You search for something ordinary. The results come back and you have the sensation of having read them already: the same handful of large publishers, the same headline construction, the same page that opens with three paragraphs of preamble before the list you wanted. Change the query and much of it stays.
That sensation is accurate and it has structural explanations. None of them requires anyone to have decided that results should look this way.
Editorial note. Everything described here is visible from outside. We make no claim about any search engine's internal workings, ranking factors or policies except where a named primary document says it, and where we have no source we say what we do not know rather than filling the gap. The documents we opened are listed at the end, with retrieval dates, and our sourcing rules are at our source ledger.
What this is, and who it is for
An explanation for someone who has noticed the sameness and is tired of reading complaints about it that never get past the complaint.
Two things this piece will not do. It will not tell you how to rank, which is a different subject belonging to different publications. And it will not attribute motives to any company, because from the outside nobody knows them, and the mechanisms described below produce this outcome without anybody intending it.
The specific thing readers notice, described precisely
The complaint is usually stated as "the results are all the same", which is too vague to explain. What people are actually noticing is four separate convergences.
Format convergence. Results share a page shape: a title matching the query almost word for word, an introduction restating the question, a numbered or subheaded body, a summary, a set of questions and answers at the bottom.
Source convergence. A small set of large publishers appears across topics that have nothing to do with each other, including topics where you would not expect that publisher to have expertise.
Answer convergence. The substantive content agrees, sometimes to the level of shared examples and shared ordering, in a way that reads less like independent agreement and more like shared ancestry.
Absence of the specific. The individual practitioner, the small workshop, the person who has done this for twenty years and writes about it occasionally, is not in the result set at all — not ranked low, simply not present.
These have different causes and it is worth keeping them apart.
Why page shapes converged
Page shape converged because of a feedback process that is entirely visible and requires no inside knowledge to describe.
Anybody producing content professionally can see which pages currently rank for a query, because the results are public. The rational thing to do, when you want to rank, is to look at what is already ranking and produce something structurally similar with more of whatever the ranking pages appear to have. Everyone with a commercial reason to rank does this, continuously, and they all look at the same public evidence.
The result is convergent evolution. The shape that is currently winning gets copied by everyone who wants to win, which reinforces the shape's dominance, which increases the incentive to copy it. This does not require the ranking system to prefer that shape. It only requires people to believe it does and to act on the belief.
There is a second effect layered on it. Where a search interface answers directly on the results page, a publisher who wants a click has to offer something the direct answer does not, and one common response is to add length and structure around the answer. The reader experiences that as preamble.
Two such surfaces are documented by the provider itself, and that documentation is the only account of them we will use. Google's Search Central documentation on featured snippets, published by Google and retrieved on 6 September 2026, describes a featured snippet as a box in which "the format of a regular search result is reversed, showing the descriptive snippet first", says that Google's systems determine whether a page would make a good featured snippet and elevate it, and states that a page cannot be marked as a featured snippet by whoever publishes it. The same documentation set's page on AI features, retrieved on the same date, describes AI Overviews and AI Mode, says both surface links onward to websites, and states that a publisher wanting to limit what appears in them uses the existing nosnippet, data-nosnippet, max-snippet and noindex controls rather than anything new.
What that documentation does not give anybody is a frequency. It does not say how often either surface appears against a given query, and we have run no measurement of our own, so this piece makes no claim about how common they are — only that where a direct answer is present, the incentive facing a publisher who wants the click changes.
Why the same publishers appear across unrelated queries
Three mechanisms, all structural.
Evidence accumulates unevenly. Ranking systems, whatever their specifics, work from evidence about pages and the sites they sit on. A publisher who has been operating at scale for a long time has more of every kind of evidence than a small site does: more pages, more links, more history, more measurable reader behaviour. Some of that evidence is about the site rather than about the individual page, which means it transfers to new pages on subjects the publisher has never covered before. A small site's first page on a subject starts from nothing.
Publishing at scale is a business model. A publisher whose economics depend on covering many subjects will cover many subjects. That is why a general-interest publisher appears on a query about a niche craft: not because anybody promoted it there, but because it published on that craft as part of covering everything, and it brought its accumulated site-level evidence with it.
Consolidation of ownership. Where several apparently independent publications share an owner and a production process, the variety a result set appears to offer can be narrower than it looks. We are not going to name examples or put a figure on how common it is. Establishing who owns a publication means opening that publication's own corporate disclosures or a companies register in the relevant jurisdiction, one publication at a time, and an ownership claim written from memory is precisely the kind of thing this site exists not to do. The part worth taking away is the reading habit rather than a list: several mastheads on a results page is not by itself evidence of several independent sources, and the way to check any particular one is to look for the owner named in its own imprint, ownership disclosure or company filing.
Again: none of this needs a decision by anyone. Change the personnel at every company involved and the same pattern falls out of the same structure.
The feedback loop between what ranks and what gets written
The ranking side is only half of the explanation. The writing side is the part people miss, and it is the more powerful of the two.
