There is a convenient pair of narratives doing the rounds at the moment, and both are useful for avoiding the real work. The first says AI is about to end marketing as we know it, possibly civilisation too. The second promises AI will hand you effortless abundance if you just lean in hard enough. Search Engine Journal reports that the truthful answer sits somewhere between those extremes, in a middle ground that does not sell headlines but is where most marketing teams actually have to operate.
The parallel drawn is Y2K. Easy to dismiss in hindsight because planes stayed in the air, but that calm outcome was the result of years of unglamorous fixing, not proof the warning was empty. AI warnings work the same way when they point at a specific risk and lead to a specific safeguard. They stop working when the message is just that the sky is falling and nothing can be done. The opposite pitch, that AI will erase drudgery and deliver a golden age, comes from the same machine wearing a different costume. Reality will likely be less cinematic: real progress next to real problems, measurable wins mixed with disappointment.
Fact-checking is no longer the bar
Most SEOs already know AI hallucinates. Knowing that is not enough anymore, because fact-checking only tells you whether a claim is technically accurate. It does not tell you whether the claim holds up under scrutiny or whether the inference behind it is sound.
The publication pushes practitioners toward ground-truthing instead. Before you ship anything AI-drafted, ask whether the underlying inference is right, whether the recommendation actually fits your market and your customer, and what evidence would prove the central assumption wrong. An AI system can hand you a clean-looking draft or a fast summary of a dataset. It cannot tell you whether the direction is correct. That remains your job, and the gap between "factually true" and "directionally right" is where most AI-assisted content fails quietly.
This is not a spellcheck problem. It is a structural one. A piece can be free of factual errors and still recommend the wrong approach, target the wrong audience, or rest on an assumption that does not survive contact with your actual customer base. Ground-truthing asks you to work backwards from the conclusion: what would have to be true for this to be the right move? If you cannot answer that clearly, the draft is not ready.
Inconsistency is now a citation risk
Here is the part practitioners underrate. AI did not just change how content gets made, it raised the cost of departments that do not talk to each other. When a press release says one thing, the landing page says another, and social says a third, human readers get confused. So does every AI system trying to summarise your brand.
Inconsistency used to be a minor annoyance. Now it is a citation risk. If an AI search engine or language model cannot reconcile what you are saying across touchpoints, it is less likely to cite you cleanly, or at all. This is not a hypothetical risk. It is already visible in how models handle brands that publish contradictory messaging. The same problem that frustrates human readers frustrates the systems trying to parse your content for training or retrieval.
SEO needs to be in the room with PR from the start, not bolted on at the end. Analytics needs to be there from the beginning of a project, not summoned to bless a decision that has already been made. The integration work that used to be optional is now load-bearing, because the systems ingesting your content do not forgive the gaps your internal handoffs create.
What to actually do about it
The advice given is practical and specific. Before publishing anything AI-assisted, run it through the ground-truthing questions above rather than a spellcheck-level review. Ask what would have to be true for the conclusion to be wrong. If the answer is unclear or uncomfortable, the piece needs more work.
Pull one person each from PR, content and analytics into the next planning conversation for any piece touching brand messaging, even if it feels like overkill. The inconsistency AI systems pick up on is usually the same inconsistency your own departments created independently. Fixing it at source is cheaper than patching it later.
When your PR team pitches a story, build the evidence in from the start. A strong hook plus real data and a sample size that survives scrutiny beats a hook alone. That difference matters more now because it is the difference between a piece that gets cited by the next model update and one that gets ignored. The hook still matters, but it has to be backed by a story that can survive an editor's second look and a machine's attempt to verify it.
The attention war is making this harder
Part of why the doom-or-utopia framing dominates is structural. Fewer journalists are covering more AI claims, and the incentives, algorithmic and human, reward the most dramatic version of any story. The pressure strips out the qualifiers, the "yes, but" and the "only if," until readers are left choosing between two extremes that were never the real choice.
The fix is not to abandon the strong hook. It is to back it with a story that can survive scrutiny. That applies as much to your own content as it does to the coverage you pitch. The bar for what counts as a credible claim is rising, both for human audiences and for the AI systems that increasingly mediate access to them. The teams that adjust to that bar early will have an advantage that compounds, because the gap between "technically true" and "actually defensible" is where trust gets built or lost.
The middle ground is the work
Neither the apocalypse nor the golden age is a useful planning assumption. The real impact of AI on marketing sits in the uncomfortable middle: tools that genuinely improve some workflows while creating new risks in others, productivity gains that are real but not evenly distributed, and a shift in what counts as good enough that makes old habits expensive.
The marketers who do well in that environment will be the ones who treat AI as a tool that requires different checking, not less checking. Ground-truthing before you publish. Consistency across departments because the cost of misalignment just went up. Evidence built in from the start, not retrofitted when someone asks for it. None of that is dramatic, and none of it is optional.
This is not the first time a technology shift has changed the bar for what professional work looks like. It will not be the last. The practitioners who adapt are usually the ones who take the boring middle seriously while everyone else is still arguing about the extremes. That pattern holds here too.