The Macro Shift · emerging evidence
Disintermediation, Reintermediation: A 30-Year Pattern in How the Internet Removes and Replaces Middlemen
Disintermediation is the removal of the middleman. The internet has spent thirty years cutting out the agent, the broker, and the branch that once sat between a buyer and what they wanted. The pattern rarely stops there. Almost every time one middleman is removed, a new one forms in its place, a move economists call reintermediation. Travel is the textbook case: the corner travel agent gave way to online booking sites, which gave way to metasearch comparison engines, which is now giving way to AI trip planning. Search ran the same loop on itself, from the human-edited directory to the ranking algorithm to the synthesized answer. Read this way, AI search is not a rupture but the latest turn of a cycle the field already knows how to describe. The practical consequence is specific and current: the AI answer layer is becoming the intermediary standing between a business and its buyer, and reaching it is now the work.
Disintermediation and reintermediation, defined
Disintermediation, in its plain economic sense, is the removal of an intermediary from a transaction so that the two ends deal more directly. The word entered wide use in banking before the web, then became the defining promise of early internet commerce: cut out the layers of agents, brokers, distributors, and desks, and pass the saved margin and friction back to the buyer and seller.
The less-quoted half of the story is reintermediation, the re-emergence of an intermediary after the old one was displaced. In practice, pure buyer-to-seller directness is unstable, because the functions the old middleman performed, aggregation, trust, discovery, and matching, do not disappear when the middleman does. They migrate to whoever can perform them next, usually a new platform that looks nothing like the one it replaced. The internet did not abolish the middleman so much as keep relocating it.
This pairing matters here because it gives a disciplined, non-hyped vocabulary for what is happening in search. Instead of asking whether AI is destroying an industry, the cycle asks a sharper question: which intermediary functions are moving, and to which new layer are they moving?
A thirty-year pattern of removing internet middlemen
The clearest way to see the cycle is to watch a single category run through it. Travel is the canonical example, and it has completed several full turns in about thirty years.
First, the internet disintermediated the human travel agent: booking sites let a traveler reach airlines and hotels without a person in the middle. But directness did not last. The aggregation and comparison the agent used to perform reintermediated into online travel platforms, which became powerful middlemen in their own right. Those platforms were then partly disintermediated again by metasearch engines that compare across platforms, and by suppliers pushing book-direct. Now a further layer is forming, in which a buyer asks an AI assistant to plan the trip and the assistant reads across all of the above and returns a shortlist.
The point is not the specific brands but the shape. Each turn removed a middleman and installed a structurally different one, and each new intermediary captured the discovery step, the moment where the buyer decides who is even in the running. Retail, financial services, media distribution, and classified advertising each show their own version of the same loop. The recurrence is what makes the pattern legible rather than unprecedented.
Search ran the same loop on itself
Search is not an exception to this pattern; it is one of its cleanest instances, and its history is unusually well documented because each transition has a founding artifact.
The first web directories, beginning with Yahoo in 1994, were human-edited taxonomies: editors decided, by hand, what belonged in the consideration set. That editorial middleman was disintermediated by crawling engines, and then decisively by PageRank, published as an academic paper by Brin and Page in 1998, which reframed relevance as a link-based vote of confidence rather than an editor's placement. The human curator was removed; a ranking algorithm took over the same gatekeeping job under new admission rules.
The synthesized AI answer is the next turn of this same loop. It does not merely reorder the ten links the algorithm produced; it reads across them and returns a single composed response, performing the selection and comparison a buyer used to perform by scanning. The function that once belonged to the directory editor, and then to the ranking algorithm, is reintermediating again, this time into a generative layer that answers rather than lists.
Why every removed middleman invites a replacement
Reintermediation is not an accident or a failure of the disintermediation promise; it follows from what intermediaries are for. A middleman survives when it performs a function neither end of the transaction can cheaply perform alone. In discovery-heavy markets, that function is reducing the buyer's search cost: aggregating options, filtering the unfit, and signaling which to trust.
