Wolfram Alpha

By Bill Hartzer · Updated 19 August 2026

Not a web index — and why that label exists

Every other engine in this directory can be placed on one axis: does it crawl the web itself, blend its crawl with a partner's, or resell somebody else's results? Wolfram|Alpha sits off that axis entirely. It has no crawler. It has no index of the web — no stored copy of pages, no ranking of documents. It also buys no results from anyone: nothing in what it shows you comes from Google, Bing or any other engine.

Wolfram says so on its own about page. Wolfram|Alpha answers questions, in its words, "not by searching the web, but by doing dynamic computations based on a vast collection of built-in data, algorithms and methods". Site owners will never see a Wolfram|Alpha bot in their logs, because there isn't one.

Calling this "own index" would be technically defensible and practically misleading, which is exactly why it gets its own verdict here. A reader who sees "own index" and pictures something that crawls and ranks the open web the way Mojeek does has been told the wrong thing. The accurate description is a different species of product: a computational knowledge engine. One line captures it. A search engine finds documents. Wolfram|Alpha computes values.

What it can answer that no web index can

The distinction stops being abstract the moment you try a question that has no answer anywhere on the web. Ask for the population of France divided by the area of Texas and Wolfram|Alpha will return it, in the right units, having reconciled two datasets that were never published together. No page on the internet contains that number, so no crawler could ever have stored it and no ranking algorithm could ever surface it. It is computed on demand.

The same property covers a wide class of question:

Symbolic mathematics — algebra, calculus, differential equations, linear algebra and statistics, with step-by-step working that makes it a teaching tool rather than only an answer machine.
Units and dimensional analysis — conversions and mixed-unit arithmetic handled correctly, which is precisely where hand calculation and language models both go wrong.
Cross-domain comparisons — quantities from two different curated datasets compared and adjusted in a single query.
Physical and chemical constants, astronomical and geographic computation, nutrition data, financial arithmetic.
Generated visualisations — a clean plot of a function without writing any plotting code.

And here is what it cannot do, stated just as plainly. It cannot find a document. It cannot find a news story, an opinion piece, a product review, a person's homepage, a recipe, a forum thread, a company's current pricing, or the best pizza near you. Not "does it badly" — cannot, because there is no web index behind it to search. Ask it what a chief executive said yesterday and there is nothing to compute and nowhere to look.

How the computation actually works

Three layers sit behind an answer. First, curated structured data: Wolfram ingests material from external primary sources — Wikipedia's article names The World Factbook and the United States Geological Survey among them — along with many hundreds of domain databases across chemistry, physics, astronomy, mathematics, geography, demographics, finance, nutrition and linguistics. Critically, this data is cleaned, reconciled and structured on ingest by people with domain expertise, not scraped at query time.

Second, the Wolfram Language and Mathematica. The whole system is built on Mathematica technology, which is why the mathematics is as good as it is — the symbolic and numeric engine evaluating your query is the same one Mathematica users run directly.

Third, natural-language parsing into computation. Your free-text question is mapped onto entities and functions in that computational stack and then evaluated. What appears on screen is generated output, not retrieved output.

One honest gap: Wolfram publishes no numbers about any of this. The about page uses only qualitative language, and gives no count of data sources, no data volume and no algorithm count. Figures like "ten trillion pieces of data" or "fifty thousand algorithms" circulate widely and could not be traced to any current Wolfram page. The URL that older writing cites for a public source list returns a 404 as of 19 August 2026. Treat any specific scale claim about Wolfram|Alpha as unsourced until someone shows you the primary page.

Where its answers turn up in other products

Most people who have used Wolfram|Alpha never visited its website. Its largest reach has always been as a back end inside other companies' products — and the pattern of those integrations tells a story worth reading carefully.

Microsoft Bing — integrated historically; no longer active.
DuckDuckGo — integrated historically; no longer active.
Apple Siri — from October 2011, launching alongside the iPhone 4S.
Amazon Alexa — from December 2018.
Microsoft Excel — 2020 to 2023; discontinued.

The Siri and Alexa start dates are sourced. Whether either integration is still live in 2026 could not be confirmed from Apple, Amazon or Wolfram, and Apple's reported rebuild of Siri's knowledge layer around large language models puts the historic dependency in question. So "Siri is powered by Wolfram|Alpha" should not be written in the present tense without a current source. Three of the five integrations above are confirmed dead. Wolfram|Alpha's era as the answer layer inside other people's assistants has substantially passed, displaced by language models.

