Why Andrea’s article is worth paying attention to
Andrea Volpini’s article, “GPT-5.4 Doesn’t Just Retrieve Your Content Anymore. It Reasons Through it,” is worth reading because it pushes the discussion beyond the usual SEO toddler toys: mentions, links, citations, and screenshots of prompts presented as if they amount to strategy. Andrea has been doing serious work in semantic search for years, so when he points to a shift in how AI systems appear to assemble answers, I pay attention.
The shift: from retrieval to synthesis
His core argument is that newer models seem less dependent on obvious retrieval patterns and more able to produce answers through relational synthesis. In practical terms, that means fewer visible citations, fewer direct brand mentions, and more answers that connect concepts without showing all the plumbing underneath.
For SEOs and CMOs, that matters because it changes the job. The question is no longer just whether your content was retrieved. It is whether your brand survives the synthesis.
Why ghost citations matter
This is where Andrea’s use of John Lovett’s ghost citations research becomes so useful. As Andrea summarises it, across 129,787 LLM responses where a brand’s content was cited, 7% were “ghost citations”: the model used the brand’s content as a source, then recommended a competitor instead. Some brands saw this in under 1% of cited responses, others in more than 40%.
That is not a quirky reporting issue. It is a warning that influence and attribution are already drifting apart.
The bigger problem: your content shapes the answer, but your brand disappears
Andrea goes a step further, and this is the part most people should be paying attention to. If citations and mentions are declining as models become more synthetic, then ghost citations may be only the visible version of a bigger problem. Your content can shape the answer, your thinking can inform the recommendation, and your brand can still disappear from the final output.
In other words, you do the intellectual labour and someone else gets the commercial benefit. That is the bit both SEO teams and CMOs should find mildly alarming.
What this means for SEO
For SEO, the implication is obvious enough: citation tracking on its own is not good enough anymore. If you are only measuring whether you were named, you may miss whether you were merely mined.
That means the work has to move beyond chasing mentions, if you really want to know how you (or your clients) are represented in LLMs. Brands need stronger entity clarity, better internal consistency, and commercial pages that make it painfully easy for machines to understand who you are, what you do, how your offer differs, and where you fit in the market.
What this means for CMOs and the wider business
For CMOs, this is larger than a visibility issue. It is a brand representation issue. If AI systems increasingly absorb category knowledge without clearly attributing it, then your real task is to make your business easy to understand, easy to associate, and difficult to misattribute.
That requires better structure, better messaging, and less reliance on generic content sludge that says a great deal while meaning very little. The risk is no longer simply being outranked. It is being conceptually useful while commercially invisible.
Where I’d still be cautious
Where I would still be a little more cautious than Andrea is around the word “reasoning.” I understand the point he is making, and I think the shift he describes is real enough, but the SEO industry does have a talent for distilling a sound observation into a slogan strong enough to intoxicate itself.
More synthesis, fewer citations, and better conceptual linking do not automatically prove that the model is “reasoning through” your content in the grand sense people will hear in that phrase. What they do show is that answer construction is becoming less transparent, and that is strategically important on its own.
The point Andrea gets right
That caveat aside, Andrea is asking the right question. The real issue is not whether AI cited you. It is whether AI understands you well enough to use your material without losing, diluting, or handing off your brand value in the process.
Once you see it that way, ghost citations stop being an oddity and start looking like an early warning.