The Search 360

In 1994, David Filo and Jerry Yang created a list of their favourite websites. They called it "Jerry and David's Guide to the World Wide Web." Within 30 days it had visitors from over 30 countries. It became Yahoo, and for several years Yahoo's human-curated directory was how most people navigated the internet. Not an algorithm. Not a crawler. A team of people deciding what was worth reading and putting it in a list.

Then the web got too large for humans to curate. Google arrived in 1998 with PageRank, automated the trust signal, and scaled where Yahoo could not. The directory gave way to the algorithm. Human judgment was replaced by something that could process billions of pages without sleeping. It was faster, more comprehensive, and for a long time, better.

Now the algorithm is giving way to something else again. AI systems synthesise the web into a single answer and deliver it without the friction of ten blue links. No browsing required. No source evaluation. Just the answer, from an invisible process, presented with the confidence of a search engine that has read everything.

And users are appending "reddit" to their queries because they do not trust it.

This is not a coincidence. It is a pattern. Every time information retrieval automates away the human layer, a portion of users eventually wants it back. We are watching the beginning of that correction in real time, and understanding it changes how you should be thinking about content, authority, and what search looks like in three years.

What Is Actually Happening to Search Right Now

The numbers are not ambiguous. Since Google launched AI Overviews in May 2024, the share of news searches ending without any click to a website grew from 56% to 69% in a single year, per Similarweb data. AI Overviews appeared on 13.14% of all US desktop queries in March 2025, more than doubling from 6.49% in January, per Semrush. General search referral traffic to 1,000 web domains dropped from 12 billion global visits in June 2024 to 11.2 billion in June 2025, a 6.7% decline year on year, per Similarweb.

When an AI Overview is present, click-through rates drop to 8%, compared to 15% for traditional results without AI summaries. HubSpot's monthly organic visits fell from approximately 13.5 million in November 2024 to around 6 million in subsequent months. HubSpot's CEO acknowledged on an earnings call that "AI Overviews are giving answers, and fewer people are clicking through to websites."

The open web is losing traffic in absolute terms, not just as a share. The query volume is growing. The clicks are not following.

Why Reddit Happened

The Reddit visibility surge is the most concrete evidence that user behaviour is already signalling a trust correction.

Sistrix data showed a 1,328% increase in Reddit's SEO visibility between July 2023 and April 2024. Reddit moved from position 68 to the fifth highest visibility domain in US organic search results. This was not primarily a Google algorithm decision made in isolation. It was Google responding to a behavioural signal: users were appending "reddit" to queries to bypass the polished content layer and get answers from actual people with actual experience. Google had to surface Reddit results more aggressively or lose those queries to Reddit's own search entirely.

Liz Reid, Google's VP and head of Search, confirmed the direction explicitly in a Wall Street Journal podcast interview: "There is a behavioral shift: people are going to short-form video, forums, and user-generated content." Danny Sullivan, Google's Search Liaison, explained the rationale in parallel: "People are trying to get more authentic information they want to hear from other people with experiences they value."

Google did not decide to surface Reddit because it thought Reddit was inherently better. It surfaced Reddit because users were telling it, through behaviour, that they trusted Reddit more than the optimised content layer. The algorithm followed the signal.

In February 2024, Google paid Reddit $60 million for API and content access. The implicit admission in that deal is significant: Reddit's human signal had become valuable enough that Google needed to contract for it rather than simply crawl it. The authenticity that users were fleeing to was now a commercial asset.

The Pattern Is Not New

What is happening now follows a structure that has repeated itself several times in the history of information retrieval, each time on a larger scale.

The first web was human-curated. Yahoo's directory, DMOZ, hand-edited lists of links compiled by people who had evaluated the sites. The model worked when the web was small enough for humans to manage. It could not scale. Google's algorithmic approach displaced it not because users preferred algorithms in principle, but because algorithms could handle a web that no human team could catalogue.

Google's dominance produced its own trust problem. Algorithmic results could be gamed, as twenty years of SEO history confirms. The content layer filled with optimised pages that ranked without being useful. Users developed workarounds: site-specific searches, navigating directly to trusted domains, and eventually appending platform names to queries to filter out the optimised layer. "Best running shoes reddit." "Best mortgage broker uk reddit." The query modification became so common that Google had to account for it in its ranking systems.

The newsletter explosion of 2020 to 2023 was part of the same correction. Readers who had lost trust in algorithmically surfaced content subscribed to individual writers whose judgment they had reason to trust. Substack grew from essentially zero to over 5 million paid subscriptions by 2025. The format is structurally different from search-optimised content because it is not trying to rank. It is trying to retain a specific reader who already chose to subscribe. That alignment of incentives produces different content.

The Substack wave, the forum preference, the Reddit appending: all of them are the same behaviour at different scales. When the automated layer degrades below a trust threshold, users route around it. They find the human signal by whatever path is available.

What Zero-Query Search Does to This Dynamic

The trajectory of AI search points toward something more fundamental than better answers. It points toward the elimination of the query itself.

The progression is already documented: Featured snippets (2015) are estimated to have produced zero-click rates of around 25%. Knowledge panel expansion (2019 to 2023) pushed that to roughly 50%. AI Overviews (2024 onwards) have taken it to 65% and rising. The next stage, which Google is already developing through AI Mode and ambient search experiences, is proactive answers before the query is formed: the system anticipates what you need and surfaces it without being asked.

This is zero-query search. Your operating system, your phone, your AI assistant answers before you type. The friction of formulating a query, evaluating sources, and navigating to a page is eliminated entirely.

For the trust correction dynamic, zero-query search is the worst version of the problem. At least with a SERP, you could see ten results and evaluate the sources. With an AI Overview, you could read the cited domains and decide whether to trust them. With zero-query, there is no visible source at all. The answer arrives. You did not ask for it in a form that implies a search process. There is no provenance to inspect.