Who owns it and how it earns

Wolfram Alpha LLC is a subsidiary of Wolfram Research, Inc., privately held and controlled by its founder Stephen Wolfram. It has never been sold, never gone public and never been acquired — which for a 2009-vintage search-adjacent product is itself remarkable. The company is based in the State of Illinois and operates under Illinois law.

Revenue is diversified and, unusually, was never built on search advertising. There is a free tier on the public site; Wolfram|Alpha Pro, a consumer and student subscription launched on 8 February 2012, adds file upload, extended computation time, step-by-step solutions and data download. The serious commercial line is API metering: eight distinct APIs — Full Results, LLM, Simple, Short Answers, Fast Query Recognizer, Summary Boxes, Instant Calculators and Spoken Results — sold with a free non-commercial allowance, tiered commercial rates, and negotiated custom or on-premise arrangements for high-volume users. Enterprise and education licensing sits alongside the broader Wolfram product line, of which Wolfram|Alpha is one commercial face. Rates change; the structure is what matters.

No advertising dependence, no acquisition, no investor exit pressure. That structure is a large part of why Wolfram|Alpha has outlived very nearly every alternative search engine of its generation.

Privacy: clean mostly by construction

Wolfram|Alpha's privacy posture is unusually good, and largely because of what the product structurally does not need. Its terms of use say queries are stored in log files and used to generate results and study usage patterns; that Wolfram does not associate individual inputs with individual human users; that it does not release individual or aggregated lists of inputs to any third party except in response to lawful court orders; and that it does not collect or solicit personal information from users of the free service.

The reason to take that more seriously than the equivalent paragraph on an advertising-funded engine is incentive rather than wording. Wolfram|Alpha runs no behavioural advertising and no ad auction, so it has no commercial reason to build a profile from your queries. An ad-funded engine making the same promise must resist its own revenue model to keep it.

Three caveats belong beside that. The no-personal-information claim is scoped to the free service — Pro subscribers have accounts, billing and logins, and are identified. Query logs exist, and United States legal process reaches them; Wolfram says as much itself. And no independent adversarial audit of Wolfram|Alpha's data handling could be found, so these are well-drafted, long-standing company claims rather than verified ones. Worth remembering too that Wolfram|Alpha queries skew mathematical, medical, chemical and financial — a sensitive mix even when unlinked from identity.

The AI era, from the other side

Wolfram's position on AI is the inverse of most engines': Wolfram|Alpha is more often the tool an AI calls than an AI itself.

Its answers are computed deterministically from curated data. It does not hallucinate a number, because it does not produce the answer by generating text probabilistically — an answer either computes or it does not. That single property is why language-model vendors integrate it, and Wolfram has productised it accordingly: one of the eight published APIs is an LLM API, explicitly described as results optimised for use by a large language model. That is the formalised version of the well-known assistant integrations, in which a model sends a query out to Wolfram|Alpha and reasons over computationally grounded facts.

Around it Wolfram now sells Wolfram MCP servers for cloud and local use, an Agent One API that returns a model response and computational intelligence in one call, and component APIs for what it calls computation-augmented generation — its counterpart to retrieval-augmented generation, grounding a model in computation rather than in retrieved documents. A Wolfram AI Assistant ships inside Mathematica and Wolfram|One as of Version 15, announced by Stephen Wolfram in June 2026, alongside AI Access Subscriptions, Chat Notebooks and language-model functions within the Wolfram Language.

One thing is not established: whether generative AI summaries have been added to the consumer result page at wolframalpha.com itself, and if so whether they can be switched off. The AI Assistant is documented as a Mathematica and Wolfram|One feature, which is a different product surface. The two should not be conflated.

Who it suits, who it does not, and what people get wrong

Use it for: mathematics above all, unit-heavy calculation, scientific constants, quantitative comparisons across domains, and any situation where you need a number to be right rather than plausible. In an era of confidently wrong generated answers, "it does not invent numbers" has become a sharper virtue, not a duller one.

Do not use it for: anything a search engine is for. It also has real weaknesses within its own territory. Its natural-language parsing is brittle in a specific and frustrating way — phrase a question slightly wrong and you get nothing useful, with little indication of why, and "speaking Wolfram|Alpha" is a genuine learned skill. Data currency varies unpredictably by domain: some datasets are near-live, others years stale, and the interface rarely tells you which. Free-tier limits on computation time and step-by-step solutions are aggressive enough that serious use is effectively a subscription. And tracing an answer back to the underlying source is harder than clicking a link on a results page, which matters if you need an auditable citation chain.

