The web assumed information required travel. That assumption is gone — and it invalidates a different category of practice than an algorithm change ever could.
The web was built on an assumption so basic that nobody stated it: that finding information meant going somewhere. Every structure we have — the page, the site, the link, the visit, the entire economics of publishing — rests on that assumption. Answer engines have quietly removed it, and most brands are still optimising for a destination-based internet while their customers have started using a different one.
It is tempting to describe this as an algorithm shift, which flatters the industry by making it a familiar kind of problem. It is not. Algorithm shifts change which destinations get chosen. This changes whether a destination is involved at all, which is a different category of event and has consequences that no amount of ranking work addresses.
The assumption being removed is that information retrieval requires travel — that you go to a place to get a thing. Every artefact of the current web encodes it: the page as a bounded document, the visit as the measured event, the site as an owned territory, and an advertising economy built entirely on arrivals. Remove the assumption and each of those becomes a legacy structure rather than a natural one.
The first unbundling is compositional. A page was a bounded thing you arrived at, read, and left. In an answer layer, retrieval operates on passages: a section of your page is extracted, combined with sections from three other sources, and presented as something none of you wrote.
This is genuinely new. Your document is no longer the unit of consumption, which means the care put into its arc, its narrative, its build toward a conclusion is largely invisible to the mechanism that now determines whether anyone encounters it. The unit of visibility has become the sentence, and almost nothing about how content is briefed, written, or measured reflects that.
The second unbundling is economic. Every measurement system, attribution model, advertising market, and content business case built over three decades assumes influence produces a visit. Answer engines break the link: your content can inform a decision, shape a comparison, and supply the fact that closes it, without a session occurring anywhere.
Publishers encountered this first because their business is the visit, and their experience is instructive precisely because they had nowhere to hide. But the same disconnection is arriving for every business with an informational content operation. What changes is not the value your content creates; it is whether any instrument you own can see it.
The third unbundling is territorial and the least discussed. A website was a place you controlled: your framing, your sequence, your context. When an answer engine composes a response, your content appears alongside competitors’ content, in a frame you did not write, in an order you did not choose, adjacent to claims you might dispute.
You are no longer publishing into your own territory. You are contributing material to somebody else’s composition, and the composition is what your customer reads. That is a fundamental change in the relationship between a brand and its own words, and it explains why so much brand-safety and message-control practice now feels like it is operating on the wrong surface.
What answer engines separate that the destination-based web held together. Each unbundling invalidates a different category of established practice.
Page → passage
The bounded document is replaced by the extracted section as the unit of consumption.
Invalidates: narrative structure as a visibility asset.
Influence → visit
Content can shape a decision without producing a session.
Invalidates: traffic as a proxy for value.
Site → composition
Your material appears inside someone else’s frame, beside competitors.
Invalidates: the site as a controlled boundary.
Every structure we built — the page, the visit, the site, the entire advertising economy — encodes that premise. Answer engines removed it, and most brands are still optimising the destination.
The change is invisible from inside conventional reporting, which is the whole problem. Rankings hold. Impressions may even rise. The instruments a marketing team owns continue producing numbers that look broadly normal while the mechanism underneath them decouples from the outcome they were built to predict.
It is also uncomfortable in a specific way. Noticing it properly means accepting that a considerable portion of accumulated expertise addresses a surface that is losing share, and that the remedy involves capabilities most search functions do not have. Organisations tend to notice things they are equipped to act on, which is a reasonable heuristic and a poor one here.
If the destination assumption is gone, what replaces it as the organising principle? Being the source rather than the destination. In a composed-answer environment the valuable position is not owning a place people go but being the material that composition is built from — which is a different asset with different properties.
It is less controllable, since you do not write the frame. It is more durable, since it rests on accumulated corroboration rather than on a position that can be displaced. And it is considerably harder to buy, which is why it advantages organisations willing to build over years and disadvantages those expecting to purchase visibility. That is a genuine reallocation of who wins, and it is only beginning.