Those needs intensify, not ease, as supply grows. Herbert Simon named the underlying economics in 1971: a wealth of information creates a poverty of attention, and a need to allocate that attention efficiently among the overabundance of sources that might consume it. When content becomes effectively infinite, the buyer's scarce resource is attention, and whoever can spend it on the buyer's behalf, by narrowing an unmanageable field to a handful, becomes the next intermediary. Simon's companion idea, that people act on the first adequate answer rather than the optimal one, explains why a single synthesized answer can capture a disproportionate share of buyer action even when it has not been verified as best.
So the cycle is self-perpetuating. Remove the middleman, and the discovery burden it carried lands back on a buyer who cannot bear it, which creates the opening for a new middleman to form. The only durable question is which layer captures the function next.
AI as intermediary: the answer layer is the new gate
Communication research supplies the mechanism by which a new intermediary exerts power, and it long predates the AI moment. Barzilai-Nahon's 2008 theory of network gatekeeping formalizes gatekeeping as control of a gate combined with how much the gated depend on it: the power to include, exclude, and order the information others rely on. A generative answer layer is a gate in exactly this sense.
The most recent work goes further. A 2026 cross-language study frames AI chatbots as meta-gatekeepers that gatekeep the other gatekeepers: the chatbot's answer decides which of the web's already-gated sources, the news outlets, the review platforms, the directories, even surface to the user. That is reintermediation stated in its strongest form. The new intermediary does not sit beside the old gates; it sits above them and decides which of them a buyer ever sees. This study is recent and awaits replication, so it is best read as an emerging framing rather than a settled finding, but it is methodologically serious and points the same direction as the older theory.
For a business, the operational meaning is blunt. Being absent from the AI answer is not the same as ranking one position lower on a list. It is being left out by the layer that now performs the buyer's shortlisting, before the buyer ever reaches the surfaces where the business is visible.
The evidence the new intermediary already routes buyers
The claim that the answer layer is capturing the discovery step is not only theoretical; it shows up in independent behavioral data, though the record is still forming.
The click is leaving the page the old intermediary owned
Clickstream analysis by SparkToro found that most US Google searches already ended without a click to any external site, at 58.5 percent in its 2024 study, and a follow-up using a separate panel put the figure at 68.01 percent across early 2026. In the same period, AI Overviews were found on more than 20 percent of searches and, when present, cut click-through to results by close to 60 percent. The intermediary that used to hand the buyer onward to a business is increasingly answering in place.
The strongest single anchor is primary panel data. A Pew Research Center study of 68,879 real searches by 900 US adults in March 2025 found users clicked a traditional result in about 8 percent of searches that showed an AI summary, versus 15 percent without one, and clicked a link inside the summary itself only about 1 percent of the time. The new layer is retaining the attention the old layer used to pass through.
A named optimization science confirms the layer exists
When a new intermediary forms, a discipline for reaching it tends to appear soon after, the way search-engine optimization appeared once the ranking algorithm became the gate. That marker is already here. A 2024 peer-reviewed paper at a top data-mining venue introduced Generative Engine Optimization and measured which content levers, such as adding cited statistics, quotations, and authoritative sources, change whether a source is cited inside a generated answer. The object being optimized is no longer a rank in a list; it is a citation inside an answer, which is precisely what you would expect once the answer layer becomes the middleman.
Generative engine optimization is the sign a new middleman formed
It is worth stating the inference plainly, because it is the load-bearing claim of the framework. The emergence of a rigorous, peer-reviewed literature on optimizing for generative answers is itself evidence of reintermediation. Disciplines like this do not form around surfaces that lack power; they form around gates that decide outcomes. Yahoo's directory had submission strategies, PageRank produced the entire SEO industry, and the generative answer layer is now producing its own optimization science.
