The Agenda-Setters: What McCombs and Shaw's 1972 Agenda-Setting Theory Predicts About AI Answer Engines

Agenda-setting theory, in one sentence

In 1972 Maxwell McCombs and Donald Shaw published a study of the Chapel Hill electorate that became one of the most replicated findings in communication science. Their claim was deceptively simple: the media are strikingly successful at telling people what to think about, even when they fail to tell people what to think. The press does not have to win an argument to shape an outcome. It only has to decide which subjects appear and which do not.

The theory calls this salience transfer. Whatever the media place high on their agenda tends to rise on the public's agenda too. The engine of that transfer is selection, and selection is a form of gatekeeping: someone, or something, decides what gets exposure before the audience ever sees it. For fifty years the gatekeeper was a newsroom editor. The point of this article is that the gatekeeper is now, increasingly, a model that writes an answer.

From newsroom to answer box: the same mechanism, a new gatekeeper

A search engine has always been an agenda-setter. It decides which of the web's pages an audience is shown, and in what order, and that ordering shapes behavior without persuading anyone. What has changed is not that search became a gatekeeper. It is that the format of the gate changed, from a ranked list you could still browse to a single synthesized answer that names a few sources and hides the rest.

This is a recognizable progression, not a rupture. The human-curated directory of the mid-1990s gave way to link-based ranking when Brin and Page published PageRank in 1998, which reframed relevance as a vote of confidence cast by links rather than editorial placement. Each generation was a new gatekeeper with new admission rules. The generative-answer layer is simply the newest one, and it filters more aggressively than any before it, because it returns a verdict instead of a menu.

The academic bridge from editors to algorithms was built before the AI-search moment arrived. Karine Barzilai-Nahon's network gatekeeping theory formalized, in 2008, the power to include, exclude, and order the information others depend on, and it has since been extended to describe search engines as gatekeepers of web content access. Agenda-setting supplies the why it matters; network gatekeeping supplies the how it operates. Both point at the same lever.

The three levels of agenda-setting map onto the AI answer

Agenda-setting research did not stop in 1972. Over five decades it developed three documented levels, and each one has a direct analogue inside a generated answer.

The consideration set is now assembled by the engine

Marketing has a long-standing name for the small group of options a buyer actually weighs before choosing: the consideration set. Classic theory assumed the buyer built it, by recalling brands and scanning what was in front of them. The structural change of the answer economy is that the engine now assembles the consideration set on the buyer's behalf and hands it over pre-built.

This is where agenda-setting stops being a media-studies abstraction and becomes a revenue question. A business can rank on the first page of classic results and still be absent from the three names the AI answer places above those results. Absence from the answer is not a weak position in the buyer's consideration set. It is exclusion from the set before the buyer even begins to deliberate, which is precisely the outcome McCombs and Shaw predicted for anything the gatekeeper leaves off the agenda.

Meta-gatekeepers: gatekeeping the gatekeepers

The most recent extension of this line of theory is worth stating carefully, because it is genuinely new and not yet widely replicated. A 2026 peer-reviewed, cross-language study by Kuai and colleagues frames generative AI chatbots as meta-gatekeepers: a chatbot's answer decides which of the web's already-gatekept sources, the news outlets, government sites, and review platforms that themselves filter information, even get surfaced to the user. The engine adds a new gatekeeping layer on top of the old ones.

The study examined political-information retrieval across five languages, so its direct evidence is not about commercial or local queries. The mechanism it describes, however, generalizes: whether the question is "who should I vote for" or "who is the best plumber in my city," a single synthesized answer performs a selection over the sources that would previously have competed for attention on a results page. That is gatekeeping squared, and it raises the stakes of being cited inside the answer rather than merely indexed behind it.

What the click data shows about where salience lands

Theory predicts that attention follows the gatekeeper's selection. The strongest primary data in this domain is consistent with that prediction. In July 2025 the Pew Research Center tracked the actual browsing of 900 consenting US adults across 68,879 Google searches and found that when an AI summary was present, users clicked through to a traditional result in 8 percent of searches, versus 15 percent for searches without one, and clicked a link cited inside the summary only about 1 percent of the time.

The same study found that 88 percent of AI summaries cited three or more sources and only 1 percent cited a single source. Read through an agenda-setting lens, that is the shape of a modern agenda: a short, curated set of named entities that captures the attention which used to be spread across a full page of links. The buyer's attention is landing on the answer, and on the handful of businesses the answer chose to name.

Answer engine optimization: entering the agenda deliberately

If the answer is the agenda, the practical question is whether a business can influence its own inclusion. A peer-reviewed literature on this already exists. The 2024 paper that introduced Generative Engine Optimization, authored by researchers from IIT Delhi, Princeton, and Georgia Tech and published at KDD, built a benchmark and tested content interventions for their effect on whether a source is cited inside a generated answer, reporting that levers such as adding cited statistics, quotations, and authoritative sources raised a source's visibility in the systems it studied.

That such a discipline exists at all confirms the shift the theory describes. The object being optimized is no longer a rank in a list, it is a citation inside an answer. Answer engine optimization, and its close cousin generative engine optimization, are the operational names for the work of getting a business onto the agenda the engine now sets. The model decides what it cites; the work is engineered visibility and corroboration that make a business the strongest candidate for that citation.

What is established, and what is still emerging

Separating what is settled from what is new matters here. Agenda-setting theory itself is about as established as findings in social science get: five decades of replication across the first, second, and third levels. Network gatekeeping theory is a mature framework. Applying them to generative answer engines is the part that is emerging. The meta-gatekeeper study is recent, single, and awaiting replication, and its direct evidence is political rather than commercial.

The wider domain also carries real hype worth naming. Gartner predicted in 2024 that traditional search volume would drop 25 percent by 2026 as chatbots absorbed queries. As of this writing that collapse has not materialized as stated, and Google still holds the large majority of the search market. The theory that AI answers set the buyer's agenda is well grounded; confident predictions about how fast and how completely search will be replaced are where the field keeps overreaching. The correct response to a genuine but young shift is to measure standing across every gate, not to assert it.