Rankings and sessions describe the destination web accurately. DUNkē measures the other one — citations across eight AI engines, per prompt, against competitors.
Not everything is being rebuilt, and it is worth being precise about what carries over. Genuine expertise survives, because composed answers still need sources that know things. Reputation survives and matters more, since corroboration is the mechanism by which a source is chosen. Original information survives most of all — data nobody else has cannot be summarised out of relevance.
What does not survive is the presumption that owning a good destination is sufficient. A brand with an excellent site, strong rankings, and no independent corroboration is well-positioned for an internet that is losing share. That is not a reason to abandon the site; it is a reason to stop treating it as the whole of the strategy.
The current state is not the end state, and treating it as such produces bad positioning. Answer engines today mostly describe and recommend. The direction of travel is toward acting — assistants that shortlist, compare, book, and purchase. When that arrives, being named stops being a visibility question and becomes a distribution one.
That transition is the argument for building the slow assets now. Corroboration accrues over years, and the brands that will be shortlisted by agentic systems are the ones whose evidence base was built before shortlisting mattered. Optimising for the current interface is a reasonable tactical choice and a poor strategic one, because the interface is the least stable thing in this picture.
The web has been through one comparable change: mobile. And the instructive part is not the technology but the pattern of denial. Desktop traffic held for years while mobile share grew, every quarter looked survivable, and the organisations that moved late did so under pressure rather than foresight.
The parallel is imperfect — mobile changed the device, not the premise — and the behavioural pattern is identical. A gradual share shift, instruments that keep reporting normally, and a rational-seeming decision to wait for clarity that in retrospect was the expensive one. The brands that adapted early to mobile were not better forecasters; they were less committed to the previous arrangement.
A structural point that gets less attention than it deserves: the open web was funded by arrivals. Advertising, affiliate revenue, subscription funnels, and lead generation all monetise a visit, and no equivalent mechanism monetises being the source of a fact inside somebody else’s answer.
That is an unresolved problem rather than a solved one, and it has consequences beyond individual brands. If contributing to answers is unfunded while producing destinations was funded, the incentive to create the material answers are built from weakens over time. Licensing arrangements are an early attempt at a fix. Whether they scale is genuinely unknown, and the outcome shapes what the next web looks like.
Transitions redistribute, and it is worth being specific about the direction. This one advantages organisations with genuine expertise, proprietary information, and accumulated independent reputation. It disadvantages organisations whose position rested on distribution advantage, aggressive publishing, or the ability to purchase visibility.
That is a meaningful reversal, because the second group substantially outperformed the first under the destination web. A brand that built a strong position through volume and technical execution finds those assets carrying less weight, while a smaller organisation with real depth and a good reputation finds its assets appreciating. The reallocation is only beginning.
Unbundling by category
Horizontal timeline with bands for publishing, informational B2B, D2C retail, local services, regulated sectors. Plot where each sits across the three unbundlings. Shows publishing complete on all three, most categories mid-way on the first two, and the third barely started outside publishing.
The general answer to any transition is to build what is durable across both states rather than optimising for either. Here that means genuine expertise, original information, independent reputation, and machine-legible structure — all of which serve the destination web and the answer layer simultaneously.
What to avoid is the two symmetrical errors: abandoning the destination web while it still carries most of the volume, and treating the answer layer as a fad while its share compounds. The defensible position is to keep the first funded and start the second early, which is unexciting and correct.
The strongest counter-argument is that the transition stalls. Regulatory pressure, licensing economics, or the simple fact that answer engines need a healthy source web to draw from could all slow or partially reverse the unbundling. Platforms have already demonstrated they will pull back coverage when it suits them.
If that happens, the destination web retains more share for longer and the urgency in this article is overstated. What would not change is the value of the assets it argues for — expertise, original data, reputation, structure — which perform well in either scenario. That asymmetry is why we would make the same recommendation even holding significant uncertainty about the pace.
The window in which a change is visible but not yet urgent is the only period in which cheap adaptation is possible. Once the shift is undeniable, everyone moves simultaneously, the assets that matter are contested, and the advantage goes to whoever started earlier.