Reading the pattern this way reframes the practical task. The question is not whether to react to AI, but how to reach the current intermediary given that a new one has demonstrably formed and a body of method for reaching it already exists. That is a question a business can actually answer, which is the advantage of using the cycle instead of the hype.
Where the pattern holds, and where it doesn't
The cycle warns against both denial and doom. What follows names what this framing does not yet prove.
On the doom side, the most-cited collapse forecast has not held. Gartner predicted in 2024 that search-engine volume would drop 25 percent by 2026 as AI absorbed queries; as of this writing that has not materialized as stated, and Google still holds the large majority of the search market. Reintermediation is not extinction. The discovery function is moving to a new layer, but the older surfaces, classic results, the local map pack, and reputation, still carry enormous volume and still decide many buyers. Attention is being reallocated across more gates, not vanishing from search.
On the rigor side, this article makes a structural argument, not a settled economic measurement. The disintermediation and reintermediation cycle is a well-described pattern and the search-specific and behavioral evidence is real, but a fully quantified economic account of the AI answer layer as an intermediary, with a second independent primary source, is not yet established. The meta-gatekeeper framing is emerging and awaits replication. The framework earns its keep by making the shift legible and testable, not by asserting a precise magnitude. The disciplined response is to measure a business's standing across the gates directly, rather than to assume it.
The evidence
Key findings, with their sources
-
Most US Google searches ended without a click to an external site, at 58.5% in 2024, rising to 68.01% in early 2026 across a separate panel.
established SparkToro, "2024 Zero-Click Search Study" (with Datos/Semrush data) and "In 2026, Less than One Third of Google Searches Still Send a Click" (with Similarweb data), sparktoro.com.
-
Users clicked a traditional result in about 8% of searches with an AI summary present, versus 15% without, and clicked a link inside the summary only about 1% of the time.
established Pew Research Center, "Do people click on links in Google AI summaries?", 2025 (browsing panel, 900 US adults, 68,879 searches).
-
AI Overviews appeared on more than 20% of Google searches and, when present, cut click-through to results by close to 60%.
established SparkToro / Similarweb analysis, 2026, covered by Search Engine Land.
-
Adding cited statistics, quotations, and authoritative sources measurably raised whether a source was cited inside generated answers in tested engines.
established Aggarwal et al., "GEO: Generative Engine Optimization", KDD 2024, arXiv:2311.09735 (peer-reviewed).
-
PageRank reframed relevance as a link-based vote of confidence, displacing the human directory editor as the gatekeeper of the consideration set.
established Brin, S. & Page, L., "The Anatomy of a Large-Scale Hypertextual Web Search Engine", Computer Networks and ISDN Systems, 1998.
-
AI chatbots can act as meta-gatekeepers, deciding which of the web's already-gated sources a user ever sees, a new intermediary layer above the old gates.
emerging Kuai et al., "AI chatbot accountability in the age of algorithmic gatekeeping", 2026 (peer-reviewed, five-language comparative study).
-
A 2024 forecast of a 25% drop in search volume by 2026 has not materialized as stated; Google still holds the large majority of search.
contested Gartner press release, 2024; reality-check reporting, 2026.
Calibration
What is proven, what is promising, what is unproven
| Evidence tier | Tactics | What the evidence says |
|---|---|---|
| established | The disintermediation and reintermediation cycle as a described pattern; the search-specific history (directory to ranking algorithm to answer); zero-click and Pew click behavior; the named GEO optimization science. | Long-standing IS and economics literature on intermediation; Brin and Page 1998; SparkToro 2024 and 2026; Pew 2025; Aggarwal et al. 2024 (KDD). |
| emerging | Framing the AI answer layer specifically as the new reintermediary and meta-gatekeeper above existing gates. | Kuai et al. 2026, methodologically serious but single-study and awaiting replication. |
| contested | Quantified claims about how large the AI-search shift already is, and predictions of search-volume collapse. | Gartner 2024 forecast did not hold as stated; vendor AI-share figures vary widely by denominator and are not yet reconciled. |
Reference
Glossary
- Disintermediation
- The removal of an intermediary from a transaction so the two ends deal more directly, cutting out an agent, broker, or platform that used to sit between buyer and seller.
