Not long ago, we gave a job candidate a straightforward assignment: review market trends and social media activity, then come back with an analysis and recommendations. No AI. Two hours later, they handed in something that looked like a full day’s work from a senior practitioner complete with tidy tables, clean formatting, and confident point-by-point analysis.
We were impressed, but skeptical. Then we started asking questions. The candidate couldn’t recall the sources behind specific claims. They struggled to explain how they reached their conclusions. The output had outrun their understanding of the task entirely. The problem wasn’t that they used AI. It was that they outsourced the reading.
What they revealed, without meaning to, is the central challenge now facing everyone in communications: there is a profound difference between letting AI read something first and reading it yourself. That distinction now determines how your organization gets understood, before you’ve said a word.
We passed on that candidate. We eventually made a different hire, someone who uses AI constantly, asks questions, and is building a deep understanding of how and why it works. That hire has been one of the best decisions we’ve made. The difference between the two had nothing to do with AI. It had to do with understanding.
From Search Results to AI Summaries
Traditional search engines indexed documents and ranked them by relevance, leaving interpretation to users. Even as search became more sophisticated, this basic contract held: systems pointed to information; humans made sense of it.
But with recent advances in Answer Engine Optimization (AEO) and Generative Engine Optimization (GEO), that model has changed. The systems no longer just retrieve information and present it, they synthesize and (often) interpret it. The implication is straightforward but significant, users are no longer encountering raw materials first; they are encountering a reading of those materials.
In my own searches, I regularly see a generative summary at the top of the results window and find myself questioning whether the system understands the context of my query. A closer look at the citations often reveals that the answer is built from only one or two sources that turn out to be owned content clearly created to feed the models rather than to inform a human reader. It is a small but telling illustration of AI as the first reader: the system does not simply find the right answer, it is inadvertently biased toward the content that has been most carefully structured for it.
The First Reader Problem
Communication professionals have long recognized that information flow is shaped by intermediaries, such as journalists, editors, analysts, and policymakers who influence what information reaches the public and how it is understood. Generative AI represents a further evolution of that dynamic, and a more disruptive one.
Earlier algorithm-based systems (think Google search) filtered or ranked content and presented a list of options for the viewer to select. Today’s generative systems compose answers, burying source material. They merge multiple sources into a single response, make their best guess when sources conflict, and hand you a conclusion rather than the ingredients. This does not make AI a journalist. The absence of verification or contextualization is a critically missing component. But it does place AI in a role historically occupied by human intermediaries: reading first and shaping the initial frame through which others engage.
The practical stakes are easy to see. A prospective customer might ask an AI assistant, “Is this company a credible cybersecurity provider?” and receive a synthesized narrative built from owned content, media coverage, review platforms, and analyst commentary, all before visiting a single channel you control. Likewise, a reporter researching an organization may start with generative summaries that blend official statements with social posts, comments, and outdated material. In both cases, the AI has already read your story and formed a version of it before any human has.
The conventional sequence looked like this:
Organization → Journalist or Influencer (first reader) → Public
Now it looks more like this:
Organization → AI (first reader) → Human audiences (journalist, influencer, public)
This is the core of the First Reader concept. Describing AI as an “audience” understates its role; describing it as a “journalist” overstates it. The designation is useful because it emphasizes sequence rather than authority. Similar language has emerged in radiology and clinical imaging, where AI reads scans before human experts review them. The same logic applies here: AI does not make final judgments, but it increasingly shapes the information on which judgments are made. For the first time, the customer and the journalist share a common upstream interpreter, one that neither discipline controls.
What this Changes
When I talk with skeptical executives, I do not start with abstractions; I start with their day. AI is already in their search results, underlining phrases in email, suggesting replies in Teams and Slack, drafting performance review language, and producing meeting summaries with action items, often before they have formed their own view of what happened. It has become as unavoidable in the texture of work as email or social media once were. Which is exactly why what it does to your organization’s story matters.
The first thing AI as the first reader changes is how organizations need to think about consistency. If your website describes your company one way, your CEO’s LinkedIn another, and your latest press release a third, AI will not reconcile those differences thoughtfully. It will blend them, or default to whichever version it encountered most often. Messaging that was merely inconsistent before is now actively working against you, being synthesized into something no one intended and presented as the authoritative read.
The second is how credibility gets established. AI systems do not take your word for who you are. They infer it from what others have said about you. A well-crafted About page counts for less than a single credible third-party reference. The systems are looking for the same signals of authority that humans always have, they just automated the assessment.
Which brings us to what may be the most counterintuitive implication of all: earned media is becoming more important, not less. There is a widespread assumption that owned content and direct publishing are the future, and that a shrinking traditional media landscape makes PR less relevant. That gets it backwards.
The self-published content optimized for search is already visibly losing ground in AI results to material with some institutional backing behind it. As these systems mature and look for more reliable signals of authority, that pattern is likely to hold. Placements in credible outlets carry weight that a company blog post cannot replicate, because AI is learning to use the same proxies for authority that readers always have.
Reporters and editors are not just intermediaries to human audiences anymore. They are, increasingly, the intermediaries to the AI layer too.
The first uncomfortable truth is not that AI might become the first reader; it is that it already is. The second is that the rules keep changing as models and ranking signals evolve, so any playbook is at best a starting point.
For most organizations, the immediate task is straightforward: communications and marketing leaders need to sit down, look at their content, and ask a simple question: how does AI currently read us and is that how we want to be seen?
Same Skills, New Intermediary
There is a version of this story that treats AI as a threat to the communications profession. I do not think that is the right read.
The skills that have always mattered in PR—understanding how intermediaries shape meaning, knowing what makes a narrative credible, earning coverage rather than just publishing content—turn out to be exactly the skills that matter most when the intermediary is a machine. A placement in a credible outlet does what a company blog post cannot: it introduces a third-party voice and validation, something models will weigh accordingly.
Organizations that invest in being covered well, not just published widely, are the ones that will be read accurately.
The sequence has changed. The underlying work has not.