The question this raises is not whether zero-query search is useful. It often will be. The question is what happens to the trust correction mechanism when the search box disappears. Users could not append "reddit" to a zero-query answer. There is no query to modify. The route around the automated layer that users found in the 2020s requires a query to exist in the first place.

The Content Layer Problem

There is a structural issue underneath all of this that connects to the dark matter argument made in an earlier piece on this site.

AI search systems are built on the public web. The public web is the content layer. That content layer is degrading simultaneously on two fronts: expert knowledge has migrated to private platforms, and AI-generated content is filling the gap left behind. Research published in Nature in 2024 confirmed that training AI models on AI-generated content leads to progressive model collapse, where errors compound and the distribution of knowledge narrows over successive generations.

The trust correction users are already demonstrating, the flight to Reddit, to forums, to newsletters, is in part a response to a content layer that feels different even if users cannot articulate exactly why. Optimised content looks correct. It has the right structure, the right length, the right signals. But it often lacks the texture of something written by someone who actually knows the subject and is trying to communicate it, rather than rank for it.

AI-generated content accelerates this. It produces the surface properties of trustworthy content without the underlying expertise. The content layer becomes harder to distinguish from genuine knowledge, and the degradation is invisible in a way that keyword-stuffed pages from 2003 were not. A page from 2003 that was clearly low quality looked low quality. An AI Overview synthesising AI-generated sources looks authoritative.

This is the real stakes of the zero-query trajectory. Zero-query search does not remove the dependency on a content layer. The system still needs to synthesise from somewhere. It is being built on a layer that is degrading while presenting answers with the confidence of a system that has read everything. The worse the layer gets, the more important the provenance of the answer becomes, and the less visible that provenance is to the user.

Reddit's Own Problem

Here is the uncomfortable turn.

The trust correction that drove users to Reddit was a response to optimised content. Reddit felt authentic because it was produced by people without ranking incentives. That is changing.

Google's 2025 updates reduced Reddit's search visibility significantly, targeting low-authority community-driven content that did not meet newly enforced quality thresholds. But the more significant problem is upstream: the moment Reddit's content became valuable enough for Google to pay $60 million for access to it, and valuable enough for SEOs to treat Reddit presence as a ranking strategy, Reddit became a target in exactly the Goodhart sense. The human signal that made it trustworthy is now being manufactured.

Coordinated inauthentic behaviour in forums is not new. What is new is the cost of producing it. Generating varied, contextually plausible forum comments at scale required human operators before late 2022. It now requires a prompt and a few seconds. The fake consensus that existed at low volume before AI is now producible at industrial scale, and it is difficult to detect because it mimics the surface properties of genuine discussion.

If the trust correction drives users to forums, and the forums are being gamed, the correction corrects to a layer that is also compromised. The users who fled optimised content for authentic discussion may be fleeing into optimised discussion. The pattern completes one more turn.

What This Means for Content Strategy

The practical implications pull in two directions simultaneously, and any honest account has to hold both.

In the short to medium term, the trust correction is real and actionable. Users are demonstrating a preference for human-sourced, experience-based answers over synthesised ones. Content that reads like it was written by someone who has actually done the thing, that contains the specific texture of genuine experience rather than summarised knowledge, is more likely to be cited by AI systems, more likely to be the result users click when they choose to click, and more likely to earn the kind of links and references that survive Goodhart pressure.

Google's own behaviour confirms this. Surfacing Reddit, paying for Reddit's content, adjusting ranking to reward UGC: these are signals that user behaviour is being read as a preference for the human layer, and that preference is being incorporated into the algorithm. For content producers, the implication is that demonstrable experience and genuine specificity are, right now, advantages that the automated content layer cannot fully replicate.

In the longer term, the picture is less comfortable. If zero-query search removes the search interface, the trust correction mechanism that users have relied on changes in ways that are not yet clear. If forums are gamed at scale, the destination of the trust correction is compromised. If the content layer continues to degrade, the systems being built on top of it get worse in ways that compound silently.

For iGaming specifically: the queries where AI Overviews are least likely to appear are commercial and transactional ones. Informational queries trigger AI Overviews 88.1% of the time. Commercial queries trigger them at 6 to 8%. This is a meaningful distinction. The informational layer, where affiliates and operators have historically built topical authority, is the one being most disrupted. The transactional layer, where conversion happens, is currently more resilient. Content strategy that has been built on informational volume needs to account for this asymmetry.

The Direction of Travel

Search has completed a partial 360 from human curation to algorithmic automation to AI synthesis, with a correction back toward human sources already underway. The correction is real, measurable in Reddit's visibility data, in Google's own statements, in the $60 million it paid to secure human-generated content.

But the correction is running into two problems simultaneously. The automated layer it is correcting away from is becoming more pervasive, not less: zero-query is not a feature, it is a direction. And the human layer it is correcting toward is itself under pressure from the same economics that degraded the automated layer. Fake consensus at scale is cheap now. Forum authenticity is not guaranteed by the format.

What this does not mean is that quality content is worthless. The evidence runs the other way. Content that is genuinely useful to people who know the subject earns citations, earns direct traffic, earns the kind of references that survive both Goodhart pressure and algorithmic change. The bar is higher than it was, not lower, because the competition from the automated layer has never been cheaper to produce.

What it does mean is that "authentic" is not a format property. It cannot be manufactured by adding first-person language to AI-generated text. It cannot be faked in forums at a quality level that experienced practitioners do not notice. The trust correction users are already demonstrating is a correction toward something specific: information from people who have done the thing and are trying to communicate it honestly. That is a narrow target. It is also the only one that holds.