghost. He sees genuine efficiency advantages in the technology, but considers improved AI visibility a possibility rather than a demonstrated outcome.

His description of the features as a bet on future developments seems apt.

Jon also argues that established publishing fundamentals remain more important than speculative GEO techniques: worthwhile writing, interesting subjects and genuine connections that help others discover and share content.

I find this perspective persuasive, especially as claims about AI optimisation multiply faster than the evidence supporting them.

What the hosting data reveals

Jon also contacted two independent Ghost hosting providers, Magic Pages and Synaps Media, to investigate whether AI systems were actually using the new features.

Magic Pages examined 30 days of traffic logs across approximately 1,630 Ghost sites. It recorded around 32,000 requests for llms.txt files, but only 3.6% came from identifiable AI crawlers or assistants. Synaps Media reported a similar pattern across about 300 hosted sites, according to the providers' analysed logs.

Markdown versions told a somewhat different story. Magic Pages recorded around two million requests for these alternative article formats, with some AI crawlers making appreciable use of them. However, the figures suggest that agents fetching content in response to individual user questions generally used Markdown far less.

These observations come from hosting providers rather than a controlled study, but they add valuable context.

Evidence suggests AI systems are using some of the formats Ghost provides. What remains unclear is whether that activity translates into greater visibility or citations for publishers.

💡
We have evidence of use, but not yet evidence of benefit.

What independent research suggests

The hosting data isn't the only evidence worth considering.

In October, SISTRIX published a study examining 96,418 webpages, comparing those cited by AI services with similar pages that weren't cited.

It found no measurable citation advantage associated with llms.txt. Clear content structure, visible authorship and other familiar indicators of useful, credible material were more strongly associated with citations.

The researchers acknowledge that these are correlations, not proof of cause and effect. Their findings also don't establish whether Markdown versions make a difference.

Nevertheless, the study suggests that established publishing practices may matter more than specialised GEO techniques.

A familiar problem for communicators

The measurement challenge extends beyond Ghost or even publishing.

A publisher might want to establish several things: whether AI systems access their content, whether they cite it in answers, and whether people follow those citations to visit the original website. These activities are related, but not interchangeable. An AI system might retrieve an article without referencing it, or cite it without generating a single website visit.

Even if citations increase, that doesn't establish that a particular GEO feature caused the improvement. Other factors, including the content itself, its reputation and changes in AI search systems, could be responsible.

For professional communicators, this should sound familiar. We've long distinguished between what we produce and distribute, what audiences notice or understand, and what changes as a result. Publishing material is not the same as demonstrating that it has influenced anyone. Likewise, enabling an AI-related capability doesn't establish that it has delivered a worthwhile outcome.

This distinction matters more as organisations introduce AI tools promising greater productivity, effectiveness, and reach. The ease of implementation can be seductive, but the more important question is whether the technology achieves something valuable.

I'll leave it switched on

I don't intend to disable Ghost's GEO tools on Studio6. They're straightforward to use, the technical reasoning is credible, and I appreciate that Ghost is exploring how independent publishing might evolve as AI systems become more prominent.

But I can't currently demonstrate that enabling them has improved my work's visibility or increased its chances of being cited.

Emerging evidence suggests some AI crawlers are taking advantage of cleaner content formats, while the benefits to publishers remain uncertain.

💡
For now, I'll concentrate on what matters most to me: worthwhile ideas, thoughtful writing, credible sources and conversations with people interested in the subjects I explore.

I'll also continue watching how AI search develops and how its impact can be measured.

Ghost has introduced a promising capability. Whether it delivers the hoped-for results is still an open question.

Making content AI-friendly is becoming easier. Knowing whether doing so makes a worthwhile difference is proving much harder.


References:

(AI-generated illustration at top created with ChatGPT.)

Neville Hobson

Somerset, England
Communicator, writer, blogger from the beginning, and podcaster shortly after that.