The loop runs like this:
- A writer or publisher decides what to write by looking at what people search for and at what currently ranks.
- They produce something matching that shape, on that topic, at that length.
- Whatever ranks reinforces the belief about what ranks, and gets copied more.
- Subjects with no search volume do not get written about, because nothing in the process suggests them.
- Formats that do not resemble the current winners do not get produced, because producing them is a bet against visible evidence.
Two consequences follow, and they are the important part of this article.
The corpus narrows. Not only the ranking, the corpus. Things that would have been written are not written, because the process that decides what to write is looking at a set of signals in which they do not appear. You cannot rank a page nobody wrote.
Idiosyncrasy is selected against. The person with an unusual view, an unusual format, or a subject too small to register produces work that the loop has no way to notice, so producing it is unrewarded, so less of it gets produced over time.
This is the same dynamic, at a different layer, as the one described in how a top ten list gets assembled: the ordering is often decided before anyone has examined the things being ordered.
What this does to a business too small to enter the loop
A small operator faces a specific version of this, and it is worth being exact rather than sympathetic.
They cannot publish at the volume the loop rewards, because publishing is not their business. They generate little of the evidence a ranking system can use — few links, few reviews, little structured data, little measurable behaviour — because they are small, not because they are bad. They cannot match the page shape without hiring someone to write it. And they are competing for the same finite result positions as publishers whose entire business is occupying those positions.
The outcome is not that they rank badly. It is that they are absent, and absence looks identical to non-existence from the reader's side. That is the mechanism behind why the small stuff got harder to find, and it is why a route that does not run through ranked results has to exist for small operators to be encountered at all. What that route used to be is described in what the blogroll used to do.
What a reader can do about it
Concrete behaviours, not a change of attitude.
Search for the conversation instead of the answer. Restrict a query to a forum, community or platform where practitioners talk. You get recommendations from people with no stake in ranking, and the archives are searchable.
Search for the phrasing people use when recommending. Queries built around how a person would ask another person — who actually makes this, who did you use, where did you get it — surface different pages than queries built around the product name.
Use an index that is not the main one. Independently built search indexes crawl a different slice of the web and return genuinely different results, including small sites that a general index ranks far down.
Ask a human who would know. A trade counter, a supplier, a specialist shop, someone already in the field. This is the highest-quality route available and the least convenient.
Follow a person, not a platform. A newsletter or a links page run by one identifiable person gives you a recommender whose taste you can calibrate over time.
Check the maintenance date on anything curated. A list with no visible last-updated signal is a claim about whenever the author last cared. How the listing category makes its money, and how to read one sceptically, is at how directories work.
What we do about it, and its limits
Small and honestly scoped, so here it is with the limits attached.
This site keeps a register of small independent businesses that a named person has looked at, with the date of the last check on the entry, and shows them one at a time on the stumble. Nothing about a position can be bought, so nothing in it reflects who can afford to compete for a result.
The limits. A manually checked register is necessarily small, because checking entries is recurring work and recurring work sets the ceiling. It can cover only a fraction of the businesses that would qualify. Its gaps are the editors' gaps rather than a statement about the field. And it does nothing at all about the loop described above, which keeps running whether or not small collections exist.
What it can do is be one route that does not select for the ability to publish at volume. That is a modest claim and it is the accurate one. The rest of this cluster is at the independent web.
In Plain English
Four different things are converging and people experience them as one. Page shapes look alike because everyone producing content can see what currently ranks and copies it. The same big publishers appear everywhere because evidence about a site accumulates and transfers to new pages on subjects they have never covered. The answers agree because they are frequently descended from each other. And small specific operators are missing entirely rather than ranked low.
The strongest mechanism is the one on the writing side. What gets written is chosen by looking at what people search for and what already ranks, so subjects nobody searches for do not get written about and formats that do not resemble the winners do not get produced. The corpus narrows before any ranking happens.
None of this requires anyone to have intended it. Replace every person at every company involved and the same structure produces the same results.
What helps a reader is going around the ranked result set: search the communities where practitioners talk, use an index built by someone else, follow individual recommenders, and ask a person who would know.
Sources
Each entry below is recorded in the source ledger under the identifier shown.
- S-030 · Featured snippets documentation — Featured Snippets and Your Website, Google Search Central. https://developers.google.com/search/docs/appearance/featured-snippets Retrieved 2026-09-06. Supports: the description of a featured snippet as a reversed-format result box, the statement that Google's systems select and elevate it, and the statement that a publisher cannot mark a page as one.
- S-029 · AI features documentation — AI Features and Your Website, Google Search Central. https://developers.google.com/search/docs/appearance/ai-features Retrieved 2026-09-06. Supports: the description of AI Overviews and AI Mode, that both surface links onward to websites, and that publishers limit their content's appearance using
nosnippet,data-nosnippet,max-snippetandnoindex.
Try the stumble for one small business at a time, chosen by a person rather than by a position anyone could buy.