That is the entire argument for writing this now rather than in two years when it will be obvious and useless. Most brands will read it, agree, and do nothing until their numbers force the issue — which is a rational individual response and the reason the brands that act early will find the ground less crowded than it will ever be again.
An article about structural change risks leaving readers with nothing to do, so it is worth landing on the controllable. You cannot influence whether the unbundling proceeds. You can determine whether your expertise is documented, whether your information is original, whether your reputation is independently corroborated, and whether your content is machine-legible.
All four are within reach of an ordinary marketing function and all four appreciate in the environment being described. That is the entire practical content of this argument: the assets that matter are ones you can build, and they happen to be the ones that were always worth building. The transition changes their relative value rather than requiring new capabilities.
If one thing should change this quarter it is instrumentation. Almost no organisation can currently observe the surface this article describes, which means every strategic conversation about it is conducted on intuition while the conversations about the destination web are conducted on data.
That asymmetry guarantees the wrong allocation. Teams fund what they can measure, and the measurable thing is the surface losing share. Closing the instrumentation gap is cheap, fast, and changes the terms of every subsequent decision — which makes it the highest-leverage move available regardless of what one believes about the pace of the transition.
There is a structural reason large brands adapt late to this specific change, beyond ordinary inertia. Their existing position is built on destination assets — site authority, ranking positions, direct traffic, brand recall driving branded search — and those assets continue performing while the environment shifts underneath them.
The signals reach them last precisely because they are best insulated. A smaller competitor with less destination advantage feels the change earlier and adapts sooner, which is the mechanism by which transitions redistribute. Incumbency delays the pain and, in doing so, delays the response — which is why the advantage of moving early accrues disproportionately to challengers.
The board version is three sentences. A growing share of how customers find and evaluate us happens inside interfaces we cannot currently see or measure. The assets that determine our position there — documented expertise, original information, independent reputation — accrue over years rather than quarters. We should start instrumenting and accruing now, because the cost of doing so early is trivial and the cost of doing so late is that everyone else has too.
That framing avoids both errors that usually kill these conversations: overstating the immediacy, which invites scepticism, and understating the compounding, which invites deferral. It is also true, which is a useful property in a board paper.
The web assumed information required travel. That assumption produced pages, sites, visits, advertising, and an entire discipline devoted to being the place people go. Answer engines removed the assumption without replacing the structures built on it, which is why everything feels simultaneously familiar and slightly wrong.
What replaces being the destination is being the source. It is less controllable, more durable, harder to purchase, and accrues to organisations willing to build over years. Most brands will recognise this and act when their numbers force it. The ones that act while the recognition is still optional will find the ground considerably less crowded than it will ever be again.
Three limits, stated plainly. We are not claiming the destination web is ending — it is losing share, which is a different and smaller claim. We are not claiming a timeline, because the pace is set by platform product decisions rather than by any trend with momentum. And we are not claiming the transition is irreversible.
What we are claiming is that a foundational assumption has been removed, that three specific unbundlings follow from it, and that the assets which matter in the resulting environment accrue slowly enough to make early accumulation worthwhile even under significant uncertainty about the pace. That is a narrower argument than the title suggests and it is the one the evidence supports.
If a customer could get everything they need to know about your category — the options, the trade-offs, the recommendation — without visiting any company’s website, what would make yours the one they are told about?
That question has an answer for some brands and not for others, and the difference is not size or budget. It is whether anything about the organisation is documented, independent, and specific enough to be worth a machine repeating. Brands that can answer it are already positioned. Brands that cannot have found their actual strategic problem, and it was not a search problem.
The three unbundlings, before and after
Split diagram. Left: the destination web — page as unit, visit as event, site as bounded territory, with arrows showing a user travelling to a source. Right: the answer layer — passages extracted from multiple sources, composed into one response, no travel. Annotate what each side makes measurable and what it makes invisible.