- Reintermediation
- The re-emergence of an intermediary after an earlier one was displaced. The discovery, trust, and matching functions the old middleman performed migrate to a new layer rather than disappearing.
- Reintermediary
- The new intermediary that captures a market's discovery step after the previous one is removed. In search, the generative answer layer is the current reintermediary.
- Network gatekeeping
- Control of a gate combined with how much the gated depend on it: the power to include, exclude, and order the information others rely on. A generative answer layer is a gate in this sense.
- Generative engine optimization
- The practice, and the peer-reviewed literature, concerned with whether a source is retrieved and cited inside an AI-synthesized answer, as distinct from ranking a link in a list.
Straight answers
Frequently asked questions
What is disintermediation?
Disintermediation is the removal of a middleman from a transaction so the two ends deal more directly. On the internet it has meant cutting out agents, brokers, and distributors, as when online booking removed the travel agent or PageRank removed the human directory editor.
What is reintermediation, and why does it keep happening?
Reintermediation is the re-emergence of an intermediary after the old one is displaced. It keeps happening because the functions a middleman performs, aggregating options, filtering, and signaling trust, do not vanish when the middleman does. They migrate to whoever can perform them next, usually a new kind of platform.
Is AI search a genuinely new middleman, or just hype?
The structural evidence is real: independent panel data shows the answer layer retaining clicks the older results page used to pass on, and a peer-reviewed optimization literature has already formed around being cited in AI answers. What is still contested is the magnitude. Forecasts of search-volume collapse have not materialized as stated, so the read is a new intermediary forming, not an industry ending.
How is reaching the AI answer layer different from SEO?
Classic SEO optimizes for a position in a ranked list of links. Reaching the AI answer layer means being named and cited inside a synthesized answer, which depends more on entity consistency, extractable content, and third-party corroboration than on ranking position alone. They are related but distinct bodies of work.
How would a business know if the new intermediary is routing buyers away from it?
It has to be measured directly, because no engine publishes this. A structured read samples a panel of the business's real buyer questions across each engine, records how often the business is named, and reports it with the engine, locale, and date. That reading is the starting point before any plan is scoped.
Provenance
Sources
- Wikipedia, "Disintermediation" and "Reintermediation" (encyclopedic summary of the general pattern; established as a described concept, primary economics source still to be added)
- Barzilai-Nahon, K., "Toward a Theory of Network Gatekeeping", Journal of the American Society for Information Science and Technology, 59(9), 2008 (established)doi.org
- Simon, H. A., "Designing Organizations for an Information-Rich World", in Computers, Communications, and the Public Interest, 1971 (established)veryinteractive.net
- Brin, S. & Page, L., "The Anatomy of a Large-Scale Hypertextual Web Search Engine", Computer Networks and ISDN Systems, 30(1-7), 1998 (established)doi.org
- Aggarwal et al., "GEO: Generative Engine Optimization", KDD 2024, arXiv:2311.09735 (peer-reviewed, established)arxiv.org
- Kuai, J., Brantner, C., Karlsson, M., Van Couvering, E. & Romano, S., "AI chatbot accountability in the age of algorithmic gatekeeping", 2026 (peer-reviewed, emerging, awaiting replication)
- Pew Research Center, "Do people click on links in Google AI summaries?", 2025 (established, primary panel data)pewresearch.org
- SparkToro & Datos, "2024 Zero-Click Search Study", and SparkToro & Similarweb, 2026 zero-click follow-up (established)
- Gartner, press release forecasting a 25% decline in search volume by 2026, 2024 (contested, used as a forecast-accuracy check)gartner.com
Every figure above is attributed to a real, dated source and tagged with its evidence tier. Where a claim could not be verified to a primary source, it is not stated as fact.