Who should stay away: anyone looking for a Google replacement. Framing Wolfram|Alpha as an alternative search engine guarantees disappointment, and it is by far the most common way people bounce off it.

Three misreadings recur. That it failed because it never beat Google — it was never trying to; the 2009 press invented the "Google killer" frame, Wolfram did not, and seventeen years later it is independently owned and shipping major releases. That it is an AI that generates answers — the opposite; it is the anti-hallucination layer other systems call. And that it is just Wikipedia with a calculator — Wikipedia is one input among hundreds of curated sources, and the computation layer is the Mathematica engine, not a lookup. As for whether anybody still uses it: direct web traffic is modest and third-party estimates suggest it is declining, but embedded usage through APIs and model integrations is invisible to traffic estimates and has never been published. Neither direction can honestly be asserted.

The right comparison set is not Google, Bing and DuckDuckGo. It is Mathematica, MATLAB, a good reference library and a scientific calculator — plus, increasingly, the computational tool layer beneath other people's AI. Judged there it is excellent and has no real equivalent. Judged as a search engine it fails, because it is not one.

Frequently asked questions

Is Wolfram Alpha a search engine?

No, and Wolfram says so itself. Its about page states that it answers questions "not by searching the web, but by doing dynamic computations based on a vast collection of built-in data, algorithms and methods". There is no Wolfram|Alpha crawler, no index of web pages, and no partner search feed. It is a computational knowledge engine: it parses your question, pulls values from data curated in advance, and calculates an answer rather than finding a page that already contains one.

Does Wolfram Alpha crawl the web?

No. It operates no crawler at all, so it will never appear in a site's server logs the way Googlebot or Bingbot does, and it stores no copy of the open web. Its data comes from external primary sources ingested and structured in advance — Wikipedia's article names The World Factbook and the United States Geological Survey among them — plus hundreds of domain databases. Nothing is fetched from the live web at query time.

What can Wolfram Alpha do that Google cannot?

Compute an answer that exists on no web page. Divide the population of one country by the area of a US state, adjust two economies' figures to a common basis, or solve a differential equation with step-by-step working, and Wolfram|Alpha produces a result by calculation. A search engine can only return a document somebody has already written. Wolfram|Alpha is also unusually reliable on units and dimensional analysis, where both manual calculation and language models tend to fail.

What can't Wolfram Alpha do?

Anything that requires finding a document. It cannot surface a news story, a blog post, an opinion, a product review, a person's homepage, a recipe, a forum thread or a company's current pricing, because it holds no web index to search. Current events, local queries and general research are all outside it. Its language parsing is also brittle: a slightly misphrased question can return nothing useful with little explanation of why.

Who owns Wolfram Alpha?

Wolfram Alpha LLC, a subsidiary of Wolfram Research, Inc., privately held and controlled by its founder Stephen Wolfram. It has never been acquired, never gone public and never changed hands since launching on 18 May 2009. The company is based in Illinois and operates under Illinois law. That independence is partly explained by its revenue mix: subscriptions, API metering and enterprise licensing rather than search advertising.

Is Wolfram Alpha still used by Siri?

It was, from October 2011, and that integration became its largest source of real-world usage. Whether it remains live in 2026 is unconfirmed — no current statement from Apple, Amazon or Wolfram could be found, and Apple's reported rebuild of Siri's knowledge layer around large language models puts the dependency in question. Worth noting that the Bing, DuckDuckGo and Microsoft Excel integrations are all confirmed to have ended.

Is Wolfram Alpha private?

Its stated position is strong. The terms of use say Wolfram does not associate individual inputs with individual users, does not release lists of inputs to third parties except under lawful court order, and collects no personal information from users of the free service. Because it sells no behavioural advertising, it has no incentive to profile you. Caveats: Pro accounts are identified, query logs exist and are reachable by US legal process, and no independent audit has been published.

How much of the web does Wolfram Alpha cover?

None of it, and no market-share figure for it exists either. StatCounter measures referral traffic to websites, and Wolfram|Alpha does not refer traffic — it answers in place — so it does not appear in search-engine share tables at any level. Wolfram publishes no query volume, user count or subscriber count, and the third-party traffic estimates that exist are modelled rather than measured. Any precise usage statistic you see should be treated as unsourced.