Three moves, in order. Instrument the surface you cannot currently see, because every subsequent decision depends on observation and almost no organisation has it. Audit whether your expertise, your data, and your reputation are documented anywhere a machine could find them. And stop treating your website as the strategy rather than as the foundation.
None requires believing any particular forecast about the pace of this transition. All three improve your position under the destination web as well, which is the property that makes them safe bets under uncertainty — and uncertainty about the pace is the correct epistemic state here.
Arguments about structural change in this industry tend toward two failure modes: apocalyptic, which is unfalsifiable and sells consulting, or dismissive, which is comfortable and ages badly. We have tried to avoid both by separating what the evidence supports from what we infer, and by stating where we could be wrong.
The honest position is that something foundational has changed, the pace is genuinely uncertain, and the assets worth building are the same ones under every plausible scenario. That is a less exciting claim than either alternative and it is the one we would still defend in three years, which seems the appropriate test for an article of this kind.
This is the argument the rest of our publication operates inside. The state-of-AI-search piece sets out what is measured and what is merely asserted; the frameworks describe how to build for the environment it argues we are entering; the research hub collects the underlying studies.
Read alone this is a thesis. Read alongside the measurement and framework material it becomes the reason those exist — and readers who find the argument overstated should read the research pieces first, since they are considerably more conservative and reach compatible conclusions from evidence rather than from interpretation.
The web assumed information required travel, and answer engines removed the assumption without replacing what was built on it. Three unbundlings follow: page to passage, influence to visit, site to composition. What replaces being the destination is being the source.
That asset is less controllable, more durable, harder to buy, and accrues over years. Which means the brands positioned for whatever this becomes are the ones building it now, while the recognition is still optional and the ground is still uncrowded.
Three indicators will tell us whether this argument is holding. Whether answer coverage continues expanding or stabilises, since platforms have already demonstrated they will pull back. Whether licensing arrangements between AI companies and publishers scale into a genuine economic mechanism for source contribution. And whether agentic behaviour moves from demonstration into ordinary use.
The third is the one that decides how consequential this becomes. Descriptive answers redistribute attention; acting assistants redistribute transactions. If that arrives at scale, being named stops being a marketing concern and becomes a distribution one — and every argument in this article becomes considerably more urgent than it currently reads.
Is this overstated? Most of our traffic still comes from search.
Almost certainly, and that is consistent with a transition rather than a completed change. The relevant question is the trajectory of the share, not the current level, and the trajectory is what argues for building now.
Does this mean websites do not matter?
They matter as the foundation and no longer as the whole. Content has to exist somewhere retrievable, and the site remains where conversion happens. What changes is that owning a good destination stopped being sufficient on its own.
What is the first structural change a brand should make?
Measurement. Every subsequent decision depends on being able to see the surface, and almost no organisation currently can. It is also the cheapest change on the list.
Who is best positioned for this shift?
Organisations with genuine expertise, original information, and accumulated independent reputation — frequently not the ones with the strongest current rankings. That mismatch is the interesting part of this period.
The web assumed information required travel, and every structure we built encodes that assumption: the page as unit, the visit as event, the site as boundary, and an entire economy of arrivals. Answer engines removed the assumption without announcing it, which is why the change is invisible in reporting that continues to describe the old arrangement accurately.
Three unbundlings follow. Your document is no longer the unit of consumption — the passage is. Your influence no longer produces a measurable visit. And your content now appears inside compositions you did not write, beside competitors, in frames you do not control. What replaces the destination as the organising asset is being the source: less controllable, more durable, harder to buy, and accruing on a timescale that rewards brands who start before the picture is clear.
Methodology note: this is an argument rather than a study. The empirical claims it rests on — click displacement, referral decline without substitution, cross-surface divergence — are cited to published research. The interpretation built on them, including the unbundling framework and the transitional-period argument, is ours and is not measured. Readers should weight the evidence and the interpretation separately.
“Every structure we have assumes that getting information means going somewhere. That assumption is gone, and most brands are still optimising the destination.” The Age’X Research Team
Get a free GEO audit — the same analysis behind every article here.