A vertical search engine is one whose index is restricted to a single kind of thing: images, scholarly papers, onion services, business listings, or computable facts. The term contrasts with horizontal search, in which an engine such as Google or Bing tries to cover the whole open web and answer any question typed into it. A vertical engine gives up that breadth on purpose. It decides in advance what it will hold, builds its index around that one object, and is useless the moment you ask it for anything else.
Why specialized search engines exist when general engines are so large
If a general engine already holds a very large share of the public web, why would anyone need a smaller index? Because size is the wrong measure. A general web index is built around one unit, the text of a web page, and some questions cannot be asked of that unit at all. Three structural gaps account for every engine in this category.
- A different unit of indexing. TinEye stores a compact digital signature (a short fingerprint of the image) for each image it finds, and it is queried with an image rather than with words. Google Scholar indexes scholarly records, clustering a publisher's version, a preprint and a repository copy into one work with authors, venue and citations attached. Wolfram|Alpha indexes no documents at all; it computes answers from curated, structured data.
- A different network. Ordinary crawlers fetch pages over the public internet. A Tor onion service (a site reachable only through the Tor anonymity network, at an address ending in .onion) does not resolve in public DNS, so a crawler must run Tor to reach it. Since the version 3 onion design, Tor's own directory cannot be read as a list of addresses either, which leaves link discovery and operator submission as the only ways in. Ahmia exists to do that work.
- Different inclusion rules. Google's own inclusion page states that "only scholarly papers are appropriate for inclusion in Google Scholar." Jayde admits only business sites whose owners submit them, prove ownership and pass human review. In both cases the gate defines the product.
Specialist search also predates Google: Jayde's own about page dates the directory to 1996, although other pages from the same corporate lineage give 1993 and 1998.
The verdict: "Not a web index"
Every profile on this site leads with a single verdict about where an engine's results come from: its own index, a hybrid, results resold from another engine, or discontinued. That scale presupposes web retrieval, and niche search engines break it. Most here are not resellers: TinEye and Ahmia run their own crawlers, and Google Scholar is a separately built index rather than a filtered view of Google web search. Calling them "own index" would still mislead, because a reader would infer that they crawl and rank the open web the way Mojeek or Brave Search does.
The verdict used for this category is therefore Not a web index. It is a neutral, structural label, not a criticism. It says the engine answers a different kind of question, and that comparing it with a general engine on coverage or market share is a category error. Wolfram|Alpha is the clearest case. Its about page says it answers "not by searching the web, but by doing dynamic computations based on a vast collection of built-in data, algorithms and methods." It has no crawler, and no meaningful market share exists for it, because it answers in place rather than referring traffic to websites. A search engine finds documents; Wolfram|Alpha computes values.
What the five engines here actually are
- TinEye is a reverse image search engine run by Idée, Inc. of Toronto since 2008. It finds copies of the same image, including resized, cropped and edited ones, but deliberately does not return merely similar images and does not perform facial recognition. It is also unusual in publishing its own index count: more than 77.6 billion images in September 2025, per its counter as reported on Wikipedia, against just over 2 billion in September 2011. It typically does not add social-media images.
- Google Scholar, launched in beta on 20 November 2004, is the broadest academic search engine on every published measure and the least transparent. Google has never published its size. Gusenbauer's 2019 study in Scientometrics estimated 389 million records in January 2018, a figure the author himself rated "questionable" because the hit counts were unstable. Citation studies are firmer: across 2.45 million citations to highly cited 2006 documents, Martín-Martín and colleagues found Scholar located 94 percent, against 60 percent for Scopus and 52 percent for Web of Science; the seed set was English-language and highly cited, so obscure work may fare differently.
- Ahmia indexes Tor onion services. Developed by Juha Nurmi with the Tor Project through Google Summer of Code 2014, it filters abuse material, and widened that filter in late 2023 to all sexual and suspicious searches. A 2024 Scientific Reports paper analyzed 110,133,715 Ahmia search sessions and found 11.1 percent sought child sexual abuse material; those logs describe Ahmia's users, not all Tor users. For scale, Tor Metrics estimated about 656,000 unique v3 onion addresses a day between 26 September and 25 October 2026, but an address is not a website; a 2021 to 2022 academic crawl found only 39,536 onion domains.
- Wolfram|Alpha, launched on 18 May 2009 and owned by Wolfram Research, is a computational knowledge engine. It is excellent at mathematics and unit-aware comparisons, and fails at anything a document search is for: news, reviews, a company's homepage. Wolfram publishes no figures for the size of its data.
- Jayde is a business-to-business directory with a search box over it, still online, whose footer names iEntry, Inc. as operator. It ranks listings by what it calls ClickRelevance, ordering results by those "clicked the most by unique searches," and it claims more than 2 million business websites. That figure is the operator's own, and it counts submitted listings, a different kind of object from a crawled web index.
How to read a vertical search engine
Because these engines share no unit, no headline number compares them. Four questions do the work instead. What is the object being indexed? How does an item get in: crawling, submission, or editorial curation? What does a published size measure, and who measured it? TinEye's count is its own counter, Scholar's is an outside estimate flagged as questionable by its author, and Jayde's counts listings rather than pages. And what does the engine refuse to do? TinEye's refusal of face matching, Ahmia's abuse filter and Scholar's scholarly-only rule are design decisions.
A vertical search engine is the right tool exactly when your question is about its one kind of thing, and the wrong tool for everything else. None of these five is a Google alternative, and judging any of them as one guarantees disappointment.
